# few-shot

Few-shot learning is a machine learning setting in which a model generalizes to a new task or class from a very small number of labeled examples.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## The Price of an Explanation: What Demanding Verifiable Artifacts Costs, and What It Buys Back

DevFeed: [The Price of an Explanation: What Demanding Verifiable Artifacts Costs, and What It Buys Back](<https://devfeed.tech/articles/the-price-of-an-explanation-what-demanding-verifiable-artifacts-costs-and-what-it-buys-back-145115.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25896>)

Author: Edgar A. Du\\'e\\~nez-Guzm\\'an, Eduardo Due\\~nez, Edgar Said Hernandez Sanchez, Suzanne Sadedin

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>), [reproducibility](<https://devfeed.tech/topics/reproducibility.md>)

Tags: [agi](<https://devfeed.tech/tags/agi.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [bottleneck](<https://devfeed.tech/tags/bottleneck.md>), [budget](<https://devfeed.tech/tags/budget.md>), [costs](<https://devfeed.tech/tags/costs.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ne](<https://devfeed.tech/tags/cs-ne.md>), [decoder](<https://devfeed.tech/tags/decoder.md>), [error-correction](<https://devfeed.tech/tags/error-correction.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [llm](<https://devfeed.tech/tags/llm.md>), [mitigation](<https://devfeed.tech/tags/mitigation.md>), [program](<https://devfeed.tech/tags/program.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [score](<https://devfeed.tech/tags/score.md>), [search](<https://devfeed.tech/tags/search.md>), [synthesis](<https://devfeed.tech/tags/synthesis.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [test](<https://devfeed.tech/tags/test.md>), [verification](<https://devfeed.tech/tags/verification.md>)

### AI overview

On ARC-AGI-2, requiring an LLM to produce a verifiable program reduced performance from 93/120 tasks (77%) to 68/120 (57%). The authors report that most answer-only wins remained out of reach for the program-producing configurations they tested, even when a program matched every demonstration, pointing to generalization as the key bottleneck. A short evolutionary search over two sampled programs improved on best-of-two sampled code by 9.2 ± 1.4 tasks without additional model cost. Checking programs against training examples also enabled early stopping, reducing budget use by up to 38% for best-of-n sampling and 49% for the evolved arm; answer-only runs could not stop this way. The same methodology was applied to quantum error-correction decoder synthesis, producing interpretable decoders evaluated on held-out shots against a certified panel of published decoders.

### Source excerpt

arXiv:2611.25896v1 Announce Type: cross Abstract: Generative models produce many candidate solutions but provide no intrinsic signal by which to rank them. One mitigation is to demand a \emph{verifiable artifact}: a program that can be executed against the available evidence. A common concern is that requiring executable verification incurs costs in both solution quality and computational budget. We measure these two costs on the $120$-task ARC-AGI-2 evaluation set. Demanding a verifiable artifact from an LLM reduced the score from $93$ to $68$ of $120$ tasks ($77\%$ to $57\%$). Of those $25$ tasks, $21$ were reached by no program-producing configuration we ran, at any budget, by sampling or by search; yet for $27$ of the $28$ tasks that only the answer-only arm solved, some program arm held a program that reproduced every demonstration and still failed the test. The bottleneck is obtaining an explanation that \emph{generalizes}, not finding one. A short evolutionary search over two already-sampled programs, at no additional model cost, beats best-of-$2$ sampled code by $9.2 \pm 1.4$ tasks, and over a single sampled program beats $1$-shot sampling by $9.9 \pm 1.3$. Moreover, because a program can be checked against the training examples, any program-producing protocol can terminate early once a task is explained: under the same stopping rule, best-of-$n$ sampling declines up to $38\%$ of its budget and the evolved arm up to $49\%$, whereas an answer-only baseline cannot terminate early. Applying the same methodology to decoder synthesis for quantum error correction yields interpretable decoders evaluated against a certified panel of published decoders on held-out exam shots.

## The ROI of Small Examples: Optimizing In-Context Learning Under Token Budgets

DevFeed: [The ROI of Small Examples: Optimizing In-Context Learning Under Token Budgets](<https://devfeed.tech/articles/the-roi-of-small-examples-optimizing-in-context-learning-under-token-budgets-145450.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24971>)

Author: Anxo Perez, Jorge Gab\\'in, Javier Parapar

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Inference](<https://devfeed.tech/topics/inference.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>)

Tags: [budget](<https://devfeed.tech/tags/budget.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [examples](<https://devfeed.tech/tags/examples.md>), [in-context-learning](<https://devfeed.tech/tags/in-context-learning.md>), [inference](<https://devfeed.tech/tags/inference.md>), [model-performance](<https://devfeed.tech/tags/model-performance.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [qa](<https://devfeed.tech/tags/qa.md>)

### AI overview

Under a fixed prompt token budget, the study finds that using more short demonstrations generally improves in-context learning for question answering compared with spending the budget on a few long demonstrations. The authors grouped examples by length and tested six instruction-tuned models on Beat the AI and Natural Questions. The effect was strongest on Beat the AI; it also appeared on Natural Questions, where gains were smaller and saturated earlier with long contexts. In an ANLI R3 inference experiment, one of the two shorter length groups performed best for five of six models, though longer examples remained competitive. The results suggest demonstration length is a relevant prompt design choice alongside example count.

### Source excerpt

arXiv:2611.24971v1 Announce Type: cross Abstract: In-context learning (ICL) improves large language model performance at inference time without updating model parameters. However, these gains are not free: in practical deployments, longer prompts directly increase inference cost and latency. Under a fixed token budget, prompt construction therefore becomes a resource-allocation problem. While prior work has often focused on how many examples to include in ICL, we study the complementary question of whether the available budget is better spent on a few long demonstrations or on a larger number of shorter ones. We study this aspect in question answering (QA), by organizing candidate demonstrations into length-based bins and constructing prompts from examples within the same length range under fixed budgets. Experiments on two QA benchmarks (\textit{Beat the AI} and \textit{Natural Questions}), using six instruction-tuned LLMs, show that shorter demonstrations are generally the most effective use of prompt space. This effect is especially strong on \textit{Beat the AI}, and remains visible on NQ, although gains are smaller and saturate earlier in the long-context setting. An additional natural language inference experiment on ANLI R3 finds that one of the two shorter bins performs best for five of six models, although longer demonstrations remain competitive. Overall, our results highlight demonstration granularity as an important factor in prompt design.

## Beyond Alignment: Whitening Geometry, Profile Synthesis, and Heterogeneous Ensembles for Cross-Environment WiFi Action Recognition

DevFeed: [Beyond Alignment: Whitening Geometry, Profile Synthesis, and Heterogeneous Ensembles for Cross-Environment WiFi Action Recognition](<https://devfeed.tech/articles/beyond-alignment-whitening-geometry-profile-synthesis-and-heterogeneous-ensembles-for-cross-environment-wifi-action-recognition-145108.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25831>)

Author: Lei Wang

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [action-recognition](<https://devfeed.tech/tags/action-recognition.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decision](<https://devfeed.tech/tags/decision.md>), [eess-sp](<https://devfeed.tech/tags/eess-sp.md>), [encoder](<https://devfeed.tech/tags/encoder.md>), [environment](<https://devfeed.tech/tags/environment.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [profile](<https://devfeed.tech/tags/profile.md>), [wifi](<https://devfeed.tech/tags/wifi.md>)

### AI overview

In the reported leave-one-domain-out WiFi action recognition evaluation, accuracy fell from 0.89+ in training environments to 0.408 with K-shot labels; the analysis attributes the failure mainly to encoder extrapolation into unseen environments, not a lack of action signal or a domain-alignment problem. Training-time, prototype, adversarial, covariance (CORAL), and subspace alignment degraded performance. The Dec-H combination instead uses closed-form source within-class whitening (+8.6 percentage points at zero cost), profile-mixing augmentation (+1.0), few-shot adaptation with alternating whitening refits (+1.5), and heterogeneous decision ensembles (+1.7). Fixed preconditioners were negative in 20/20 tests, while homogeneous scaling and source-weighted stacking were saturated or harmful. Dec-H reached 0.560 at K10 and 0.812 at K40, a 15.2-point gain over baseline, reported as consistent across four environments and seeds. These are abstract-level results; the supplied text does not provide further experimental details.

### Source excerpt

arXiv:2611.25831v1 Announce Type: new Abstract: Cross-environment deployment is the central obstacle to practical WiFi sensing: a pipeline reaching 0.89+ top-1 in its training environments collapses to 0.408 under Leave-One-Domain-Out (LODO) evaluation with only K-shot labels. A ceiling analysis shows the collapse is not an alignment problem: the action signal is strong across seen environments (about 0.89 K10), environment identity is decodable from the encoder (0.965), and the failure concentrates in the encoder's extrapolation to the novel environment. A pre-registered falsification chain confirms that "align the domains" methods--training-time alignment, readout-level prototype alignment, adversarial confusion, and covariance/subspace alignment (CORAL -3.8pp, subspace -8.3pp)--degrade performance, and that source-domain statistics anti-correlate with cross-environment contribution. The effective levers lie elsewhere; we combine them into Dec-H: (i) closed-form source within-class whitening, provably re-geometrizing the readout's L2 regularization (+8.6pp, zero cost); (ii) profile-mixing augmentation (+1.0pp); (iii) few-shot adaptation with alternating whitening refit (+1.5pp; fixed preconditioners are 20/20 negative); (iv) heterogeneous decision ensembles (+1.7pp); homogeneous scaling and source-weighted stacking are saturated or harmful. Dec-H reaches 0.560 K10 (0.812 K40), +15.2pp over the baseline, consistent across all four environments and seeds; all thirteen configurations were pre-registered with frozen criteria.

## From Data Alignment to Parameter Generation: Few-Shot Cross-City Traffic Prediction with Conditional Diffusion

DevFeed: [From Data Alignment to Parameter Generation: Few-Shot Cross-City Traffic Prediction with Conditional Diffusion](<https://devfeed.tech/articles/from-data-alignment-to-parameter-generation-few-shot-cross-city-traffic-prediction-with-conditional-diffusion-145073.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25592>)

Author: Shengchun Wang, Yuhang Lin, Tong Li, Pengyuan Zhou, Andy Chi Lok Tai, Wenjie Chen, Zhu Xiao, Zhu Han

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Internet Traffic](<https://devfeed.tech/topics/internet-traffic.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [cities](<https://devfeed.tech/tags/cities.md>), [clustering](<https://devfeed.tech/tags/clustering.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [imagery](<https://devfeed.tech/tags/imagery.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [node](<https://devfeed.tech/tags/node.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [region](<https://devfeed.tech/tags/region.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [traffic](<https://devfeed.tech/tags/traffic.md>)

### AI overview

CDPG addresses few-shot mobile traffic forecasting in data-scarce cities by generating base-station-specific predictor parameters instead of aligning raw traffic data across cities. It learns a conditional distribution of strong source-city predictor parameters and conditions generation on spatial context from satellite imagery and LLM-derived descriptions, plus temporal patterns identified through unsupervised clustering. The authors report that across three large-scale datasets CDPG outperforms cross-city few-shot baselines by up to 9.72% on standard forecasting metrics and remains robust for few-shot multi-step prediction. These findings are reported in the abstract; details of the metrics and experimental setup are not provided here.

### Source excerpt

arXiv:2611.25592v1 Announce Type: new Abstract: Accurate mobile traffic forecasting is crucial for network service management, supporting proactive capacity provisioning, congestion mitigation, and energy-efficient operation. However, many cities and regions are data-scarce (e.g., newly deployed or poorly instrumented areas), thus making traffic prediction a few-shot learning problem and limiting the practicality of standard data-driven models. Existing cross-city solutions primarily rely on data-space alignment to match traffic distributions between data-rich and data-scarce cities, but such alignment is often brittle under real-world distribution shifts, measurement noise, and city-specific bias. To address these limitations, we propose a Conditional Diffusion-based Parameter Generation (CDPG) framework. Instead of aligning raw data, CDPG learns the conditional distribution of high-performing predictor parameters for source cities and directly generates node-specific model parameters for the target city. To support reliable adaptation for network traffic prediction, CDPG conditions parameter generation on multi-modal spatio-temporal context. Specifically, CDPG captures spatial environment cues by combining satellite imagery with large language model (LLM)-derived semantic descriptions, and summarizes complex temporal behaviors via unsupervised pattern clustering. These contextual components guide CDPG in producing parameters tailored to each base station's operational context, thereby improving robustness under cross-city and cross-region shifts. Extensive experiments on three large-scale datasets show that CDPG consistently outperforms state-of-the-art cross-city few-shot baselines by up to 9.72% on standard forecasting metrics and remains robust for few-shot multi-step prediction. The code is available at https://github.com/LingoodBoy/CDPG.

## Capturing In-Context Learning Dynamics with Task Operators

DevFeed: [Capturing In-Context Learning Dynamics with Task Operators](<https://devfeed.tech/articles/capturing-in-context-learning-dynamics-with-task-operators-144953.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.01054>)

Author: Guangzhi Xiong, Zhenghao He, Bohan Liu, Sanchit Sinha, Wenqian Ye, Aidong Zhang

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [circuit](<https://devfeed.tech/tags/circuit.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [examples](<https://devfeed.tech/tags/examples.md>), [in-context-learning](<https://devfeed.tech/tags/in-context-learning.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [layers](<https://devfeed.tech/tags/layers.md>), [lexical](<https://devfeed.tech/tags/lexical.md>), [models](<https://devfeed.tech/tags/models.md>), [projection](<https://devfeed.tech/tags/projection.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [scaling](<https://devfeed.tech/tags/scaling.md>)

### AI overview

Task Operator (TO) captures how in-context learning changes attention outputs, allowing those changes to be replayed as an analytically derived update to the attention output projection. The authors report that the transformation parameters are empirically stable across samples for a given task. Across lexical, algorithmic, and reasoning tasks, TO performs best overall among the prior methods evaluated and narrows the gap between zero-shot inference and in-context learning. The extracted knowledge concentrates in a task-specific sparse circuit; averaging operators from disjoint demonstration batches supports many-shot scaling without expanding the context window. These claims are based on the abstract; it does not provide numerical results or further implementation details.

### Source excerpt

arXiv:2610.01054v2 Announce Type: replace-cross Abstract: In-context learning (ICL) enables language models to perform new tasks from demonstrations without weight updates. However, every ICL inference requires processing the full set of examples, resulting in inefficient deployments, and how ICL works mechanistically is not fully understood. Prior work compresses ICL into fixed activation vectors extracted from specific layers or positions, but these input-independent interventions fail on complex tasks where the output depends on fine-grained interactions with the input. By analyzing the ICL forward pass, we show that each attention head's output is an affine transformation of its context-masked counterpart, and that the parameters of this transformation are empirically stable across samples for a given task. Building on this, we introduce Task Operator (TO), which replays this transformation as an analytically derived update to the attention output projection. Across lexical, algorithmic, and reasoning tasks, TO achieves the best overall performance among prior methods and substantially narrows the gap between zero-shot inference and ICL. We further show that the extracted knowledge concentrates in a task-specific sparse circuit across layers and positions, and that averaging operators from disjoint demonstration batches enables effective many-shot scaling without expanding the context window. Our code is available at https://github.com/gzxiong/task_operator.

## Is a Description Worth a Thousand Demonstrations? Specificity and Reliability of Task Descriptions in In-Context Learning

DevFeed: [Is a Description Worth a Thousand Demonstrations? Specificity and Reliability of Task Descriptions in In-Context Learning](<https://devfeed.tech/articles/is-a-description-worth-a-thousand-demonstrations-specificity-and-reliability-of-task-descriptions-in-in-context-learning-145070.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25560>)

Author: Bruce Nguyen, Bei Jiang, Linglong Kong

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [bayes](<https://devfeed.tech/tags/bayes.md>), [bayesian](<https://devfeed.tech/tags/bayesian.md>), [bayesian-statistics](<https://devfeed.tech/tags/bayesian-statistics.md>), [computer-science](<https://devfeed.tech/tags/computer-science.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [in-context-learning](<https://devfeed.tech/tags/in-context-learning.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [linear-regression](<https://devfeed.tech/tags/linear-regression.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reliability](<https://devfeed.tech/tags/reliability.md>), [sample-size](<https://devfeed.tech/tags/sample-size.md>), [specificity](<https://devfeed.tech/tags/specificity.md>), [training](<https://devfeed.tech/tags/training.md>), [transformers](<https://devfeed.tech/tags/transformers.md>)

### AI overview

The paper models task descriptions in in-context learning as informative priors and uses effective sample size (ESS) to express their predictive value in equivalent demonstrations. In its linear regression analysis, a reliable description's ESS increases with specificity and task dimension; if a description may be irrelevant, specificity adds little without demonstrations. Experiments find that meta-trained Transformers approach Bayes-optimal ESS on their training distribution. Pretrained language models tend to underuse highly specific descriptions without reasoning, while most closely match the Bayes-optimal predictor when reasoning is enabled.

### Source excerpt

arXiv:2611.25560v1 Announce Type: new Abstract: A prompt for in-context learning (ICL) can describe a task in addition to, or instead of, demonstrating it through examples. Yet how task descriptions work remains poorly understood theoretically. Through the Bayesian view of ICL, we model descriptions as informative priors characterized by their specificity and reliability. We adapt the effective sample size (ESS) from Bayesian statistics, quantifying a description's predictive value as an equivalent number of demonstrations. For in-context linear regression, we show that a reliable description's ESS grows with its specificity and with the task's dimension. A description that may be irrelevant, however, gains little from its specificity unless demonstrations are given. In practice, do in-context learners use descriptions as a Bayes-optimal predictor would? Our experiments show that meta-trained Transformers come close to the Bayes-optimal ESS when tested on their training distribution. Pretrained language models, however, make too little use of highly specific descriptions when answering without reasoning; most of them closely match the Bayes-optimal predictor only when reasoning is enabled. These results extend the Bayesian analysis of ICL from demonstrations to descriptions.

## Collision-Free Is Not Enough: Passage-Conditioned Navigation for Wheeled-Legged Robots

DevFeed: [Collision-Free Is Not Enough: Passage-Conditioned Navigation for Wheeled-Legged Robots](<https://devfeed.tech/articles/collision-free-is-not-enough-passage-conditioned-navigation-for-wheeled-legged-robots-145587.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25788>)

Author: Wai Ka Sun, Waiyiu Pong, Bowen Wang

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Robot Navigation](<https://devfeed.tech/topics/robot-navigation.md>), [navigation](<https://devfeed.tech/topics/navigation.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>)

Tags: [competition](<https://devfeed.tech/tags/competition.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [legged-robots](<https://devfeed.tech/tags/legged-robots.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [robots](<https://devfeed.tech/tags/robots.md>)

### AI overview

Passage-conditioned navigation records a successful taught crossing for each stair or ledge and, after every local replan, uses an entry gate to make the robot turn or back up and retry when its approach differs by more than 10 cm or 10°; it requires neither a robot model nor additional training. With path, gait schedule, and sensing unchanged, the method increased disturbed ledge crossings within those limits from 24.1% to 88.6%. In contact simulation, robots pushed off course before stairs completed 180 climbs without falling, compared with 9 falls when the stored direction was ignored; after a 1.2 m displacement, they rejoined the route and ended an average 0.14 m from it, versus 1.10 m without rejoining. A real wheeled-legged robot completed six competition runs across 271 m of multi-obstacle terrain and 7.0 m of elevation, finishing fully autonomously.

### Source excerpt

arXiv:2611.25788v1 Announce Type: cross Abstract: A collision-free detour can still bring a legged robot to the next stair or ledge a few centimetres or degrees off, and a fixed locomotion skill, factory or learned, can stall or fall. Existing methods prevent this with a robot model or extensive trial and error, but on a new site with unknown skills neither is available, so they cannot be used. We find that a single 27-minute teaching session already supplies what these methods have to derive or learn: for every stair and ledge, one crossing the skill has completed. Our method, passage-conditioned navigation, makes the route remember how each transition was entered, not only where the path went. It stores each taught crossing as an event and, without training or a robot model, enforces it after every local replan: before each ledge, an entry gate makes the robot turn or back up and retry if it is more than 10 cm or 10{\deg} off the taught crossing. Under localization noise, with path, gait schedule, and sensing unchanged, this raises the share of disturbed ledge crossings that stay within these limits from 24.1% to 88.6%. In contact simulation with our learned policies, a robot pushed off course before stairs never falls in 180 climbs, against 9 falls when the stored direction is ignored; pushed 1.2 m away, it rejoins the route, ending on average 0.14 m from it (1.10 m without rejoining), and finishes as often as an undisturbed robot. This suits any robot with fixed skills that repeats a route over terrain transitions, as in inspection and patrol, where one wrong entry can end the task. We deployed it in competition on a real wheeled-legged robot: over 271 m of multi-obstacle terrain spanning 7.0 m of elevation, it completed all six runs, won third place at the 2026 Global Open-source AI Challenge, which drew 2,999 submissions, and was the only finalist in its task to finish fully autonomously.

## Parameter-Efficient Adaptation of Self-supervised Vision Backbones Survives Distribution Shift in Radar Micro-Doppler HAR

DevFeed: [Parameter-Efficient Adaptation of Self-supervised Vision Backbones Survives Distribution Shift in Radar Micro-Doppler HAR](<https://devfeed.tech/articles/parameter-efficient-adaptation-of-self-supervised-vision-backbones-survives-distribution-shift-in-radar-micro-doppler-har-145113.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25874>)

Author: Keryan Chelouche (ETIS UMR 8051, CY Cergy Paris Universit\\'e, ENSEA, CNRS, Cergy, France, Fondation L\\'eonie Chaptal, Sarcelles, France), Petr Dobias (ETIS UMR 8051, CY Cergy Paris Universit\\'e, ENSEA

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [application](<https://devfeed.tech/tags/application.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [doppler](<https://devfeed.tech/tags/doppler.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [fewer](<https://devfeed.tech/tags/fewer.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [image](<https://devfeed.tech/tags/image.md>), [improvements](<https://devfeed.tech/tags/improvements.md>), [layer](<https://devfeed.tech/tags/layer.md>), [orientation](<https://devfeed.tech/tags/orientation.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [peft](<https://devfeed.tech/tags/peft.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [probing](<https://devfeed.tech/tags/probing.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [radar](<https://devfeed.tech/tags/radar.md>), [shift](<https://devfeed.tech/tags/shift.md>), [vision](<https://devfeed.tech/tags/vision.md>)

### AI overview

In the reported radar micro-Doppler HAR evaluations, parameter-efficient fine-tuning (PEFT) avoided two failures seen with alternatives: full fine-tuning hurt cross-dataset transfer, while frozen features under-adapted across orientations. DINOv3 adapted with PiSSA ranked first among PEFT methods in every tested setting. Its Small variant matched the prior Base-size PEFT state of the art in in-distribution accuracy with 4.5 times fewer total parameters and only 3% trainable; it also beat a size-matched baseline across evaluated settings, with transfer gains of 2.1, 6.6, and 8.9 percentage points as shift severity increased. Ablations credited gains to both pretraining and adapter design, with pretraining contributing most under compound shift.

### Source excerpt

arXiv:2611.25874v1 Announce Type: new Abstract: Micro-Doppler radar enables contactless, privacy-preserving human activity recognition (HAR), but deployment robustness remains underexplored. We compared full fine-tuning, frozen probing, and parameter-efficient fine-tuning (PEFT) using subject-disjoint in-distribution cross-validation and few-shot cross-age, cross-orientation, and cross-dataset transfer. Full fine-tuning suffered negative cross-dataset transfer, while frozen features under-adapted on cross-orientation. Only PEFT avoided both failures. We paired DINOv3 with all-layer principal-singular-direction adaptation (PiSSA), to our knowledge the first application of a self-supervised image foundation backbone to radar micro-Doppler HAR. DINOv3-PiSSA ranked first among PEFT methods in every evaluated setting. Under our protocol, its Small variant matched the in-distribution accuracy of the prior Base-size PEFT state of the art using 4.5x fewer total parameters, with only 3% trainable. It surpassed a size-matched variant of that baseline in every evaluated setting, with transfer gains increasing with shift severity: 2.1, 6.6, and 8.9 percentage points, respectively. Ablations attributed improvements to both pretraining and adapter recipe, with the pretraining contributing most under compound shift. Code and protocol are available at https://github.com/KeryanChelouche/udoppler-ssl.

## How Far Does Online RL Generalize in Mobile GUI Agents? A Controlled Study with AndroidWorld-Generalization

DevFeed: [How Far Does Online RL Generalize in Mobile GUI Agents? A Controlled Study with AndroidWorld-Generalization](<https://devfeed.tech/articles/how-far-does-online-rl-generalize-in-mobile-gui-agents-a-controlled-study-with-androidworld-generalization-143237.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.07432>)

Author: Li Gu, Zihuan Jiang, Zhixiang Chi, Huan Liu, Ziqiang Wang, Yuanhao Yu, Linqiang Guo, Yang Wang

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-hc](<https://devfeed.tech/tags/cs-hc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [gui](<https://devfeed.tech/tags/gui.md>), [gui-agents](<https://devfeed.tech/tags/gui-agents.md>), [mobile](<https://devfeed.tech/tags/mobile.md>), [online](<https://devfeed.tech/tags/online.md>), [ppo](<https://devfeed.tech/tags/ppo.md>), [qwen3](<https://devfeed.tech/tags/qwen3.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [release](<https://devfeed.tech/tags/release.md>), [rl](<https://devfeed.tech/tags/rl.md>), [test-time-adaptation](<https://devfeed.tech/tags/test-time-adaptation.md>)

### AI overview

In AndroidWorld-Generalization, online RL gains shrink as mobile GUI tasks move farther from training contexts: from unseen instances to templates to applications. Using the same RL algorithm with UI-TARS-7B-SFT and Qwen3-VL-8B, the study reports gains of +17.5/+11.1/+4.2 and +10.3/+7.4/+0, respectively. Training performance kept rising while evaluation performance saturated; the decay was also reproduced with PPO, and RL helped less with the stronger prior. Few-shot adaptation in the target application recovered +22.9, suggesting a complement to further RL training. The authors release the benchmark, training system, recipes, and model weights.

### Source excerpt

arXiv:2603.07432v2 Announce Type: replace-cross Abstract: Online reinforcement learning (RL) has advanced mobile GUI agents on AndroidWorld to over 80\% success, but these systems train and evaluate in the same environment, and their training tasks are not released for auditing. How far RL generalizes beyond its training contexts therefore remains unmeasured. We formalize the mobile environment as a Contextual MDP and instantiate a ladder of three progressively distant context shifts as AndroidWorld-Generalization: Unseen Instance, Unseen Template, and Unseen Application, with separation verified manually. In addition, we provide an open RL training system that makes studies at this scale feasible. Applying the identical RL algorithm to two priors of different origin and competence, UI-TARS-7B-SFT at 23.0\% and Qwen3-VL-8B at 52.6\%, we find that RL improves both in distribution but that gains fall monotonically up the ladder (+17.5/+11.1/+4.2 and +10.3/+7.4/+0), with training performance rising throughout while evaluation performance saturates. The decay is algorithm-agnostic, reproduced under PPO, and RL contributes less as the prior strengthens. Few-shot adaptation in the target application recovers +22.9, motivating test-time adaptation as a complement to further RL training. We release the benchmark, training system, recipes, and model weights.

## GRAFT: Graph-Regularized Prototype Fields for Realistic Test-Time Adaptation of VLMs

DevFeed: [GRAFT: Graph-Regularized Prototype Fields for Realistic Test-Time Adaptation of VLMs](<https://devfeed.tech/articles/graft-graph-regularized-prototype-fields-for-realistic-test-time-adaptation-of-vlms-143408.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23591>)

Author: Maxx Richard Rahman, Wolfgang Maass

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [vlm](<https://devfeed.tech/topics/vlm.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [anchor](<https://devfeed.tech/tags/anchor.md>), [bayes](<https://devfeed.tech/tags/bayes.md>), [class](<https://devfeed.tech/tags/class.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [features](<https://devfeed.tech/tags/features.md>), [fields](<https://devfeed.tech/tags/fields.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [graph](<https://devfeed.tech/tags/graph.md>), [inference](<https://devfeed.tech/tags/inference.md>), [markov](<https://devfeed.tech/tags/markov.md>), [model](<https://devfeed.tech/tags/model.md>), [online](<https://devfeed.tech/tags/online.md>), [protocols](<https://devfeed.tech/tags/protocols.md>), [prototype](<https://devfeed.tech/tags/prototype.md>), [prototypes](<https://devfeed.tech/tags/prototypes.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [test-time-adaptation](<https://devfeed.tech/tags/test-time-adaptation.md>), [vision-language-model](<https://devfeed.tech/tags/vision-language-model.md>), [vlms](<https://devfeed.tech/tags/vlms.md>), [zero-shot](<https://devfeed.tech/tags/zero-shot.md>)

### AI overview

GRAFT addresses cases where test-time adaptation of vision-language models underperforms zero-shot classification because batches contain few classes or online streams are correlated. It represents adapted prototypes as text anchors plus class-specific displacements, partially pooled across a semantic class graph using a Gaussian Markov random field prior, with an explicit model of class presence. Inference is closed-form and gradient-free on cached features. The prior spans anchor-only shrinkage and a single global shift; absent classes are harmonic extensions of present classes, and pooling strength can be selected by empirical Bayes. The abstract reports that GRAFT improves on zero-shot classification across realistic batch and online protocols on eleven benchmarks and four ImageNet shifts, and beats anchor-based transduction when few classes are present.

### Source excerpt

arXiv:2611.23591v1 Announce Type: cross Abstract: Unlabeled test data can improve the zero-shot classifier of a vision-language model, yet transductive and test-time adaptation methods often fall below the zero-shot baseline when only a few vocabulary classes are present in a batch or when streams are correlated. We trace this failure to a shared design: every class owns independent parameters estimated from its own samples, so absent or rare classes are left at biased text anchors or fitted to foreign samples. We propose Graph-Regularized Anchor-displacement Field Transduction (GRAFT), which models the adapted prototypes as text anchors plus a displacement field that is partially pooled over a semantic class graph through a Gaussian Markov random field prior, together with an explicit class-presence model. Inference is closed-form and gradient-free on cached features. Anchor-only shrinkage and a single global shift are the two extreme settings of the prior; absent classes are provably harmonic extensions of the present ones, and the pooling strength can be chosen by empirical Bayes. Under the realistic batch and online protocols on eleven benchmarks and four ImageNet shifts, GRAFT improves over the zero-shot model in every setting and over anchor-based transduction where few classes are present.

## Meta-LSQ: A Closed-Form Sobolev Inner Solver for Few-Shot Meta-Learning

DevFeed: [Meta-LSQ: A Closed-Form Sobolev Inner Solver for Few-Shot Meta-Learning](<https://devfeed.tech/articles/meta-lsq-a-closed-form-sobolev-inner-solver-for-few-shot-meta-learning-143547.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24718>)

Author: Antonin Sulc

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [few-shot](<https://devfeed.tech/topics/few-shot.md>), [few-shot-learning](<https://devfeed.tech/topics/few-shot-learning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>)

Tags: [basis](<https://devfeed.tech/tags/basis.md>), [beam](<https://devfeed.tech/tags/beam.md>), [cross-validation](<https://devfeed.tech/tags/cross-validation.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-lo](<https://devfeed.tech/tags/cs-lo.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [meta](<https://devfeed.tech/tags/meta.md>), [random](<https://devfeed.tech/tags/random.md>), [regression](<https://devfeed.tech/tags/regression.md>), [sensor](<https://devfeed.tech/tags/sensor.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [solve](<https://devfeed.tech/tags/solve.md>)

### AI overview

Meta-LSQ fits each task with one regularised least-squares solve over a fixed random Fourier basis, meta-learning only bandwidth, ridge and derivative weight. It can use derivative observations and yields exact closed-form derivatives. On 10-shot sinusoid regression, it trained in 14 seconds and achieved 5.5 times lower error than a first-order MAML baseline that trained 36 times longer; a Gaussian-process control suggests shared-kernel learning explains much of the gain. In simulated DESY beam tuning, it slightly outperformed an ALPaCA-style model when model derivatives were available and exceeded gradient descent. Across the full simulated target-beam range, its best setting had 7.3 μm median error versus 14.4 μm for the published final beams of a reinforcement-learning agent, though that agent had a lower median on 300 small-target trials. On real sensor-calibration data, Meta-LSQ beat MAML on two of four targets, while an exact Gaussian process with the same prior-mean change and an ALPaCA-style model were more accurate on most targets. It also broke down on non-smooth targets; for noiseless targets outside the basis span, generalised cross-validation selected too little ridge.

### Source excerpt

arXiv:2611.24718v1 Announce Type: new Abstract: Gradient-based meta-learners such as model-agnostic meta-learning (MAML) adapt to a new task with an inner loop of gradient descent that has to be differentiated through, and they meta-learn every network weight. Meta-LSQ instead fits each task with one regularised least-squares solve over a fixed random Fourier basis and meta-learns only three scalars, a bandwidth, a ridge and a derivative weight. Since the derivative of a sinusoid is again a sinusoid, the same solve can also take derivative observations, and the fitted model has exact closed-form derivatives of any order. On 10-shot sinusoid regression Meta-LSQ trains in 14 s and ends with a 5.5 times lower error than a first-order MAML baseline that trains 36 times longer, and a Gaussian-process control suggests that most of this gain comes from sharing one kernel across tasks. In simulated beam tuning at DESY's ARES facility with model derivatives it is slightly more accurate than an ALPaCA-style model and well ahead of gradient descent. Over DESY's full range of target beams, again in simulation and from measurements alone, its best setting has a median error of 7.3 $\mu$m, against 14.4 for the published final beams of their reinforcement-learning agent, which has the lower median on their 300 trials of small targets. On real sensor-calibration data a learned prior mean puts it ahead of MAML on two of four targets, but an exact Gaussian process with the same change and an ALPaCA-style model stay more accurate on most targets, and it breaks down on non-smooth targets. On noiseless targets outside the span of the basis, in-sample criteria such as generalised cross-validation choose too small a ridge.

## JEPA as a Transferable Optimizer: Learning and Understanding Black-Box Optimization from Reward-Agnostic Pretraining

DevFeed: [JEPA as a Transferable Optimizer: Learning and Understanding Black-Box Optimization from Reward-Agnostic Pretraining](<https://devfeed.tech/articles/jepa-as-a-transferable-optimizer-learning-and-understanding-black-box-optimization-from-reward-agnostic-pretraining-142678.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23608>)

Author: Mayukh Debnath, Manan Roy Choudhury

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [model architecture](<https://devfeed.tech/topics/model-architecture.md>), [Self-supervised learning](<https://devfeed.tech/topics/self-supervised-learning.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [bayesian](<https://devfeed.tech/tags/bayesian.md>), [black-box](<https://devfeed.tech/tags/black-box.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [history](<https://devfeed.tech/tags/history.md>), [information](<https://devfeed.tech/tags/information.md>), [learning](<https://devfeed.tech/tags/learning.md>), [llm](<https://devfeed.tech/tags/llm.md>), [memory](<https://devfeed.tech/tags/memory.md>), [model](<https://devfeed.tech/tags/model.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [policy](<https://devfeed.tech/tags/policy.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [random](<https://devfeed.tech/tags/random.md>), [reward](<https://devfeed.tech/tags/reward.md>)

### AI overview

JEPA-PS learns to propose black-box optimization candidates from short evaluation histories, predicting the next point in a trajectory rather than rewards or gradients. Pretrained on random rollouts with parameter vectors and scalar returns, it proposes candidates in one forward pass without task descriptions, reward shaping, or test-time gradients. Across 15 control environments and 12 optimizers, it was competitive with tuned classical baselines, transferred zero-shot to environments excluded from pretraining, and beat an LLM optimizer by more than 2x on Swimmer at 71x lower per-step cost. Its useful history varied by landscape: smooth tasks benefited from the full trajectory, while on rugged tasks the best-seen point carried nearly all the signal. Evidence that this pattern relates to fitness-distance correlation is weak and was not corroborated by a separate reward-ordering probe. Decoder ablations also found every trained decoder variant worse than the default frozen representation.

### Source excerpt

arXiv:2611.23608v1 Announce Type: cross Abstract: Black-box optimization evaluates candidates through simulations, rollouts, or experiments, often without gradients. Learned optimizers are usually judged by whether they outperform existing methods. We ask a complementary question: what information does a learned optimizer use from its search history? We introduce JEPA-PS (JEPA Policy Search), based on a Joint-Embedding Predictive Architecture. It is pretrained on random rollouts using parameter vectors and scalar returns. Its latent-prediction objective predicts the next point in an optimization trajectory rather than its reward or gradient, allowing us to examine what was tried and how well it performed. At test time, JEPA-PS reads a short evaluation history and proposes a candidate in one forward pass, without task descriptions, reward shaping, or test-time gradients. We evaluate it on 15 control environments against 12 optimizers, including evolutionary methods, Bayesian optimization, CMA-ES, and an LLM optimizer. JEPA-PS is competitive with tuned classical baselines and outperforms the LLM optimizer by more than 2x on Swimmer at 71x lower per-step cost. It also transfers zero-shot to environments excluded from pretraining. Null-model comparisons and decoder-training ablations indicate that its latent representation carries information beyond simple history statistics; every trained decoder variant performs worse than the default frozen representation. Our main finding is that the history JEPA-PS uses depends on the task. On smooth landscapes it uses the full trajectory, whereas on rugged landscapes the best-seen point carries nearly all the signal. This memory requirement has a weak positive correlation with fitness-distance correlation (FDC; r=0.112, n=7), while an independent reward-ordering sensitivity probe does not corroborate the relationship (r=-0.024, n=7). Code: https://github.com/mdebnath22/jepa-props.

## Improving LLM Final Representations with Inter-Layer Geometry

DevFeed: [Improving LLM Final Representations with Inter-Layer Geometry](<https://devfeed.tech/articles/improving-llm-final-representations-with-inter-layer-geometry-143245.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.22665>)

Author: Tom Ulanovski, Eyal Blyachman, Maya Bechler-Speicher

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [graph theory](<https://devfeed.tech/topics/graph-theory.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [lora](<https://devfeed.tech/topics/lora.md>)

Tags: [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diameter](<https://devfeed.tech/tags/diameter.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [encoder](<https://devfeed.tech/tags/encoder.md>), [explainability](<https://devfeed.tech/tags/explainability.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [fully-connected](<https://devfeed.tech/tags/fully-connected.md>), [gnn](<https://devfeed.tech/tags/gnn.md>), [intermediate](<https://devfeed.tech/tags/intermediate.md>), [layers](<https://devfeed.tech/tags/layers.md>), [llm](<https://devfeed.tech/tags/llm.md>), [lora](<https://devfeed.tech/tags/lora.md>), [multiple](<https://devfeed.tech/tags/multiple.md>), [overfitting](<https://devfeed.tech/tags/overfitting.md>), [performance](<https://devfeed.tech/tags/performance.md>), [topology](<https://devfeed.tech/tags/topology.md>)

### AI overview

Using intermediate LLM layers can improve downstream prediction over relying only on the final-layer representation. The work first applies a lightweight GNN over a fully connected graph of layers, then introduces Cayley-Encoder, which uses a sparse, regular, low-diameter Cayley graph over SL(2, Zn) to make aggregation more efficient and constrain its structure. Across 13 tasks and 13 LLMs, Cayley-Encoder reportedly outperforms baselines, with accuracy gains up to 48 percentage points and added parameters capped at 0.1% of LLM size. The abstract also reports effectiveness in few-shot settings and better performance than LoRA fine-tuning while keeping the LLM frozen. Its explainability analysis finds meaningful contributions from multiple layers. These are abstract-level claims; the supplied material gives no task-specific results or experimental details.

### Source excerpt

arXiv:2603.22665v4 Announce Type: replace-cross Abstract: The standard in LLM-based prediction is to use the final-layer representation as the input to a downstream predictor. However, intermediate layers may encode complementary task-relevant signals. Existing approaches therefore either search for the best layer for each task or apply expensive attention-based mechanisms to learn inter-layer aggregation. In this work, we first show that such complexity is unnecessary: a lightweight Graph Neural Network over a fully connected graph of LLM layers is more efficient and achieves significantly stronger predictive performance than existing approaches. We then introduce the Cayley-Encoder, which further improves both efficiency and predictive performance by replacing the fully connected graph with a Cayley graph over SL(2, Zn). These Cayley graphs provide a mathematically grounded topology that is sparse, regular by construction, and has low diameter. This enables effective communication across layers while constraining the aggregation structure and reducing the risk of GNN overfitting. In an evaluation of Cayley-Encoder across 13 tasks and 13 LLMs, Cayley-Encoder consistently outperforms baselines, achieving improvements of up to 48 percentage points in accuracy, while introducing at most 0.1% additional parameters relative to the LLM size. We further show that Cayley-Encoder is effective in few-shot regimes. Finally, we show that Cayley-Encoder outperforms LoRA fine-tuning while operating on the frozen LLM. We conclude with an explainability analysis showing that multiple layers contribute meaningfully to the final prediction, supporting our hypothesis.

## Towards Pose-Robust Few-Shot Vision-Language-Action Models

DevFeed: [Towards Pose-Robust Few-Shot Vision-Language-Action Models](<https://devfeed.tech/articles/towards-pose-robust-few-shot-vision-language-action-models-142785.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24360>)

Author: Wenhao Li, Zhongyi Han, Yuhao Wei, Liang Han, Yilong Yin, Liqiang Nie

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Robot Navigation](<https://devfeed.tech/topics/robot-navigation.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [anchor](<https://devfeed.tech/tags/anchor.md>), [arm](<https://devfeed.tech/tags/arm.md>), [base](<https://devfeed.tech/tags/base.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [camera](<https://devfeed.tech/tags/camera.md>), [canonicalization](<https://devfeed.tech/tags/canonicalization.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [data](<https://devfeed.tech/tags/data.md>), [data-collection](<https://devfeed.tech/tags/data-collection.md>), [estimator](<https://devfeed.tech/tags/estimator.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [frame](<https://devfeed.tech/tags/frame.md>), [framework](<https://devfeed.tech/tags/framework.md>), [image](<https://devfeed.tech/tags/image.md>), [language](<https://devfeed.tech/tags/language.md>), [models](<https://devfeed.tech/tags/models.md>), [robust](<https://devfeed.tech/tags/robust.md>), [vision](<https://devfeed.tech/tags/vision.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vision-language-action-models](<https://devfeed.tech/tags/vision-language-action-models.md>)

### AI overview

WACO addresses camera and robot-base shifts that disrupt few-shot vision-language-action policies by treating both as changes in the camera-to-base relationship. From one RGB-D frame, it uses the robot arm's forward-kinematic pose as a metric calibration target and the static scene as a world anchor, then canonicalizes the image, state, action, and start pose to the original calibration frames. It requires neither ground-truth extrinsics nor extra data, and fine-tunes the unmodified policy using artifacts from its own estimator. The authors introduce LIBERO-PoseShift, a benchmark with graded camera, base, and combined shifts. Across that benchmark, LIBERO-Plus's camera track, and four UR5e real-world tasks, WACO reportedly improves shifted success by 13.2 points in simulation and 36.7 points on the robot, recovering 90% and 87% of the respective gaps to a ground-truth oracle, without test-time calibration or perturbed training data.

### Source excerpt

arXiv:2611.24360v1 Announce Type: cross Abstract: Fine-tuning a vision-language-action (VLA) model on a few dozen demonstrations is the standard way to put a manipulation policy on a robot, yet the resulting policy silently assumes that the camera and the robot base stay exactly where they were during data collection: a bumped tripod or a base re-docked a few centimetres away makes it reach for the wrong place. We show that both failures are one phenomenon, a change of the camera-to-base extrinsic that the policy has memorised, and that the robot carries everything needed to undo it. We introduce WACO, a world-anchored canonicalization framework for VLA models: the arm, placed by forward kinematics at the measured joint angles, serves as a metric calibration target, and the static scene serves as the world anchor that tells a moved camera from a moved robot. From a single RGB-D frame, WACO estimates both shifts without ground-truth extrinsics or extra data, canonicalises the image to the calibration viewpoint and the state, action and start pose to the calibration base frame, and fine-tunes the unmodified policy under the artefacts of its own estimator. We also introduce LIBERO-PoseShift, the first benchmark for RGB VLAs with graded camera, base and compound shifts. Across LIBERO-PoseShift, the LIBERO-Plus camera track and four real-world tasks on a UR5e robot, WACO consistently outperforms the standard few-shot recipe and prior viewpoint-robust methods, raising shifted success by 13.2 points in simulation and by 36.7 points on the real robot, recovering 90% of the simulation gap to a ground-truth oracle and 87% of the real-robot gap, and establishing a new state of the art among methods that use no perturbed training data, with no test-time calibration. Project page: https://peacelwh.github.io/WACO/.

## When Does Subspace Direction Matter for LoRA? Regime Analysis of the Magnitude Principle in Few-Shot Adaptation

DevFeed: [When Does Subspace Direction Matter for LoRA? Regime Analysis of the Magnitude Principle in Few-Shot Adaptation](<https://devfeed.tech/articles/when-does-subspace-direction-matter-for-lora-regime-analysis-of-the-magnitude-principle-in-few-shot-adaptation-143392.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23428>)

Author: Nischal Subedi, Cencheng Shen, Peng Zhao

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [lora](<https://devfeed.tech/topics/lora.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [bert](<https://devfeed.tech/tags/bert.md>), [classification](<https://devfeed.tech/tags/classification.md>), [completion](<https://devfeed.tech/tags/completion.md>), [covariance](<https://devfeed.tech/tags/covariance.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dimension](<https://devfeed.tech/tags/dimension.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [llama](<https://devfeed.tech/tags/llama.md>), [lora](<https://devfeed.tech/tags/lora.md>), [magnitude](<https://devfeed.tech/tags/magnitude.md>), [span](<https://devfeed.tech/tags/span.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [structure](<https://devfeed.tech/tags/structure.md>), [subspace](<https://devfeed.tech/tags/subspace.md>), [theory](<https://devfeed.tech/tags/theory.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

The paper finds that LoRA initialization direction matters most when labeled data are scarce relative to representation dimension and the initialization subspace captures task structure. Its theory includes a finite-sample stability analysis and a first-step account that motivates an alignment coefficient, ρ, which tracks classification gains. PivotLoRA uses activation PCA at prediction-relevant positions, then applies an output-preserving orthogonal pivot early in training. In few-shot BERT-base classification it ranks best on average among 13 methods and beats vanilla LoRA in 29 of 30 settings despite a smaller input-factor initialization norm. Gains diminish with more labeled data; targeted experiments also report improvements on GSM8K and StrategyQA. The supplied text does not detail those evaluation setups or effect sizes.

### Source excerpt

arXiv:2611.23428v1 Announce Type: new Abstract: We show that the importance of a LoRA adapter's initial input-subspace direction is regime-dependent, and we characterize this dependence in theory and experiment. This refines the recent view that spectral initializers help through update magnitude alone. Direction matters most when labeled data are scarce relative to the representation dimension and the activation subspace used for initialization captures supervised task structure. In this regime, the adapter's row span remains sensitive to its starting point, making task-aligned initialization especially useful. A spiked-covariance perturbation analysis characterizes the finite-sample stability of the estimated subspace. A first-step analysis of LoRA training explains why the subspace's alignment with the task matters and motivates an alignment coefficient $\rho$ that closely tracks gains across classification tasks. We test this account with PivotLoRA, which initializes adapters from activation PCA at prediction-relevant positions and applies one output-preserving orthogonal pivot early in training. In few-shot BERT-base classification, PivotLoRA achieves the best average rank among 13 methods and improves over vanilla LoRA in 29 of 30 settings, even though it starts from a smaller input-factor initialization norm. Additional classification experiments extend the evaluation to 7B and 8B Llama decoders. Classification gains diminish with more labeled data, consistent with the proposed regime dependence. A targeted extension to completion-only reasoning also shows improvements on GSM8K and StrategyQA. These findings identify conditions under which initialization direction matters, with subspace-task alignment offering a low-cost diagnostic before training for classification.

## Measuring and Improving Transfer of Neural Surrogates for Vehicle Aerodynamics

DevFeed: [Measuring and Improving Transfer of Neural Surrogates for Vehicle Aerodynamics](<https://devfeed.tech/articles/measuring-and-improving-transfer-of-neural-surrogates-for-vehicle-aerodynamics-143571.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24843>)

Author: Favour Nerrise, Julian Busch, Gurpreet Singh Hora, Martin Sipka, Danielle C. Maddix, Georgia Olympia Brikis

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Computational Fluid Dynamics (CFD)](<https://devfeed.tech/topics/cfd.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>)

Tags: [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [cfd](<https://devfeed.tech/tags/cfd.md>), [computational-fluid-dynamics](<https://devfeed.tech/tags/computational-fluid-dynamics.md>), [computational-fluid-dynamics-cfd](<https://devfeed.tech/tags/computational-fluid-dynamics-cfd.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [domain](<https://devfeed.tech/tags/domain.md>), [error](<https://devfeed.tech/tags/error.md>), [external](<https://devfeed.tech/tags/external.md>), [fast](<https://devfeed.tech/tags/fast.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [independent](<https://devfeed.tech/tags/independent.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [neural](<https://devfeed.tech/tags/neural.md>), [physics](<https://devfeed.tech/tags/physics.md>), [physics-flu-dyn](<https://devfeed.tech/tags/physics-flu-dyn.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [results](<https://devfeed.tech/tags/results.md>), [risk](<https://devfeed.tech/tags/risk.md>), [vehicle](<https://devfeed.tech/tags/vehicle.md>)

### AI overview

Across four target datasets that vary in geometry, CFD solver, turbulence model, and flow conditions, fine-tuning five DrivAerML-pretrained vehicle-aerodynamics surrogates beat training from scratch in fewer than half the comparisons and reduced accuracy on the source domain. The authors introduce GeoReplay, which retains source simulations geometrically similar to the target and applies their updates only when aligned with the target gradient. It matches fine-tuning's target accuracy while reducing its source error in 77 of 80 settings. The study is based on these benchmark settings; the abstract does not report broader validation.

### Source excerpt

arXiv:2611.24843v1 Announce Type: new Abstract: Neural surrogates for external vehicle aerodynamics are fast alternatives to physics-based computational fluid dynamics (CFD) simulations. Their foundational limitation is a lack of zero-shot generalization across vehicle classes, which results in an expensive, custom training campaign for each new vehicle design. Similar to pretrained models in other fields, a pretrained surrogate should adapt to a new vehicle class using only a few simulations and stay accurate on its pretraining domain. No prior study has systematically tested this transfer behavior. We conduct a large-scale, controlled benchmarking study on five DrivAerML-pretrained surrogates across four independent target datasets. These datasets differ from the source in geometry, CFD solver, turbulence model, and flow conditions. We find that fine-tuning these surrogates outperforms training from scratch in fewer than half of the comparisons and lowers the accuracy of the surrogate on its source dataset. We introduce GeoReplay, a geometry- and gradient-aligned source replay that keeps source simulations resembling the target in training and applies their updates only when they agree with the target gradient. A risk bound separates geometric coverage from differences in the simulated physics, and we prove that every sufficiently small GeoReplay step decreases the target training objective. GeoReplay lowers the source error of fine-tuning in 77 of 80 settings and matches its target accuracy.

## Weight-Space Programming Doesn't Need Transformers

DevFeed: [Weight-Space Programming Doesn't Need Transformers](<https://devfeed.tech/articles/weight-space-programming-doesn-t-need-transformers-139953.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22055>)

Author: Hyunjong Ok, Jaeho Lee

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Frontier AI](<https://devfeed.tech/topics/frontier-ai.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [few-shot-learning](<https://devfeed.tech/topics/few-shot-learning.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [gradient-descent](<https://devfeed.tech/tags/gradient-descent.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [memory](<https://devfeed.tech/tags/memory.md>), [table](<https://devfeed.tech/tags/table.md>), [transformer](<https://devfeed.tech/tags/transformer.md>), [weight-space-programming](<https://devfeed.tech/tags/weight-space-programming.md>)

### AI overview

In weight-space programming, an LLM writes model weights directly, so the authors test whether it should design a model rather than fill a fixed Transformer. With minimal guidance, ten labeled examples, and a sandbox, three frontier LLMs mainly solve multiple-choice tasks by encoding plain-text facts in a sparse table queried by a fixed procedure, reaching up to 54.4% test accuracy. One LLM-designed model beats the most accurate released Transformer written by that same LLM using about 1,800 times fewer parameters. The table makes weights interpretable and lets the LLM locate and rewrite errors; removing a row affects about one other answer. The authors propose an associative-memory model reported up to 17 times faster on GPUs than a same-sized Transformer. Pretrained LLMs remain more accurate, and the structural claim is specifically about models whose weights are written directly; the abstract also mentions tests on other benchmarks and answer-generating tasks without reporting their results.

### Source excerpt

arXiv:2611.22055v1 Announce Type: cross Abstract: Standard neural architectures such as the Transformer were chosen because they train well with gradient descent and run fast on GPUs. In weight-space programming, a large language model (LLM) writes target weights by hand without gradient-based training, so those reasons no longer apply. Should we keep the Transformer? Given minimal guidance, ten labeled examples, and a sandbox, mainly on multiple-choice questions, we task three frontier LLMs with either generating the weights of a fixed architecture or designing the architecture themselves. When free to design, all three compile plain-text facts into a sparse table queried by a fixed procedure, reaching up to 54.4% test accuracy. An LLM-designed model achieves higher test accuracy than the most accurate released Transformer that the same LLM wrote while using about 1,800x fewer parameters. Without gradients, they choose a table, in which the LLM can state what each weight means and can find and rewrite a wrong one. Each wrong answer points to one row, and removing a row changes about one other answer. From this, we propose an associative-memory-based model up to 17x faster on GPUs than a same-sized Transformer. We also test pretrained weights, other benchmarks, and tasks where models generate answers. Pretrained LLMs remain more accurate; the claim concerns which structure to build when writing weights.

## Feature Selection for Tabular Foundation Models: Exploring Feature Utility in Low-Data Molecular Prediction

DevFeed: [Feature Selection for Tabular Foundation Models: Exploring Feature Utility in Low-Data Molecular Prediction](<https://devfeed.tech/articles/feature-selection-for-tabular-foundation-models-exploring-feature-utility-in-low-data-molecular-prediction-141605.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.30789>)

Author: Yiqi Yao, Miquel Duran-Frigola

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Feature Engineering](<https://devfeed.tech/topics/feature-engineering.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Predictive AI](<https://devfeed.tech/topics/predictive-ai.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data-provenance](<https://devfeed.tech/tags/data-provenance.md>), [feature-selection](<https://devfeed.tech/tags/feature-selection.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [mol](<https://devfeed.tech/tags/mol.md>), [tabpfn](<https://devfeed.tech/tags/tabpfn.md>)

### AI overview

In few-shot molecular property prediction with TabPFN, broad RDKit descriptor and fingerprint families provide a strong default across 77 FS-Mol development tasks and 157 held-out ChEMBL assays. Rank-averaging TabPFN with support-fitted fingerprint forests outperforms either model alone, while learned weighting of family predictions adds gains at a larger prediction budget. A utility-trained selector produces a traceable 16-column input from 238 source-assay predictors: it beats the baseline but trails wider representations and equal-width structural SVD coordinates. On Therapeutics Data Commons ADME/Tox assays, endpoint-matched assay predictions improve a compact RDKit baseline; a provenance audit finds most audited query molecules already in predictor training sources. The study also tests selected outputs through paired removal and replacement, assessing each alongside the other inputs.

### Source excerpt

arXiv:2609.30789v4 Announce Type: replace Abstract: Choosing features is difficult with few labeled molecules, even for a tabular foundation model that learns in context. We study this problem with TabPFN for low-data molecular property prediction, separating input width, feature meaning, and predictive usefulness while keeping contexts, query molecules, and predictor settings matched. On ADME/Tox assays from Therapeutics Data Commons, endpoint-matched assay predictions improve a compact RDKit descriptor baseline, and a provenance audit shows that most audited query molecules already occur in the predictors' training sources. Across 77 FS-Mol development tasks and 157 held-out test assays drawn from ChEMBL, broad RDKit descriptor and fingerprint families provide a strong default, rank-averaging TabPFN with support-fitted fingerprint forests improves upon both models, and learned weighting of family predictions adds further gains at a larger prediction budget. A utility-trained selector builds a traceable 16-column input from 238 sourceassay predictors, improving upon the baseline while trailing wider representations and equalwidth structural SVD coordinates. Named outputs identify what each feature predicts and where it comes from. Their usefulness for a new assay is assessed alongside the other input features, and paired removal and replacement tests measure what each selected output contributes. Together, these studies offer a protocol for evaluating feature selection, representation choice, and data provenance with a fixed tabular foundation model for few-shot molecular property prediction.

## OCCAM: Improved Single-Cell Representation Learning and Perturbation Effect Prediction

DevFeed: [OCCAM: Improved Single-Cell Representation Learning and Perturbation Effect Prediction](<https://devfeed.tech/articles/occam-improved-single-cell-representation-learning-and-perturbation-effect-prediction-139939.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21959>)

Author: Ihab Bendidi, Kian Kenyon-Dean, Alisandra K. Denton, Auguste Genovesio, Emmanuel Noutahi

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [mlp](<https://devfeed.tech/topics/mlp.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [mlp](<https://devfeed.tech/tags/mlp.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [q-bio-qm](<https://devfeed.tech/tags/q-bio-qm.md>), [representation-learning](<https://devfeed.tech/tags/representation-learning.md>), [single-cell](<https://devfeed.tech/tags/single-cell.md>)

### AI overview

Occam is an MLP framework that combines denoising-learned cell representations with gene interaction, text, and protein-sequence features to predict genetic perturbation effects. The abstract reports the highest aggregate representation scores among evaluated methods for perturbation information and cell-type preservation with batch integration. In few-shot tests where perturbation and cell context were separately seen but not together, it captured measured response patterns better than four retrained competitors and direct transfer from other cell lines. Without fine-tuning, it also best distinguished perturbations in held-out cell lines across seen and unseen assays, though accurate expression-level prediction remained challenging. For double perturbations, composing learned single-gene responses reduced expression error versus competing models and simply summing single-gene effects, including for strongly and weakly additive responses.

### Source excerpt

arXiv:2611.21959v1 Announce Type: cross Abstract: Predicting the effects of genetic perturbations requires models that generalize beyond observed cell contexts and gene combinations. We introduce Occam, a multilayer perceptron (MLP) framework for representation learning and perturbation-effect prediction. Occam combines cell representations learned through denoising with gene interaction, text, and protein-sequence features. It achieves the highest aggregate representation scores for perturbation information and cell-type preservation with batch integration. In cross-context few-shot prediction, a perturbation and a cell context are each observed during training, but never together. Occam better captures the measured response patterns for these unobserved combinations than four retrained competitors or direct transfer of the same perturbation's measured effects from other training cell lines. Without fine-tuning, Occam best distinguishes perturbations among the evaluated models and simple baselines in cell lines held out from predictor training, across seen and unseen assays. Predicting expression levels accurately in these settings remains challenging. For double perturbations, which target two genes simultaneously, Occam composes learned single-gene responses to better capture the combined response and reduce expression error compared with competing models and summed single-gene effects. These gains extend to both strongly additive and weakly additive responses. Together, these results support simple MLPs for cell representation learning and perturbation-effect prediction across contexts and gene combinations.

## MAGIC-SSCIL: Manifold Anchoring and Geometric Incremental Calibration for Semi-Supervised Class Incremental Learning

DevFeed: [MAGIC-SSCIL: Manifold Anchoring and Geometric Incremental Calibration for Semi-Supervised Class Incremental Learning](<https://devfeed.tech/articles/magic-sscil-manifold-anchoring-and-geometric-incremental-calibration-for-semi-supervised-class-incremental-learning-141590.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2608.07586>)

Author: Yousef Abdi, Mohammad Asadpour, Yousef Seyfari

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [class](<https://devfeed.tech/tags/class.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cub](<https://devfeed.tech/tags/cub.md>), [features](<https://devfeed.tech/tags/features.md>), [fine](<https://devfeed.tech/tags/fine.md>), [graph](<https://devfeed.tech/tags/graph.md>), [incremental](<https://devfeed.tech/tags/incremental.md>), [learning](<https://devfeed.tech/tags/learning.md>), [magic](<https://devfeed.tech/tags/magic.md>), [model](<https://devfeed.tech/tags/model.md>), [nearest-neighbor](<https://devfeed.tech/tags/nearest-neighbor.md>), [phantom](<https://devfeed.tech/tags/phantom.md>), [resnet](<https://devfeed.tech/tags/resnet.md>), [semi](<https://devfeed.tech/tags/semi.md>), [statistics](<https://devfeed.tech/tags/statistics.md>)

### AI overview

MAGIC improves exemplar-free semi-supervised class-incremental learning by propagating labels across a k-nearest-neighbor graph of frozen features, then using propagated confidence both to weight unlabeled loss and, through Soft-Weighted Geometry Calibration, to refine Gaussian class statistics used for replay. In frozen ResNet-18 experiments on CIFAR-100, CUB-200, and ImageNet-R with 1%, 5%, and 10% labels, it gained from unlabeled data in all nine tested settings, with the largest gains at 1%, and led on CUB-200 at every label ratio. At 1%, it matched the best ImageNet-R result and trailed TACLE by 2.2 points on CIFAR-100. Ablations attribute most of the gain to weighted statistics rather than the loss; at 1% labels, graph pseudo-labels were more accurate than the model's at equal coverage, while a fixed 0.95 confidence threshold removed most of the gain. At higher label ratios, backbone-adapting or full-covariance baselines overtook MAGIC on CIFAR-100 and ImageNet-R.

### Source excerpt

arXiv:2608.07586v2 Announce Type: replace-cross Abstract: Exemplar-free semi-supervised class-incremental learning (SSCIL) asks a model to learn new classes from a few labeled and many unlabeled images without storing past data. A common recipe replays old classes from stored Gaussian statistics and pseudo-labels unlabeled data with a confidence threshold. With one or two labels per class, however, the model's confidence is a poor guide to which pseudo-labels are right, so the statistics that replay depends on rest on few or noisy samples. We propose MAGIC (Manifold Anchoring and Geometric Incremental Calibration), which propagates labels over a k-nearest-neighbor graph of frozen features and uses the propagated confidences as weights in the unlabeled loss and, through Soft-Weighted Geometry Calibration (SWGC), in the class statistics from which phantom features replay old classes. With a frozen ResNet-18 on CIFAR-100, CUB-200, and ImageNet-R at 1%, 5%, and 10% labels, we compare MAGIC with nine supervised class-incremental methods equipped with FixMatch and two native SSCIL methods. MAGIC gains from unlabeled data in all nine settings, most at 1%, and leads on fine-grained CUB-200 at every ratio; at 1% it is level with the best on ImageNet-R and 2.2 points behind the native SSCIL method TACLE on CIFAR-100. Experiments inside MAGIC show that most of this gain flows through the weighted statistics, not the loss, that at 1% labels graph pseudo-labels are more accurate than the model's own at equal coverage, and that a fixed threshold of 0.95 removes most of the gain. At higher ratios, baselines that adapt the backbone or model full covariances overtake MAGIC on CIFAR-100 and ImageNet-R.

## Network Intrusion Detection Adaptability under Spurious Labels: A Noise-Aware Few-Shot Class-Incremental Learning Framework

DevFeed: [Network Intrusion Detection Adaptability under Spurious Labels: A Noise-Aware Few-Shot Class-Incremental Learning Framework](<https://devfeed.tech/articles/network-intrusion-detection-adaptability-under-spurious-labels-a-noise-aware-few-shot-class-incremental-learning-framework-137037.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21473>)

Author: Zijian Zhao, Davide Di Monda, Domenico Ciuonzo, Yipeng Wang, Antonio Pescap\\\`e

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>), [arXiv: Networking](<https://devfeed.tech/sources/arxiv-networking.md>)

Topics: [few-shot-learning](<https://devfeed.tech/topics/few-shot-learning.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [data augmentation](<https://devfeed.tech/topics/data-augmentation.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>)

Tags: [classifiers](<https://devfeed.tech/tags/classifiers.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ni](<https://devfeed.tech/tags/cs-ni.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [detection](<https://devfeed.tech/tags/detection.md>), [edge](<https://devfeed.tech/tags/edge.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [framework](<https://devfeed.tech/tags/framework.md>), [iiot](<https://devfeed.tech/tags/iiot.md>), [incremental](<https://devfeed.tech/tags/incremental.md>), [intrusion-detection](<https://devfeed.tech/tags/intrusion-detection.md>), [iot](<https://devfeed.tech/tags/iot.md>), [logistic-regression](<https://devfeed.tech/tags/logistic-regression.md>), [memory](<https://devfeed.tech/tags/memory.md>), [nearest-neighbor](<https://devfeed.tech/tags/nearest-neighbor.md>), [network](<https://devfeed.tech/tags/network.md>), [network-traffic](<https://devfeed.tech/tags/network-traffic.md>), [noise](<https://devfeed.tech/tags/noise.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [statistics](<https://devfeed.tech/tags/statistics.md>)

### AI overview

NAFN aims to make few-shot class-incremental network intrusion detection more resilient to mislabeled examples in newly observed attack classes. It detects and removes suspected mislabeled biflows, then estimates class-conditional embeddings from the remaining clean data, calibrates them with base-class statistics, and resamples synthetic features to restore the original biflow count. The framework uses the Rethink Few-Shot teacher-student paradigm with nearest-neighbor or logistic regression classifiers and different pretraining and memory-selection strategies. On CSE-CIC-IDS2018, Edge-IIoT, and IoT-NID, the abstract reports retained prior knowledge and strong incremental performance with up to 50% label noise; new-class F1 improved over non-denoised baselines by as much as 52%. It also reports time efficiency and robustness across networks and high-noise settings.

### Source excerpt

arXiv:2611.21473v1 Announce Type: new Abstract: Network Intrusion Detection Systems (NIDSs) must incrementally adapt to emerging attacks, even when only a limited amount of labeled network traffic is available. Consequently, Few-Shot Class-Incremental Learning (FSCIL)-based NIDSs have gained growing attention. However, existing FSCIL-based NIDSs are highly vulnerable to label noise in newly-observed attack classes. To address this limitation, we propose NAFN, a noise-aware FSCIL-based NIDS framework that reduces the impact of mislabeled data during incremental adaptation. NAFN consists of two components: (i) Noise Detection, which identifies and removes mislabeled biflows using noise-detection algorithms, and (ii) Feature Sampling, which estimates class-conditional embeddings from clean biflows, calibrates them with base-class statistics, and resamples synthetic features to recover the original biflow count. The framework follows the Rethink Few-Shot (RFS) teacher-student paradigm with lightweight nearest-neighbor or logistic regression classifiers, and various pretraining & memory selection strategies. Experiments on CSE-CIC-IDS2018, Edge-IIoT, and IoT-NID datasets show that NAFN effectively preserves prior knowledge and maintains strong incremental performance even under 50% label noise, consistently improving new-class F1-scores over non-denoised baselines, up to +52%, while being time-efficient and robust in cross-network and high-noise incremental settings. The implementation is publicly available to foster reproducibility.

## Revisiting What Makes a Good Context Subset: A Sparse Latent Basis Perspective

DevFeed: [Revisiting What Makes a Good Context Subset: A Sparse Latent Basis Perspective](<https://devfeed.tech/articles/revisiting-what-makes-a-good-context-subset-a-sparse-latent-basis-perspective-137418.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19220>)

Author: Cen-Jhih Li, Siddhant Chaudhary, Aditya Bhaskara

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [diagnostic](<https://devfeed.tech/tags/diagnostic.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [hidden-states](<https://devfeed.tech/tags/hidden-states.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [layer](<https://devfeed.tech/tags/layer.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [mmr](<https://devfeed.tech/tags/mmr.md>), [search](<https://devfeed.tech/tags/search.md>), [selection](<https://devfeed.tech/tags/selection.md>), [similarity](<https://devfeed.tech/tags/similarity.md>), [spaces](<https://devfeed.tech/tags/spaces.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [zero-shot](<https://devfeed.tech/tags/zero-shot.md>)

### AI overview

The study frames few-shot demonstration selection as sparse reconstruction: Residual Explanation Search (RES) uses query-conditioned residuals to find demonstrations that collectively explain task-relevant information. Across seven benchmarks and eight open-weight LLMs, no retrieval objective wins universally; RES beats maximal marginal relevance (MMR) in 34 of 56 model-dataset pairs, with its clearest advantage on sentiment tasks. Relevance retrieval remains strong on template-heavy reasoning benchmarks, while all retrieval methods average below zero-shot on StrategyQA, where analysis suggests the query may not expose needed external knowledge. Results also depend on representation layer: middle-layer hidden states generally favor similarity retrieval, while last-layer states often favor residual- and diversity-based selection.

### Source excerpt

arXiv:2611.19220v1 Announce Type: cross Abstract: Selecting context subsets is increasingly important in modern LLM inference, yet what makes a subset effective remains unclear. We study this question in the controlled setting of few-shot demonstration selection. We examine a hypothesis that demonstrations should collectively provide task-relevant information needed by a target query. Assuming that the query representation provides a useful proxy for these requirements, we formulate selection as sparse reconstruction and instantiate it with Residual Explanation Search (RES), a Matching-Pursuit-style diagnostic procedure. Across seven benchmarks and eight open-weight LLMs, no retrieval objective is universally best. RES outperforms MMR in 34 of 56 model--dataset pairs, with its clearest advantage on sentiment tasks, suggesting that query-conditioned collective reconstruction may capture useful complementarity that pairwise dissimilarity misses. Relevance-based retrieval remains strong on template-heavy reasoning benchmarks, while all retrieval methods underperform zero-shot prediction on average on StrategyQA, where a solution-reconstruction analysis is consistent with the query not exposing the required external knowledge. Target-model hidden states further show that retrieval objectives interact differently with representation spaces: middle-layer representations generally favor similarity retrieval, whereas last-layer representations are often stronger for residual- and diversity-based selection. These findings support a conditional, representation-dependent account of context subset quality rather than a universally superior selector.

## ST-GAT: A Leakage-Free Spatio-Temporal Graph Attention Framework for Robust Network Intrusion Detection

DevFeed: [ST-GAT: A Leakage-Free Spatio-Temporal Graph Attention Framework for Robust Network Intrusion Detection](<https://devfeed.tech/articles/st-gat-a-leakage-free-spatio-temporal-graph-attention-framework-for-robust-network-intrusion-detection-137394.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19109>)

Author: Nikhil Parkar, Sanskar Pandey, Siddhi Prajapati, Suganeshwari G, Syed Ibrahim

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Endpoint Security & XDR](<https://devfeed.tech/topics/endpoint-security-xdr.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [Graphs](<https://devfeed.tech/topics/graphs.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [detection](<https://devfeed.tech/tags/detection.md>), [framework](<https://devfeed.tech/tags/framework.md>), [graph](<https://devfeed.tech/tags/graph.md>), [intrusion-detection](<https://devfeed.tech/tags/intrusion-detection.md>), [network](<https://devfeed.tech/tags/network.md>), [network-intrusion](<https://devfeed.tech/tags/network-intrusion.md>), [robust](<https://devfeed.tech/tags/robust.md>), [temporal](<https://devfeed.tech/tags/temporal.md>)

### AI overview

ST-GAT represents network traffic as graph snapshots, using k-NN edges and multi-head graph attention for spatial aggregation, then a BiLSTM with temporal attention to model sequences. Its evaluation addresses two leakage risks by using non-overlapping temporal windows and fitting preprocessing only on the training split. Across three random seeds, inverse class-frequency weighting achieved 96.65% Macro-F1 (±0.87) and 99.14% accuracy, with Macro-F1 14.3 points above the per-flow Random Forest baseline. Dynamic balanced sampling scored 94.37% Macro-F1, 2.3 points below weighting; the authors suggest aggressive resampling may hurt on small, noisy minority classes. The abstract also names few-shot metric prototyping as a compared strategy but gives no result for it.

### Source excerpt

arXiv:2611.19109v1 Announce Type: cross Abstract: Intrusion detection systems are a central defence in modern networks. Deep learning models report strong anomaly-detection results, but evaluations on standard benchmarks such as CSE-CIC-IDS-2018 are often inflated by two methodological problems: temporal leakage from overlapping sliding windows, and contamination from preprocessing fitted before the train/test split. Per-flow classifiers also ignore the spatio-temporal dependencies across sequences of network flows. We propose a Spatio-Temporal Graph Attention Network (ST-GAT) that represents traffic as a sequence of graph snapshots. It builds k-Nearest Neighbour (k-NN) edges within each window, applies multi-head Graph Attention Network (GAT) layers for spatial aggregation, and uses a Bidirectional Long Short-Term Memory (BiLSTM) network with temporal attention for sequence modelling. To prevent leakage, we use non-overlapping temporal windows (stride equal to window length) and fit all preprocessing on the training split only. We compare three class-imbalance strategies: inverse class-frequency weighting, dynamic balanced sampling, and few-shot metric prototyping. Across three random seeds, ST-GAT with inverse class weighting reached a Macro-F1 of 96.65% (+/- 0.87) and an accuracy of 99.14%. This is 14.3 percentage points higher in Macro-F1 than a per-flow Random Forest baseline (82.34%). Balanced sampling reduced Macro-F1 by 2.3 points (to 94.37%) relative to loss weighting. This suggests that aggressive resampling of small, noisy minority classes can hurt detection in heavily imbalanced traffic.

## Foundation Pareto Flow Policy for Multi-Objective Reinforcement Learning

DevFeed: [Foundation Pareto Flow Policy for Multi-Objective Reinforcement Learning](<https://devfeed.tech/articles/foundation-pareto-flow-policy-for-multi-objective-reinforcement-learning-138943.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19825>)

Author: Zhanjiang Yang, Lijun Sun, Yueming Li, Meng Li, Yijun Yang

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [flow](<https://devfeed.tech/tags/flow.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [fp2](<https://devfeed.tech/tags/fp2.md>), [learning](<https://devfeed.tech/tags/learning.md>), [modal](<https://devfeed.tech/tags/modal.md>), [multiple](<https://devfeed.tech/tags/multiple.md>), [objectives](<https://devfeed.tech/tags/objectives.md>), [offline](<https://devfeed.tech/tags/offline.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>), [policy](<https://devfeed.tech/tags/policy.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>)

### AI overview

FP2 targets offline multi-objective reinforcement learning for continuous robotic control, where policies must balance potentially conflicting rewards and generalize trade-offs to unseen tasks. It uses a multimodal diffusion transformer with a likelihood-free flow-matching loss to learn diverse trade-off trajectories. Its three-stage pipeline pre-trains on a large MORL dataset, mid-trains on a curated high-quality dataset, then post-trains on a small amount of trajectory data from new tasks and trade-offs. The abstract reports better performance than prior methods on most D4MORL tasks, including zero-shot interpolation and extrapolation across complex Pareto manifolds and few-shot adaptation to new tasks.

### Source excerpt

arXiv:2611.19825v1 Announce Type: new Abstract: The overarching goal of reinforcement learning (RL) is to efficiently optimize the agent's policy for various environments, tasks, dynamics, and even user preferences, which often consist of multiple and potentially conflicting rewards/objectives. This necessitates the learning of Pareto-optimal policies that can navigate different trade-offs among objectives. However, existing multi-objective RL (MORL) methods, whether trained online or offline, are typically limited to task- or objective-specific policy learning and do not address generalization of learned trade-offs to unseen tasks/objectives. In this work, we propose the Foundation Pareto Flow Policy (FP2) for offline MORL in continuous robotic control, inspired by the pre-mid-post-training paradigm underlying recent foundation models. Specifically, we instantiate FP2 as a multi-modal diffusion transformer equipped with a likelihood-free flow matching loss, enabling unified, scalable, and instructable learning of diverse trade-off trajectories. We further develop a three-stage training pipeline to gradually optimize the policy: (1) pre-training on a large-scale MORL dataset, (2) mid-training on a carefully curated high-quality dataset, and (3) post-training on few collected trajectory data from new tasks and trade-offs. Empirical results demonstrate that FP2 outperforms previous state-of-the-art methods on most tasks of the D4MORL benchmark, achieving superior zero-shot interpolation and extrapolation on complex Pareto manifolds as well as efficient few-shot adaptation of the learned trade-off prior to new continuous-control tasks. Code is available at https://github.com/FibonacciRabbit/FP2.

[Next page](<https://devfeed.tech/topics/few-shot.md?cursor=WyIyMDI2LTExLTE3VDA1OjAwOjAwKzAwOjAwIiwgIjgzZjlmNDM5LTQzMjItNDhmNC04OWNiLWVmMWI0OGQwMWM3NCJd>)