# enhancing AI model accuracy

Improving the accuracy and scalability of AI models through big-data techniques and cloud computing.

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

## SimVLA: Attributing Gains in VLA Models Through Controlled Ablation

DevFeed: [SimVLA: Attributing Gains in VLA Models Through Controlled Ablation](<https://devfeed.tech/articles/simvla-attributing-gains-in-vla-models-through-controlled-ablation-144878.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2602.18224>)

Author: Yuankai Luo, Woping Chen, Tong Liang, Zhenguo Li

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

Content type: article

Language: en

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

Topics: [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [ablation](<https://devfeed.tech/tags/ablation.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [attribution](<https://devfeed.tech/tags/attribution.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [delta](<https://devfeed.tech/tags/delta.md>), [driver](<https://devfeed.tech/tags/driver.md>), [encoder](<https://devfeed.tech/tags/encoder.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [minimal](<https://devfeed.tech/tags/minimal.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [results](<https://devfeed.tech/tags/results.md>), [robot](<https://devfeed.tech/tags/robot.md>), [small](<https://devfeed.tech/tags/small.md>), [training](<https://devfeed.tech/tags/training.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>)

### AI overview

SimVLA's controlled ablations identify training dynamics as the largest measured driver of VLA performance: one-knob-at-a-time tests across LIBERO, CALVIN, WidowX, and Google Robot found swings of about 54-94%, while task configuration and architecture had smaller effects under matched settings. SimVLA pairs a standard vision-language backbone with a lightweight continuous-action head. Despite a small backbone and no additional large-scale robot-data VLA pretraining before benchmark-specific training, it achieved 98.6% average success on LIBERO, remained competitive on the other benchmarks, and transferred to held-out real-robot scenes. The work presents a controlled attribution study and minimal-baseline calibration, rather than a new technique.

### Source excerpt

arXiv:2602.18224v2 Announce Type: replace-cross Abstract: Vision-Language-Action (VLA) models often bundle architectural changes with different pretraining data, backbone scales, and optimization recipes, obscuring what actually drives progress. We introduce SimVLA, a deliberately minimal VLA, a standard vision-language backbone with a lightweight continuous-action head, as a controlled testbed for this attribution problem; our contribution is methodological and empirical (a controlled attribution study and a minimal-baseline calibration), not a new technique. Across LIBERO, CALVIN, WidowX, and Google Robot, one-knob-at-a-time ablations show that training dynamics are the largest measured driver, producing performance swings of $\Delta$$\approx$54--94\%, while task configuration and architecture have smaller effects under matched settings. Despite using a small backbone and no additional large-scale robot-data VLA pretraining before benchmark-specific training, SimVLA reaches 98.6\% average success on LIBERO, remains competitive across the other benchmarks, and transfers to held-out real-robot scenes. These results suggest that simple, carefully controlled VLA baselines remain underexplored and provide a calibration point for future architectural claims within the VLM-encoder plus lightweight-head family. Website: https://frontierrobo.github.io/SimVLA

## Genie One Accuracy: Why Curated Agents Need Governed Context

DevFeed: [Genie One Accuracy: Why Curated Agents Need Governed Context](<https://devfeed.tech/articles/genie-one-accuracy-why-curated-agents-need-governed-context-144361.md>)

Original publisher: [Read original article](<https://www.databricks.com/blog/genie-one-accuracy-why-curated-agents-need-governed-context>)

Author: Richa Sethi; Patrick Chan

Published: 2026-11-19T20:15:41Z

Content type: news

Language: en

Sources: [Databricks](<https://devfeed.tech/sources/databricks.md>)

Topics: [AI-Ready Data](<https://devfeed.tech/topics/ai-ready-data.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>), [Data Quality](<https://devfeed.tech/topics/data-quality.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>)

Tags: [1-5](<https://devfeed.tech/tags/1-5.md>), [2025](<https://devfeed.tech/tags/2025.md>), [2027](<https://devfeed.tech/tags/2027.md>), [about](<https://devfeed.tech/tags/about.md>), [access-control](<https://devfeed.tech/tags/access-control.md>), [account](<https://devfeed.tech/tags/account.md>), [added](<https://devfeed.tech/tags/added.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-instructions](<https://devfeed.tech/tags/agent-instructions.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [all](<https://devfeed.tech/tags/all.md>), [also](<https://devfeed.tech/tags/also.md>), [best-practices](<https://devfeed.tech/tags/best-practices.md>), [context](<https://devfeed.tech/tags/context.md>), [data-strategy](<https://devfeed.tech/tags/data-strategy.md>), [finance](<https://devfeed.tech/tags/finance.md>), [genie](<https://devfeed.tech/tags/genie.md>), [governed-context](<https://devfeed.tech/tags/governed-context.md>), [need](<https://devfeed.tech/tags/need.md>), [one](<https://devfeed.tech/tags/one.md>), [ratio](<https://devfeed.tech/tags/ratio.md>)

### AI overview

In the fictional insurance benchmark, Genie One reached 93-95% accuracy when curated team agents were combined with governed Unity Catalog knowledge, compared with 64% using tables alone and 83% for either addition by itself. Agents helped answer questions requiring team documents; Metric Views and Pages supplied authoritative business definitions, reducing reliance on unrelated Pages and making definition answers accurate despite variable routing. Planted draft and prompt-injection documents did not change answers. A smaller retailer benchmark showed overall accuracy rising from 54% to 82% when governed knowledge was added to agents. The excerpt also shows limits: Genie One may answer directly without inheriting agent instructions, documents available only through agent search need agents, and formulas left in agents can become stale after a governed definition changes. The results come from fictional datasets and benchmarks, including a smaller retail check.

### Source excerpt

Ask Finance and Underwriting whether the loss ratio improved from 2024 to 2025, and...

## How Well Do Generalist CT Foundation Models Differentiate Liver Tumors?

DevFeed: [How Well Do Generalist CT Foundation Models Differentiate Liver Tumors?](<https://devfeed.tech/articles/how-well-do-generalist-ct-foundation-models-differentiate-liver-tumors-142757.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24122>)

Author: Yosuke Yamagishi, Tomohiro Kikuchi, Shouhei Hanaoka, Takeharu Yoshikawa, Osamu Abe

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

Content type: article

Language: en

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

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [AI Models](<https://devfeed.tech/topics/ai-models.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [generalist](<https://devfeed.tech/tags/generalist.md>), [linear-probing](<https://devfeed.tech/tags/linear-probing.md>)

### AI overview

In this retrospective study of 517 cases from a public single-institution multiphase CT dataset, RADAR and Merlin had limited ability to differentiate four liver tumor classes zero-shot. At the portal venous phase, RADAR reached 54.4% accuracy and labeled 31% of malignant tumors as hemangioma. Training a linear classifier on frozen four-phase RADAR features raised accuracy to 79.9% and reduced those hemangioma predictions to 1%. Merlin achieved 60.9% accuracy. These findings suggest linear probing substantially improved performance in this dataset; they do not establish performance beyond the study setting.

### Source excerpt

arXiv:2611.24122v1 Announce Type: cross Abstract: Generalist CT foundation models detect many findings with high performance, but their ability to differentiate detected lesions remains largely unevaluated. This retrospective study evaluated two generalist models, RADAR and Merlin, for differentiating liver tumors using a public single-institution multiphase CT dataset (517 cases: hepatocellular carcinoma, intrahepatic cholangiocarcinoma, breast and colorectal cancer liver metastases, and hemangioma). The models were evaluated in a zero-shot setting using their existing outputs and with linear probing, in which only a linear classifier was trained on frozen image features (stratified fivefold cross-validation repeated five times). In the zero-shot setting, RADAR achieved an accuracy of 54.4% (95% CI: 49.9, 58.8) and a $\kappa$ of 0.37 for four-class differentiation at the portal venous phase and predicted 31% (129 of 421) of malignant tumors as hemangioma. With linear probing of four-phase features, its accuracy improved to 79.9% (95% CI: 76.6, 83.4) with a $\kappa$ of 0.72, and 1% (five of 421) of malignant tumors were predicted as hemangioma; Merlin achieved an accuracy of 60.9% (95% CI: 56.9, 65.0). Generalist CT foundation models showed limited zero-shot ability to differentiate liver tumors, which improved substantially with linear probing alone.

## Beyond the Sink Token: Sink-Guided Key Editing in Vision-Language Models

DevFeed: [Beyond the Sink Token: Sink-Guided Key Editing in Vision-Language Models](<https://devfeed.tech/articles/beyond-the-sink-token-sink-guided-key-editing-in-vision-language-models-142667.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23516>)

Author: Hui Zhu, Saleh Zare Zade, Rafi Ibn Sultan, Xiangyu Zhou, Dongxiao Zhu

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

Content type: article

Language: en

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

Topics: [vlm](<https://devfeed.tech/topics/vlm.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [model architecture](<https://devfeed.tech/topics/model-architecture.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [attention-steering](<https://devfeed.tech/tags/attention-steering.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [boosting](<https://devfeed.tech/tags/boosting.md>), [components](<https://devfeed.tech/tags/components.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [gemma-3](<https://devfeed.tech/tags/gemma-3.md>), [generation](<https://devfeed.tech/tags/generation.md>), [image](<https://devfeed.tech/tags/image.md>), [keys](<https://devfeed.tech/tags/keys.md>), [layers](<https://devfeed.tech/tags/layers.md>), [llava](<https://devfeed.tech/tags/llava.md>), [magnitude](<https://devfeed.tech/tags/magnitude.md>), [operations](<https://devfeed.tech/tags/operations.md>), [points](<https://devfeed.tech/tags/points.md>), [position](<https://devfeed.tech/tags/position.md>), [question](<https://devfeed.tech/tags/question.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reference](<https://devfeed.tech/tags/reference.md>), [steering](<https://devfeed.tech/tags/steering.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>)

### AI overview

Across four probed vision-language models, increasing positive sink-aligned key components raised attention to image and question tokens on average, while increasing negative components lowered it; the effects persisted in later layers. Using this relationship, the authors propose SHRINK, a training-free method that partially reduces sink-aligned components across visible keys during generation. On nine benchmarks with sampled reasoning, it raised mean accuracy by 5.1 percentage points for LLaVA-OneVision-7B and 2.7 points for Qwen2.5-VL-3B, with smaller gains for Qwen2.5-VL-7B and Gemma-3-4B. Rescaling these key components also replaced attention operations in GIFT and ACG with comparable accuracy on hallucination benchmarks.

### Source excerpt

arXiv:2611.23516v1 Announce Type: cross Abstract: In vision-language models (VLMs), a few sink tokens attract a disproportionate share of attention. While previous studies have examined why these tokens attract attention, we investigate how the sink-key direction affects attention to image and question tokens. Across four probed VLMs, boosting the magnitude of positive sink-aligned components increases attention on average, whereas boosting negative components decreases it; these responses persist through later layers. This identifies the native sink key as a reference for controlling attention beyond its own position. Motivated by this finding, we propose SHRINK, a training-free method that partially reduces sink-aligned components across visible keys during generation. On nine benchmarks under sampled reasoning, it improves mean accuracy by 5.1 and 2.7 percentage points on LLaVA-OneVision-7B and Qwen2.5-VL-3B, with smaller gains on the other two main models, Qwen2.5-VL-7B and Gemma-3-4B. Beyond SHRINK, rescaling sink-aligned key components can replace the attention operations of two existing attention-steering methods, GIFT and ACG, with comparable accuracy on hallucination benchmarks.

## Cross-Vehicle Transferability of Lateral Velocity Estimation on Production Vehicles through Input Selection and Data Augmentation

DevFeed: [Cross-Vehicle Transferability of Lateral Velocity Estimation on Production Vehicles through Input Selection and Data Augmentation](<https://devfeed.tech/articles/cross-vehicle-transferability-of-lateral-velocity-estimation-on-production-vehicles-through-input-selection-and-data-augmentation-143554.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24747>)

Author: Hendrik Sch\\"afke, Simon F. G. Ehlers, Thomas Seel, Askar Vagapov, Christoph Schweers, Ahmed Trabelsi, Daniel O. M. Weber

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

Content type: article

Language: en

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

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [data augmentation](<https://devfeed.tech/topics/data-augmentation.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [accelerometer](<https://devfeed.tech/tags/accelerometer.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [based](<https://devfeed.tech/tags/based.md>), [bias](<https://devfeed.tech/tags/bias.md>), [but](<https://devfeed.tech/tags/but.md>), [can](<https://devfeed.tech/tags/can.md>), [car](<https://devfeed.tech/tags/car.md>), [combined](<https://devfeed.tech/tags/combined.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-sy](<https://devfeed.tech/tags/cs-sy.md>), [data](<https://devfeed.tech/tags/data.md>), [data-augmentation](<https://devfeed.tech/tags/data-augmentation.md>), [eess-sy](<https://devfeed.tech/tags/eess-sy.md>), [input](<https://devfeed.tech/tags/input.md>), [lateral](<https://devfeed.tech/tags/lateral.md>), [production](<https://devfeed.tech/tags/production.md>), [selection](<https://devfeed.tech/tags/selection.md>), [vehicle](<https://devfeed.tech/tags/vehicle.md>), [vehicles](<https://devfeed.tech/tags/vehicles.md>), [velocity](<https://devfeed.tech/tags/velocity.md>)

### AI overview

A neural network trained on a compact car transfers poorly to an SUV and van: its lateral-velocity estimation error on the unseen vehicles is an order of magnitude higher than on the training vehicle. Comparing inputs drawn from ten production CAN signals identified a five-signal set with the most robust transfer. Of three physically motivated data augmentations, exploiting left-right vehicle symmetry and emulating accelerometer bias improved generalization; together they reduced error on validation maneuvers by 13% for the SUV and 15% for the van. On the SUV, the augmented network matched a model-based estimator tuned on the same training data. The findings are based on real-world drives of three production vehicles and zero-shot evaluation against a GNSS-aided inertial reference.

### Source excerpt

arXiv:2611.24747v1 Announce Type: cross Abstract: Lateral velocity is safety-critical for vehicle-dynamics control but cannot be measured with production sensors. Model-based estimators such as Kalman filters are established, whereas data-driven estimators are more accurate within the training regime but suspected to lose accuracy beyond it. Their transfer to a structurally different vehicle has rarely been tested. We record real-world test drives of three series-production vehicles, train a neural-network estimator on production CAN signals of a compact car, and evaluate it zero-shot on an SUV and a van against a GNSS-aided inertial reference. On the unseen vehicles, the network's error exceeds that on the training vehicle by an order of magnitude, a gap not quantified on production vehicles before. A systematic comparison of input sets from ten production CAN signals identifies a five-signal set that transfers most robustly. To narrow this gap, we introduce three physically motivated augmentations of training data. Two improve generalization, namely exploiting the vehicle's left-right symmetry and emulating accelerometer bias. Combined, they reduce the error on the validation maneuvers of the SUV and van by 13% and 15%. Contrary to common expectation, on the SUV the data-driven estimator trained with both augmentations matches a model-based estimator tuned on the same training data.

## Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy

DevFeed: [Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy](<https://devfeed.tech/articles/same-simulation-different-evidence-where-retrieved-tool-output-costs-llm-agents-accuracy-139843.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.26836>)

Author: Ke Zhang, Sahchit Chundur, Jerry Peprah Owusu, Mohammad Javad Abdolhosseini Qomi, Maziar Raissi

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

Content type: article

Language: en

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

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [context-management](<https://devfeed.tech/tags/context-management.md>), [context-window](<https://devfeed.tech/tags/context-window.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [tool](<https://devfeed.tech/tags/tool.md>)

### AI overview

In a PHREEQC task set of 158 execution-verified questions, retrieving simulator reports through a section index reduced accuracy for GPT-5.1 and GPT-5.2 by 11.3 and 7.9 points, while Claude Opus 4.6 and Sonnet 4.6 showed no detectable cost. Replaying 11,007 simulator calls indicated that the GPT models computed the keyed values about as often as before, but often answered from higher-precision output files instead of the report and rounded to a neighboring option ("evidence substitution"). Holding simulations fixed, retrieval still cost GPT-5.2 9.4 points; for GPT-5.4, more correct simulations masked an evidence-use loss of about ten points. Naming the authoritative output improved accuracy by 6.3-7.9 points for the three GPT models; inlining outputs that fit worked as well for GPT-5.1 and GPT-5.2, while prompting for correct rounding did not. The findings concern this scientific tool and question set.

### Source excerpt

arXiv:2610.26836v2 Announce Type: replace Abstract: LLM agents that compute the right answer can still answer wrongly. Agent harnesses increasingly keep long tool outputs out of the context window, saving them to files that the agent searches or reads on demand. We ask where this costs accuracy with a real scientific tool: PHREEQC, a widely used geochemistry simulator from the USGS, whose reports reach 846.8k characters and whose determinism lets every printed value be traced. Agents use it to answer 158 execution-verified multiple-choice items. When agents read the simulator's report through a section index rather than in context, GPT-5.1 and GPT-5.2 lose 11.3 and 7.9 points of accuracy, while Claude Opus 4.6 and Claude Sonnet 4.6 show no detectable cost. Re-running all 11,007 logged simulator calls shows that both GPT models compute the keyed value about as often as before; what changes is the evidence they answer from. Under retrieval they often skip the report, answer from their own higher-precision output files, and misround to a neighboring option, a failure we call evidence substitution. With the simulation held fixed, retrieval still costs GPT-5.2 9.4 points. For GPT-5.4, more frequent correct simulations mask an evidence-use loss of about ten points, leaving a net change indistinguishable from zero. Naming the authoritative output raises accuracy by 6.3-7.9 points over plain retrieval for the three GPT models; for GPT-5.1 and GPT-5.2, inlining outputs that fit works as well, while asking for correct rounding does not. Context-management choices in agent harnesses should be judged by where accuracy is lost, not only by final accuracy.

## The Frontier Must Grow - Self-proposed, interpreter-labelled training roughly doubles a 4B model's accuracy on a fresh exam as its verified frontier grows

DevFeed: [The Frontier Must Grow - Self-proposed, interpreter-labelled training roughly doubles a 4B model's accuracy on a fresh exam as its verified frontier grows](<https://devfeed.tech/articles/the-frontier-must-grow-self-proposed-interpreter-labelled-training-roughly-doubles-a-4b-model-s-accuracy-on-a-fresh-exam-as-its-verified-frontier-grows-137530.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20049>)

Author: Martin Schroder

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

Content type: article

Language: en

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

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [exam](<https://devfeed.tech/tags/exam.md>), [frontier](<https://devfeed.tech/tags/frontier.md>), [interpreter](<https://devfeed.tech/tags/interpreter.md>), [model](<https://devfeed.tech/tags/model.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Qwen3-4B proposed Lua 5.1 programs, and an interpreter labelled those the model got wrong at least once across four samples. The researchers expanded this verified problem pool over three rounds (300, 900, then 2,100 programs), training a fresh adapter from the base model on each round. Accuracy on a later 2,340-item exam rose from 8.3% to 16.7%, but the registered follow-up criteria were not met: round 1 showed no gain, while rounds 2 and 3 added 5.0 and 2.9 percentage points on average. A single fit on an equally sized pool of base-proposed problems reached 14.6%, leaving the rounds' 2.1-point advantage inconclusive. Three earlier registered series met no per-lineage criterion, and the loop showed no gain on repository tasks or for a 27B agent in the tested configurations.

### Source excerpt

arXiv:2611.20049v1 Announce Type: new Abstract: A model can learn from problems it poses itself when an executable oracle supplies the answers. Qwen3-4B proposed Lua 5.1 programs; those it predicted wrong at least once in four samples joined a pool labelled by the interpreter, each round adding twice as many programs as the last (300, 900, 2100 in total), and each round trained a fresh adapter from the base on the pool. On the original exam the procedure met its registered criterion, an item-level test of each adapter against its predecessor. On a 2340-item exam drawn afterwards, accuracy rose from 8.3% to 16.7%; that follow-up's registered criteria were not met because round 1 gained nothing, while rounds 2 and 3 added 5.0 and 2.9 points on average, both intervals above zero. One fit on the same amount of base-proposed frontier data reached 14.6%, so the rounds' 2.1-point advantage is inconclusive. Three earlier registered series met no per-lineage criterion, and the same loop showed no gain on repository tasks or for a 27B agent in those configurations.

## On accuracy of density estimation in generative models based on stochastic optimal control

DevFeed: [On accuracy of density estimation in generative models based on stochastic optimal control](<https://devfeed.tech/articles/on-accuracy-of-density-estimation-in-generative-models-based-on-stochastic-optimal-control-138716.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2506.03043>)

Author: Nikita Puchkin, Denis Suchkov, Alexey Naumov, Denis Belomestny

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

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>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [approximation](<https://devfeed.tech/tags/approximation.md>), [bounded](<https://devfeed.tech/tags/bounded.md>), [convergence](<https://devfeed.tech/tags/convergence.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [kl-divergence](<https://devfeed.tech/tags/kl-divergence.md>), [math-st](<https://devfeed.tech/tags/math-st.md>), [networks](<https://devfeed.tech/tags/networks.md>), [optimal-control](<https://devfeed.tech/tags/optimal-control.md>), [oracle](<https://devfeed.tech/tags/oracle.md>), [process](<https://devfeed.tech/tags/process.md>), [relu](<https://devfeed.tech/tags/relu.md>), [risk](<https://devfeed.tech/tags/risk.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [stat-th](<https://devfeed.tech/tags/stat-th.md>)

### AI overview

The study analyzes density estimation with Schrödinger bridge models using an Ornstein-Uhlenbeck reference process. With a known compactly supported initial distribution and independent samples from an unknown target, it estimates a right Schrödinger log-potential by minimizing empirical KL risk over a prescribed function class. For bounded sub-Gaussian target densities, including ones with unbounded support, it proves a non-asymptotic high-probability oracle inequality for KL divergence to the estimated terminal marginal. With additional two-sided Gaussian tail bounds and smoothness of the target log-density, sparse ReLU networks achieve approximation guarantees and exact empirical risk minimization attains a KL rate of O-tilde(n^{-2s/(2s+d)}), matching the classical nonparametric exponent up to logarithmic factors.

### Source excerpt

arXiv:2506.03043v2 Announce Type: replace Abstract: We study the statistical accuracy of Schr\"odinger bridge models based on an Ornstein-Uhlenbeck reference process. Given a known, compactly supported initial distribution and independent samples from an unknown target distribution, we estimate a right Schr\"odinger log-potential by minimizing empirical Kullback-Leibler (KL) risk over a prescribed function class. For bounded sub-Gaussian target densities, potentially with unbounded support, we establish a non-asymptotic high-probability oracle inequality for the KL-divergence between the target and estimated terminal marginals. Under additional two-sided Gaussian bounds and smoothness assumptions on the target log-density, we derive approximation guarantees for sparse ReLU networks and establish a KL convergence rate of $\widetilde{O}(n^{-2s / (2s + d)})$ for exact empirical risk minimization. This matches, up to logarithmic factors, the classical nonparametric KL-risk convergence exponent, while allowing target densities on $\mathbb R^d$ whose tails are controlled by Gaussian envelopes.

## AdaMenu: Query-Adaptive Menus for Decision Models over Large Label Spaces

DevFeed: [AdaMenu: Query-Adaptive Menus for Decision Models over Large Label Spaces](<https://devfeed.tech/articles/adamenu-query-adaptive-menus-for-decision-models-over-large-label-spaces-138816.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19005>)

Author: Jiangze Yan, Yi Shen, Jieyun Huang, Wenjing Zhang, Zhaoxiang Liu, Ning Wang, Kai Wang, Shiguo Lian

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

Content type: article

Language: en

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

Topics: [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>), [Vectors & LLMs](<https://devfeed.tech/topics/vectors-llms.md>), [decision-making](<https://devfeed.tech/topics/decision-making.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [beam-search](<https://devfeed.tech/tags/beam-search.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [error-propagation](<https://devfeed.tech/tags/error-propagation.md>), [limits](<https://devfeed.tech/tags/limits.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [size](<https://devfeed.tech/tags/size.md>), [token-usage](<https://devfeed.tech/tags/token-usage.md>)

### AI overview

AdaMenu handles large classification label sets by retrieving labeled examples to build a query-adaptive candidate menu, then making one decision-model request. Its default setting also adjusts menu size using aggregated retrieval scores. Across four label sets with up to 1,016 labels, it beat TypeSafe's beam-search baseline on all 16 model and label-set pairs where both ran, by 16.6-34.3 percentage points in accuracy. Against the full menu on the three smaller sets, accuracy ranged from 0.5 points lower to 7.2 points higher, while input token use fell by factors of 3.2-22.9. AdaMenu also ran in settings where option limits ruled out the full menu and the hierarchical baseline was unavailable.

### Source excerpt

arXiv:2611.19005v1 Announce Type: new Abstract: Decision models such as Jev return probabilities over caller-defined options and are increasingly used for classification. Large label spaces challenge both accuracy and scalability: larger menus can impair decisions, increase input costs, and exceed option limits. Hierarchical classification reduces menu size but introduces serial calls and error propagation. We propose AdaMenu, which builds a compact candidate menu from retrieved labeled examples and makes a single decision-model request. Its candidate labels adapt to the query, and its default setting also adjusts menu size using aggregated retrieval scores. We evaluate AdaMenu with open and closed decision models on four label sets containing up to 1,016 labels. It outperforms TypeSafe's beam-search baseline on all 16 model and label-set pairs where both methods run, improving accuracy by 16.6-34.3 percentage points. Relative to the full menu on the three smaller label sets, accuracy differences range from -0.5 to +7.2 percentage points, while input token usage falls by factors of 3.2-22.9. AdaMenu also runs when option limits exclude the full menu, including settings where the hierarchical baseline is unavailable.

## iEEG Can Help EEG: Improving Clinical Scalp EEG Analysis through Intracranial Priors

DevFeed: [iEEG Can Help EEG: Improving Clinical Scalp EEG Analysis through Intracranial Priors](<https://devfeed.tech/articles/ieeg-can-help-eeg-improving-clinical-scalp-eeg-analysis-through-intracranial-priors-139050.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20435>)

Author: Xianglong Bao, Yi Ding, Zengxiang Li, Nan Liu, Kewen Wang, Zhe Wang, Ming Jin

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

Content type: article

Language: en

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

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Medical imaging](<https://devfeed.tech/topics/medical-imaging.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [brain](<https://devfeed.tech/tags/brain.md>), [code](<https://devfeed.tech/tags/code.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [eeg](<https://devfeed.tech/tags/eeg.md>), [eess-sp](<https://devfeed.tech/tags/eess-sp.md>), [framework](<https://devfeed.tech/tags/framework.md>), [hospitals](<https://devfeed.tech/tags/hospitals.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [model](<https://devfeed.tech/tags/model.md>), [monitoring](<https://devfeed.tech/tags/monitoring.md>), [neural](<https://devfeed.tech/tags/neural.md>), [q-bio-nc](<https://devfeed.tech/tags/q-bio-nc.md>), [recordings](<https://devfeed.tech/tags/recordings.md>), [signal](<https://devfeed.tech/tags/signal.md>), [synchronization](<https://devfeed.tech/tags/synchronization.md>)

### AI overview

ReBrain uses an intracranial EEG (iEEG) model as a transferable prior to improve scalp EEG analysis, which is challenged by non-neural interference and volume conduction. It trains a signal-to-signal model on iEEG alone, freezes it, then re-expresses scalp signals through that model before a downstream host model. The approach requires neither paired iEEG/scalp recordings nor downstream fine-tuning, addressing the lack of shared channel space and clean mapping targets. The abstract reports consistent improvements across recordings from multiple hospitals and architecture families, attributing the gains to neural structure in iEEG. iEEG is higher fidelity but invasive and scarce; the supplied abstract does not quantify the improvements.

### Source excerpt

arXiv:2611.20435v1 Announce Type: new Abstract: Scalp electroencephalography (EEG) is a cornerstone of clinical brain monitoring, making the quality of its signal fundamental to downstream tasks and models. However, that signal is heavily contaminated by non-neural interference and blurred by volume conduction. Existing methods operate almost exclusively within the scalp modality, so they inherit the contamination they would need to overcome. In practice, intracranial EEG (iEEG) observes the neural activity with far higher fidelity, yet it is invasive and therefore scarce, and the two modalities are rarely recorded as pairs at scale, share no common channel space, and offer no clean ground truth to directly supervise a mapping. Motivated by the neuronal synchronization mechanism shared by the two modalities, we propose a reprogramming approach that fits a signal-to-signal model on iEEG alone and freezes it as a cross-modal prior. We introduce ReBrain, a framework to transfer an iEEG-learned prior to scalp EEG, which re-expresses any scalp signal through that prior and can be inserted upstream of any downstream model with no pairing or fine-tuning. ReBrain consistently improves host models across recordings from multiple hospitals and across architecture families, and our analysis traces this gain to genuine neural structure carried by iEEG rather than to the architecture itself. This work provides evidence that a scarce clean modality can serve as a transferable prior for an abundant noisy one, offering a general recipe for cross-modal transfer among biosignals. Our code is available at https://github.com/Xianglong-Bao/ReBrain.

## Generalization Testing of Grid Foundation Models for Power Flow and Optimal Power Flow

DevFeed: [Generalization Testing of Grid Foundation Models for Power Flow and Optimal Power Flow](<https://devfeed.tech/articles/generalization-testing-of-grid-foundation-models-for-power-flow-and-optimal-power-flow-138842.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19194>)

Author: Ayushi Jolotia, Parikshit Pareek

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

Content type: article

Language: en

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

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-sy](<https://devfeed.tech/tags/cs-sy.md>), [eess-sy](<https://devfeed.tech/tags/eess-sy.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [flow](<https://devfeed.tech/tags/flow.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [framework](<https://devfeed.tech/tags/framework.md>), [grid](<https://devfeed.tech/tags/grid.md>), [power](<https://devfeed.tech/tags/power.md>), [testing](<https://devfeed.tech/tags/testing.md>)

### AI overview

The study tests GENCO, LUMINA, and GridSFM against a DC baseline across three changes: altered grid admittance matrices, load levels, and unseen grids. Adding a line degrades all three foundation models more than removing one; with longer lines, LUMINA and fine-tuned GENCO eventually perform worse than DC. All three models degrade beyond their training load range, where the loading-voltage curve steepens. On unseen grids, LUMINA degrades sharply even within its training range, while GridSFM fails on the grid farthest from its training grid. The authors conclude that outside training conditions these models act more like meta-models, requiring test-time fine-tuning data and compute. The supplied text is an abstract, so it does not provide details of the experiments or the models' reported limitations.

### Source excerpt

arXiv:2611.19194v1 Announce Type: cross Abstract: There is growing interest in foundation models (FMs) for power grids, particularly for power flow and optimal power flow. Their usefulness depends on generalization accuracy, which existing works test on their own benchmarks. We propose a three-axis framework to test their generalization limits: changed admittance matrices of training grids, load levels, and unseen grids. We test GENCO, LUMINA and GridSFM against the DC baseline, alongside their reported limitations, and discuss theoretical reasons for their failures. Experiments reveal that adding a line degrades all FMs more than removing one, and longer lines eventually make LUMINA and fine-tuned GENCO worse than DC. All three degrade beyond their training load range, where the loading-voltage curve steepens. On unseen grids, LUMINA degrades sharply even inside its training range, and GridSFM fails on the grid farthest from it. Thus, beyond training conditions, FMs behave more like meta-models requiring test-time fine-tuning data and compute.

## Auditing World-Model Explanations in Retail Forecasting: Faithfulness, Accuracy and Price Response

DevFeed: [Auditing World-Model Explanations in Retail Forecasting: Faithfulness, Accuracy and Price Response](<https://devfeed.tech/articles/auditing-world-model-explanations-in-retail-forecasting-faithfulness-accuracy-and-price-response-137614.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20811>)

Author: Kathik Iyer

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>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Time Series](<https://devfeed.tech/topics/time-series.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [auditing](<https://devfeed.tech/tags/auditing.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [forecasting](<https://devfeed.tech/tags/forecasting.md>), [retail](<https://devfeed.tech/tags/retail.md>), [world-model](<https://devfeed.tech/tags/world-model.md>)

### AI overview

Passing explanation checks did not ensure accurate retail forecasts or price-response claims. In generated store data, a world model passed checks for construction, stability, price direction and agreement with sales and LightGBM, yet overstated promotion lift by 1.33-1.40 times when merchants planned into high-demand weeks (1.23-1.42 times across eight variants). It selected the right weeks, but inventory ordered to its lift differed from the true lift by 0.44-0.48. Its stated response to lasting regular-price changes was 4.7-6.5 times the known effect; only a measured-cause approach recovered it. On Dominick's, a noisy benchmark from 38 episodes estimated a pooled eight-week response of -0.33 (95% interval -0.68 to 0.01). Tuned world models and LightGBM stated about twice that measured response, and deeper price cuts lowered forecasts in only 29%-83% of item-week steps. Across 703 settings, check pass counts were not associated with forecast accuracy (Spearman 0.02, 95% interval -0.02 to 0.05); explanation correctness could be assessed only in generated data with known effects.

### Source excerpt

arXiv:2611.20811v1 Announce Type: cross Abstract: Forecasting models that read the price and promotion plan explain each forecast as a baseline, what would sell with nothing planned, plus the lift the plan adds. We test such explanations with eight checks, six with a pass rule fixed before any result, and find that passing them does not make an explanation right. In generated store data with known effects, a world model passes the checks of construction, stability, price direction and agreement with sales and with LightGBM, yet its lifts are 1.33 to 1.40 times the truth when merchants plan into high-demand weeks, and 1.23 to 1.42 times in eight variants of that world. It still picks the right weeks to promote, but stock ordered to its lift is off by 0.44 to 0.48 of the true lift. Its stated response to a lasting regular-price change is 4.7 to 6.5 times the true -0.5, and only a measured cause finds this. On Dominick's, 38 episodes in which a store's regular prices sat apart from its zone peers give a noisy benchmark: a pooled 8-week response of -0.33 (95% interval -0.68 to 0.01). For the final tuned world models of our companion paper on nine data sets, baseline and lift add up under each model's own plan, but a deeper price cut lowers the forecast in 29% to 83% of item-week steps on Dominick's (23% to 80% within prices seen in training), and the world model and LightGBM both state about twice the measured response. Across 703 configurations from the settings searches, checks passed were not related to forecast accuracy (Spearman 0.02, 95% interval -0.02 to 0.05); with no known answer for those versions, this concerns accuracy, while whether an explanation is right is measured only in the generated data.

## Can decision models extract clinical data from cancer pathology reports? An evaluation of accuracy, calibration, and failure modes

DevFeed: [Can decision models extract clinical data from cancer pathology reports? An evaluation of accuracy, calibration, and failure modes](<https://devfeed.tech/articles/can-decision-models-extract-clinical-data-from-cancer-pathology-reports-an-evaluation-of-accuracy-calibration-and-failure-modes-137540.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20129>)

Author: Indranil Mallick, Paromita Roy

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>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>), [Data analysis](<https://devfeed.tech/topics/data-analysis.md>), [Open Source Models & Datasets](<https://devfeed.tech/topics/open-source-models-datasets.md>)

Tags: [cancer](<https://devfeed.tech/tags/cancer.md>), [ci](<https://devfeed.tech/tags/ci.md>), [cloud](<https://devfeed.tech/tags/cloud.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>), [data](<https://devfeed.tech/tags/data.md>), [database](<https://devfeed.tech/tags/database.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [failure](<https://devfeed.tech/tags/failure.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [glm](<https://devfeed.tech/tags/glm.md>), [glm-5-3-flash](<https://devfeed.tech/tags/glm-5-3-flash.md>), [local](<https://devfeed.tech/tags/local.md>), [margins](<https://devfeed.tech/tags/margins.md>), [measurements](<https://devfeed.tech/tags/measurements.md>), [nimble](<https://devfeed.tech/tags/nimble.md>), [open-weights](<https://devfeed.tech/tags/open-weights.md>), [performance](<https://devfeed.tech/tags/performance.md>), [probability](<https://devfeed.tech/tags/probability.md>), [questions](<https://devfeed.tech/tags/questions.md>), [reports](<https://devfeed.tech/tags/reports.md>)

### AI overview

In one centre's evaluation of 1,113 oral cavity cancer pathology reports, decision models extracted some explicitly stated findings quickly on local hardware, but performance varied by model and field. Tev1 4B averaged 0.917 accuracy, comparable to local Qwen3.5 9B (0.918; p = 0.26), while taking 15 seconds per report versus 61. Nimble 9B scored 0.862, and cloud-based GLM-5.3-flash scored 0.986. Decision models matched GLM on simple yes/no categories but struggled with measurements converted to categories, unreported answers, and complex decisions such as involved margins. Asking Nimble two questions per request improved its accuracy over asking all 25 at once. Probabilities were calibrated across fields but did not reveal systematic errors in individual fields. In a simulation, escalating Tev1 answers below 0.94 to GLM raised accuracy to 0.982, with 38% escalated. The authors recommend validating the intended configuration, limiting use to reliable fields, and using other methods for numeric values.

### Source excerpt

arXiv:2611.20129v1 Announce Type: cross Abstract: Decision models are a new class of language model that, instead of generating text, choose one option per question from a fixed list and return a probability for every option. They have not yet been evaluated on clinical text. We evaluated two open-weights decision models, Nimble 9B and Tev1 4B, for extracting 25 fields from 1,113 oral cavity cancer pathology reports from a single centre, against a prospectively maintained clinical database. We compared them with two generative models: Qwen3.5 9B, run locally, and GLM-5.3-flash, run through a cloud API. Tev1 reached a mean field accuracy of 0.917 (95% CI 0.913 to 0.920), the same as Qwen3.5 9B (0.918; p = 0.26), and was four times faster (15 s against 61 s per report). Nimble reached 0.862 and GLM-5.3-flash 0.986. Both decision models matched GLM-5.3-flash on simple dichotomous categories. They were weak on measurements binned into categories, on fields whose correct answer was not reported, and on some complex decisions (e.g. involved margins). We found that the main cause of the inferior performance of Nimble's came from asking 25 questions in one request: with two questions per request, its accuracy improved. Pooled over fields, the returned probabilities were well calibrated (expected calibration error 0.022 and 0.026), but they did not flag systematic errors within single fields. A simulated strategy of sending answers with a probability below 0.94 to GLM-5.3-flash raised Tev1's accuracy to 0.982, with 38% of answers escalated. Decision models can extract explicitly stated findings from pathology reports quickly on local hardware and return useful probabilities. They should be validated in the configuration in which they will be used, restricted to the fields in which they are reliable, and combined with other methods for numeric values.

## RSIMed: Scaling Medical Foundation Models beyond Limited Clinical Data via Retrieval-Grounded Self-Evolution

DevFeed: [RSIMed: Scaling Medical Foundation Models beyond Limited Clinical Data via Retrieval-Grounded Self-Evolution](<https://devfeed.tech/articles/rsimed-scaling-medical-foundation-models-beyond-limited-clinical-data-via-retrieval-grounded-self-evolution-137318.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18646>)

Author: Wei Dai, Qianchu Liu, Guanghui Qin, Xing Han, Tristan Naumann, Sheng Zhang, Paul Pu Liang

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>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>), [Vectors & LLMs](<https://devfeed.tech/topics/vectors-llms.md>), [Advanced AI workflows](<https://devfeed.tech/topics/advanced-ai-workflows.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diagnosis](<https://devfeed.tech/tags/diagnosis.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [grounded](<https://devfeed.tech/tags/grounded.md>), [less](<https://devfeed.tech/tags/less.md>), [medical](<https://devfeed.tech/tags/medical.md>), [models](<https://devfeed.tech/tags/models.md>), [recursive-self-improvement](<https://devfeed.tech/tags/recursive-self-improvement.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [rl](<https://devfeed.tech/tags/rl.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [self-improvement](<https://devfeed.tech/tags/self-improvement.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [train](<https://devfeed.tech/tags/train.md>)

### AI overview

RSIMed trains medical diagnostic models through retrieval-grounded, evolving tasks: a frozen copy of the solver creates questions from retrieved evidence and targets topics and difficulty the solver still struggles with. On rare-disease diagnosis using MIMIC-IV admissions, RSIMed-27B reaches 0.516 accuracy and RSIMed-9B 0.488 without external models; the abstract reports accuracy scaling with training compute, while a static generated-task pool collapses. The best stated GPT-5.6 setting scores 0.399 at 38 times RSIMed-27B's inference cost, and tasks from four external teachers, including GPT-5.5, produce a less accurate diagnostician than the solver's own tasks. RSIMed-RL-27B, trained only on diagnosis, also raises its adjusted HealthBench Professional score from 0.291 to 0.303. These are the reported results in the supplied abstract.

### Source excerpt

arXiv:2611.18646v1 Announce Type: cross Abstract: Rare diseases are hard to diagnose, and the labeled clinical records needed to train models for them are few, private, and slow to extend, since a specialist must read every case. A model that writes its own training tasks could learn beyond these records, but no program can check a model-written diagnosis, and a wrong answer key corrupts training. To tackle this dilemma between data scarcity and difficulty of data annotations, we introduce RSIMed, Recursive Self-Improvement in Medicine, one of the first self-evolving pipelines in which a task proposer co-evolves in context with a diagnostic solver on real clinical records. Seeded by training admissions, a frozen copy of the solver writes each clinical question from retrieved medical evidence, so its answer rests on a source, and steers topics and difficulty toward what the solver still gets wrong. On rare-disease diagnosis from MIMIC-IV admissions, the accuracy of RSIMed-27B grows linearly with training compute, while a static pool of generated tasks collapses. Without any external model, RSIMed-27B and RSIMed-9B reach 0.516 and 0.488 diagnostic accuracy, above every off-the-shelf GPT-5.6 setting, whose best reaches 0.399 at 38 times the inference cost of RSIMed-27B. Tasks written by four external teachers, including GPT-5.5, train a less accurate diagnostician than the solver's own. Trained only on diagnosis, RSIMed-RL-27B also improves on the physician-written tasks of HealthBench Professional, from 0.291 to 0.303 in adjusted score.

## Nested Ensemble Learning with Topological Data Analysis for Graph Classification and Regression

DevFeed: [Nested Ensemble Learning with Topological Data Analysis for Graph Classification and Regression](<https://devfeed.tech/articles/nested-ensemble-learning-with-topological-data-analysis-for-graph-classification-and-regression-132931.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.14900>)

Author: Innocent Abaa, Umar Islambekov

Published: 2026-11-13T05:00:00Z

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>), [Graph Mining](<https://devfeed.tech/topics/graph-mining.md>), [Model Development](<https://devfeed.tech/topics/model-development.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [classification](<https://devfeed.tech/tags/classification.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [ensemble-learning](<https://devfeed.tech/tags/ensemble-learning.md>), [graph](<https://devfeed.tech/tags/graph.md>), [regression](<https://devfeed.tech/tags/regression.md>), [stat-ap](<https://devfeed.tech/tags/stat-ap.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

The framework combines graph topology from three centrality-based filters--closeness, betweenness, and degree2--through a three-level ensemble. It computes persistence diagrams for each filter, trains separate base models, combines their predictions into filter-specific ensembles, then integrates those ensembles for a final prediction. On simulated and real-world graph datasets, the authors report better classification accuracy and lower regression error than base models and standard stacking, and placement among the top three models across benchmarks. The supplied text does not include dataset details or quantitative results.

### Source excerpt

arXiv:2611.14900v1 Announce Type: cross Abstract: We propose a nested ensemble learning framework that utilizes Topological Data Analysis (TDA) to extract and integrate topological features from graph data, with the goal of improving performance on classification and regression tasks. Our approach computes persistence diagrams (PDs) using lower-star filtrations induced by three filter functions: closeness, betweenness, and degree2 centrality. To overcome the limitation of relying on a single filter, these PDs are integrated through a data-driven, three-level architecture. At Level-0, diverse base models are independently trained on the topological features extracted for each filter function. At Level-1, a meta-learner combines the predictions of these base models for each filter to form filter-specific ensembles. Finally, at Level-2, a meta-learner integrates the outputs of these filter-specific ensembles to produce the final prediction. We evaluate our method on both simulated and real-world graph datasets. Experimental results demonstrate that our framework consistently outperforms base models and standard stacking methods, achieving higher classification accuracy and lower regression error. It also surpasses existing state-of-the-art approaches, ranking among the top three models across all benchmarks.

## SPARC: Accurate Photonic Matrix Multiplication via Representative Vectors

DevFeed: [SPARC: Accurate Photonic Matrix Multiplication via Representative Vectors](<https://devfeed.tech/articles/sparc-accurate-photonic-matrix-multiplication-via-representative-vectors-132256.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16561>)

Author: Yasuhiro Fujiwara, Tatsuhiro Nakamori, Hideyuki Kawashima

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

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

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [accelerators](<https://devfeed.tech/tags/accelerators.md>), [activation](<https://devfeed.tech/tags/activation.md>), [analog](<https://devfeed.tech/tags/analog.md>), [application](<https://devfeed.tech/tags/application.md>), [approach](<https://devfeed.tech/tags/approach.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [can](<https://devfeed.tech/tags/can.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [deep](<https://devfeed.tech/tags/deep.md>), [dnn](<https://devfeed.tech/tags/dnn.md>), [domains](<https://devfeed.tech/tags/domains.md>), [effects](<https://devfeed.tech/tags/effects.md>), [errors](<https://devfeed.tech/tags/errors.md>), [existing](<https://devfeed.tech/tags/existing.md>), [experimental](<https://devfeed.tech/tags/experimental.md>), [future](<https://devfeed.tech/tags/future.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [matrix](<https://devfeed.tech/tags/matrix.md>), [vectors](<https://devfeed.tech/tags/vectors.md>)

### AI overview

SPARC aims to improve photonic DNN inference accuracy despite low-bit quantization and analog noise. It groups similar activation vectors, computes each group's representative vector inner product digitally at high precision, then sends the residual inner products to the photonic accelerator. Because those residuals have smaller norms, the approach reduces their exposure to quantization errors and photonic noise. The paper says this preserves the photonic inner-product pipeline and requires no hardware modification. Experiments on six representative DNN models reportedly improved inference accuracy across diverse architectures and application domains; the excerpt provides no quantitative results.

### Source excerpt

arXiv:2611.16561v1 Announce Type: new Abstract: Photonic neural network accelerators have emerged as a promising solution for reducing the power consumption of deep neural network (DNN) inference by executing matrix multiplication on analog photonic hardware. However, their inference accuracy is limited by low-bit quantization required for photonic processing and by analog photonic noise. This paper proposes an approach for improving inference accuracy by exploiting the high similarity among activation vectors in DNNs. Specifically, our approach constructs a representative vector for each group of similar activation vectors, computes its inner product digitally with high precision, and offloads the residual inner products to the photonic accelerator. Since the activation vectors within each group are highly similar, the residual vectors have much smaller norms, thereby reducing the effects of quantization errors and analog photonic noise. Furthermore, our approach preserves the photonic inner-product computation pipeline, allowing seamless integration into existing and future photonic accelerators without hardware modification. Experimental results on six representative DNN models demonstrate that the proposed approach can improve inference accuracy across diverse network architectures and application domains.

## CAMRI Loss: Improving the Recall of a Specific Class without Sacrificing Accuracy

DevFeed: [CAMRI Loss: Improving the Recall of a Specific Class without Sacrificing Accuracy](<https://devfeed.tech/articles/camri-loss-improving-the-recall-of-a-specific-class-without-sacrificing-accuracy-132711.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2209.10920>)

Author: Daiki Nishiyama, Kazuto Fukuchi, Youhei Akimoto, Jun Sakuma

Published: 2026-11-13T05:00:00Z

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>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [angular](<https://devfeed.tech/tags/angular.md>), [cifar-10](<https://devfeed.tech/tags/cifar-10.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [feature](<https://devfeed.tech/tags/feature.md>), [important](<https://devfeed.tech/tags/important.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [recall](<https://devfeed.tech/tags/recall.md>), [space](<https://devfeed.tech/tags/space.md>), [variance](<https://devfeed.tech/tags/variance.md>), [weight](<https://devfeed.tech/tags/weight.md>)

### AI overview

CAMRI loss targets higher recall for a designated important class while preserving overall accuracy. It adds a class-sensitive angular penalty between that class's features and corresponding weight vectors, aiming to reduce feature variance and improve separation without substantially impairing separation of other classes. The authors report improved specific-class recall without sacrificing accuracy on CIFAR-10, GTSRB, and AwA2; on GTSRB, recall for the second-worst class rose by 9% compared with cross-entropy training.

### Source excerpt

arXiv:2209.10920v2 Announce Type: replace Abstract: In real-world applications of multiclass classification models, misclassification in an important class (e.g., stop sign) can be significantly more harmful than in other classes (e.g., no parking). Thus, it is crucial to improve the recall of an important class while maintaining overall accuracy. For this problem, we found that improving the separation of important classes relative to other classes in the feature space is effective. Existing methods that give a class-sensitive penalty for cross-entropy loss do not improve the separation. Moreover, the methods designed to improve separations between all classes are unsuitable for our purpose because they do not consider the important classes. To achieve the separation, we propose a loss function that explicitly gives loss for the feature space, called class-sensitive additive angular margin (CAMRI) loss. CAMRI loss is expected to reduce the variance of an important class due to the addition of a penalty to the angle between the important class features and the corresponding weight vectors in the feature space. In addition, concentrating the penalty on only the important class hardly sacrifices separating the other classes. Experiments on CIFAR-10, GTSRB, and AwA2 showed that CAMRI loss could improve the recall of a specific class without sacrificing accuracy. In particular, compared with GTSRB's second-worst class recall when trained with cross-entropy loss, CAMRI loss improved recall by 9%.

## Classical Energies Learn to Reason 1,000x Faster than Looped Models

DevFeed: [Classical Energies Learn to Reason 1,000x Faster than Looped Models](<https://devfeed.tech/articles/classical-energies-learn-to-reason-1-000x-faster-than-looped-models-132331.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17309>)

Author: Benjamin Hoover, Hugh Blayney, James C. R. Whittington

Published: 2026-11-13T05:00:00Z

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>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [built-in](<https://devfeed.tech/tags/built-in.md>), [compute](<https://devfeed.tech/tags/compute.md>), [cram](<https://devfeed.tech/tags/cram.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [energy](<https://devfeed.tech/tags/energy.md>), [fast](<https://devfeed.tech/tags/fast.md>), [faster](<https://devfeed.tech/tags/faster.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [inference](<https://devfeed.tech/tags/inference.md>), [models](<https://devfeed.tech/tags/models.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [scalability](<https://devfeed.tech/tags/scalability.md>), [simple](<https://devfeed.tech/tags/simple.md>), [solve](<https://devfeed.tech/tags/solve.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [test](<https://devfeed.tech/tags/test.md>), [train](<https://devfeed.tech/tags/train.md>), [training](<https://devfeed.tech/tags/training.md>), [using](<https://devfeed.tech/tags/using.md>), [version](<https://devfeed.tech/tags/version.md>)

### AI overview

CRAM, a learned Hopfield-Tank energy model, solves 99.996% of Sudoku-Extreme and 100% of Maze-Unique test puzzles in results reported after 42 ms and 132 ms of training on one GPU, respectively. The authors report at least 1,000x faster training than looped models on these tasks and matching baseline looped models' peak Sudoku-Extreme accuracy with at least 282x fewer inference FLOPs. They attribute energy descent's appeal to early stopping, compute that adapts to problem difficulty, training without differentiating through a solver, and interpretable weights that learn puzzle constraints. The authors caution that Sudoku and Maze may be too easily solved to test the scalability of looped reasoning methods.

### Source excerpt

arXiv:2611.17309v1 Announce Type: new Abstract: Looped models have emerged as a prominent paradigm for reasoning, with successes on simple reasoning puzzles such as Sudoku-Extreme and Maze-Unique. However, these models can be unstable, slow to train, and use many parameters and FLOPs to solve simple puzzles. Descent on an energy not only mitigates these issues but also offers advantages: built-in early stopping, compute that adapts to problem difficulty, fast training without differentiating through a solver, and interpretable weights that learn the constraints of the puzzle. Here we demonstrate this with a learned version of Hopfield & Tank's energy from the 80s, which we term CRAM, solves 99.996% of Sudoku-Extreme and 100% of Maze-Unique test puzzles after 42 ms and 132 ms of training on one GPU, at least 1,000x faster than looped models on these tasks. CRAM also matches baseline looped models at their highest accuracy on Sudoku-Extreme using at least 282x fewer inference FLOPs. Our work suggests two things. First, that energies are an extremely promising direction for training reasoning from data. Second, a caution against using trivially solvable benchmarks like Sudoku and Maze to test the scalability of looped reasoning techniques.

## Genie with Genie Ontology Delivers 1.6x the Accuracy with Less than Half the Tokens

DevFeed: [Genie with Genie Ontology Delivers 1.6x the Accuracy with Less than Half the Tokens](<https://devfeed.tech/articles/genie-with-genie-ontology-delivers-1-6-the-accuracy-with-less-than-half-the-tokens-129972.md>)

Original publisher: [Read original article](<https://www.databricks.com/blog/genie-genie-ontology-delivers-16-accuracy-less-half-tokens>)

Author: Jeff Fletcher; Rishabh Singh

Published: 2026-11-11T15:00:00Z

Content type: article

Language: en

Sources: [Databricks](<https://devfeed.tech/sources/databricks.md>)

Topics: [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [assets](<https://devfeed.tech/tags/assets.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [business-context](<https://devfeed.tech/tags/business-context.md>), [databricks](<https://devfeed.tech/tags/databricks.md>), [enterprise-data](<https://devfeed.tech/tags/enterprise-data.md>), [genie](<https://devfeed.tech/tags/genie.md>), [genie-ontology](<https://devfeed.tech/tags/genie-ontology.md>), [platform](<https://devfeed.tech/tags/platform.md>)

### AI overview

In a benchmark run five times per task in a live Databricks workspace, Genie achieved a 91% pass rate versus 45% and 58% for two frontier agents, completed tasks 3-4 times faster, and used less than half the tokens on average. The benchmark tests questions that require business knowledge spread across notebooks, dashboards, saved SQL, documents, and other workspace assets, not just table schemas. Genie Ontology surfaces relevant context and links it to its source. In one example, a table showed a rate applied using FY25/26 assumptions, while a notebook documented the historical FY24/25 planning rate; Genie returned the documented rate of 0.52 p/kWh in all five runs, while the other agents did not. The authors caution that differences may be smaller in new deployments or when table data and schemas alone answer the question. The excerpt describes a fictional company and a Databricks-specific setup, with the other agents connected through a toolbelt that excluded Genie MCP.

### Source excerpt

We developed a new enterprise data-agent benchmark in collaboration with Turing to...

## Projected Neural Additive Models as Universal Approximators

DevFeed: [Projected Neural Additive Models as Universal Approximators](<https://devfeed.tech/articles/projected-neural-additive-models-as-universal-approximators-125141.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10186>)

Author: Nhon N. Phan, Qiang Du, John D. Clayton, WaiChing Sun

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

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

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [mlp](<https://devfeed.tech/topics/mlp.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [approximation](<https://devfeed.tech/tags/approximation.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [interpretability](<https://devfeed.tech/tags/interpretability.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [neural](<https://devfeed.tech/tags/neural.md>), [sparsity](<https://devfeed.tech/tags/sparsity.md>), [symbolic-regression](<https://devfeed.tech/tags/symbolic-regression.md>)

### AI overview

Projected neural additive models (PNAMs) add a learnable linear transformation to neural additive models (NAMs), allowing them to represent cross-variable interactions that standard NAMs cannot, including the interaction χ₁χ₂. The authors prove PNAMs are universal approximators, characterize the minimum projections needed for exact polynomial representation, and give an approximation rate of O(M^{-s/(N−1)}) for functions with s continuous derivatives, described as optimal for ridge function approximation. Structured sparsity and post-hoc symbolic regression support input ranking, parameter pruning, and compact analytical expressions. In experiments on knot theory invariants, MNIST, and phase field fracture mechanics, PNAMs reportedly match or exceed similar architectures in accuracy while identifying salient features and producing symbolic models with far fewer parameters.

### Source excerpt

arXiv:2611.10186v1 Announce Type: new Abstract: Neural additive models (NAMs) offer interpretability but lack expressivity due to their rigid additive structure, whereas multi-layer perceptrons (MLPs) achieve universal approximation at the cost of transparency. We establish the first theoretical foundation for projected neural additive models (PNAMs), which augment NAMs with a learnable linear transformation $\boldsymbol{T}$ to capture cross-variable interactions. While standard NAMs cannot approximate even the simplest interaction term $\chi_1 \chi_2$ because their cross-derivatives are identically zero, we prove that PNAMs are universal approximators and characterize the minimum number of projections required for exact polynomial representation. Furthermore, we derive an approximation rate of $\mathcal{O}(M^{-s / (N - 1)})$ for $\mathcal{C}^s$ functions (i.e., functions with $s$ continuous derivatives), where $M$ denotes the number of projections and $N$ the input dimension; this rate is known to be optimal for ridge function approximation. To recover interpretability from the projected inputs, we introduce structured sparsity regularizers and post-hoc symbolic regression techniques that enable global input ranking, parameter pruning, and the conversion of learned bases into compact analytical expressions. Experiments on knot theory invariants, MNIST, and phase field fracture mechanics demonstrate that PNAMs match or surpass similar architectures in accuracy while discovering salient features and yielding symbolic models with orders of magnitude fewer parameters.

## Definitions, Label Names, or Demonstrations? Auditing What Governs LLM Annotation

DevFeed: [Definitions, Label Names, or Demonstrations? Auditing What Governs LLM Annotation](<https://devfeed.tech/articles/definitions-label-names-or-demonstrations-auditing-what-governs-llm-annotation-124685.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09457>)

Author: Viraj Mishra, Shubham Gupta, Dhruv Gupta, Yuehong Cassandra Tai, Qunhua Li

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

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

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>)

Tags: [annotation](<https://devfeed.tech/tags/annotation.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [definition](<https://devfeed.tech/tags/definition.md>), [examples](<https://devfeed.tech/tags/examples.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [llm](<https://devfeed.tech/tags/llm.md>), [nearest-neighbor](<https://devfeed.tech/tags/nearest-neighbor.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [science](<https://devfeed.tech/tags/science.md>), [social](<https://devfeed.tech/tags/social.md>), [training](<https://devfeed.tech/tags/training.md>), [training-data](<https://devfeed.tech/tags/training-data.md>)

### AI overview

Across three political-text datasets, LLMs rarely follow category definitions when those definitions are reassigned to output labels; standard fine-tuning reinforces the original label mapping without reliably improving adherence to the new definitions. For retrieval-augmented annotation, retrieving the nearest training examples generally beats retrieving examples separately by category, even with fewer demonstrations. On two tasks with substantial gains, direct k-nearest-neighbor classification recovers 74-93% of the improvement, suggesting neighborhood labels carry much of the signal. Relabeling fixed demonstrations redirects predictions, while aligning demonstrations to reassigned definitions improves adherence. The findings show that accuracy gains alone do not establish codebook adherence, so definitions, labels, demonstrations, retrieval, and training data need joint validation.

### Source excerpt

arXiv:2611.09457v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to annotate social-science texts, but high accuracy does not establish that they apply researcher-supplied category definitions: predictions may instead reflect prior associations with category labels or labeled demonstrations. We introduce a counterfactual audit that reassigns definitions among output labels, distinguishing adherence to supplied definitions from preservation of a dataset's original mapping. Across three political-text datasets, models rarely follow reassigned definitions, and standard fine-tuning strengthens the original mapping without reliably increasing definition adherence. We then study retrieval-augmented annotation. Retrieving the nearest training examples generally outperforms retrieving examples separately within each category despite using fewer demonstrations. On the two tasks with substantial gains, direct \(k\)-nearest-neighbor classification recovers 74--93\% of the improvement, indicating that neighborhood labels provide much of the predictive signal. Relabeling fixed demonstrations redirects predictions, whereas aligning demonstrations with reassigned definitions substantially increases adherence to the new mapping. Thus, accuracy-improving interventions need not increase codebook adherence; definitions, category labels, demonstrations, retrieval, and training data should be aligned and validated jointly.

## Active Learning of Robust Models

DevFeed: [Active Learning of Robust Models](<https://devfeed.tech/articles/active-learning-of-robust-models-123417.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09520>)

Author: Mizu Nishikawa-Toomey, Andreas Kirsch, Laurent Charlin, Dhanya Sridhar

Published: 2026-11-09T05:00:00Z

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 Development](<https://devfeed.tech/topics/model-development.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [data-collection](<https://devfeed.tech/tags/data-collection.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [information](<https://devfeed.tech/tags/information.md>), [learning](<https://devfeed.tech/tags/learning.md>), [robust](<https://devfeed.tech/tags/robust.md>), [shift](<https://devfeed.tech/tags/shift.md>), [strategy](<https://devfeed.tech/tags/strategy.md>)

### AI overview

Standard active learning can fail when deployment data has a different subpopulation mix from the data sampled for labeling. The paper proposes selecting data for labeling with collection metadata as a noisy proxy for each sample's subpopulation, aiming to train classifiers that generalize across subpopulations under this specific form of distribution shift. The abstract reports improved classifier accuracy across subpopulations in empirical evaluation; it does not specify the datasets or the size of the improvement.

### Source excerpt

arXiv:2611.09520v1 Announce Type: new Abstract: In many domains from species identification to cancer diagnosis, labelling data to train predictive models remains costly. In these settings, it is advantageous to intelligently select data points to label to achieve the maximum information gain, this is the motivation behind the study of active learning. In addition to the high costs of acquiring data in such domains, there is the added complexity of data distribution shifts that can occur between training and deployment. For example, the features of cancerous skin cells taken in one country may be different in another region of the world. These two challenges - efficient labelling of data, and learning models robust to distribution shifts have been studied separately but not together, even though these challenges often co-occur in real-life settings. We identify failure modes of standard active learning approaches when the distribution of data at deployment is different from the distribution in which the data is sampled for labelling. We propose a new strategy for efficiently learning classifiers that generalise under specific forms of distribution shift, namely subpopulation shift. Drawing on work in learning from heterogeneous data, we aim to train models that perform well across subpopulations of the data, through active learning. To handle this challenging setting, we exploit the fact that many datasets include metadata about data collection that can serve as a noisy proxy for each sample's underlying subpopulation. We demonstrate how to leverage these noisy labels during active learning selection, and demonstrate empirically that our approach improves classifier accuracy across subpopulations.

## Revisiting the Generic Transformer: Deconstructing a Strong Baseline for Time Series Foundation Models

DevFeed: [Revisiting the Generic Transformer: Deconstructing a Strong Baseline for Time Series Foundation Models](<https://devfeed.tech/articles/revisiting-the-generic-transformer-deconstructing-a-strong-baseline-for-time-series-foundation-models-123096.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2602.06909>)

Author: Yunshi Wen, Wesley M. Gifford, Chandra Reddy, Lam M. Nguyen, Jayant Kalagnanam, Anak Agung Julius

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

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

Topics: [Time Series](<https://devfeed.tech/topics/time-series.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>), [data-engineering](<https://devfeed.tech/topics/data-engineering.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [ablation](<https://devfeed.tech/tags/ablation.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [forecasting](<https://devfeed.tech/tags/forecasting.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [masking](<https://devfeed.tech/tags/masking.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [scalability](<https://devfeed.tech/tags/scalability.md>), [time-series](<https://devfeed.tech/tags/time-series.md>), [time-series-foundation-models](<https://devfeed.tech/tags/time-series-foundation-models.md>), [transformer](<https://devfeed.tech/tags/transformer.md>), [transformer-architecture](<https://devfeed.tech/tags/transformer-architecture.md>), [zero-shot](<https://devfeed.tech/tags/zero-shot.md>)

### AI overview

The study reports that a standard patch Transformer with a straightforward training protocol can achieve strong zero-shot time series forecasting performance, with accuracy substantially affected by pretraining data composition, contiguous patch masking, and context length. Its controlled ablations examine scaling, data, and training choices, and the authors report that the protocol generalizes across benchmarks and alternative architectures. They say pretrained models and detailed protocol findings are released to support a transparent, reproducible baseline.

### Source excerpt

arXiv:2602.06909v2 Announce Type: replace Abstract: The recent surge of interest in Time Series Foundation Models has rapidly advanced the field. Yet, heterogeneous training setups across studies make it difficult to determine whether performance improvements came from architectural innovations or effective data engineering. In this work, we investigate whether a standard patch Transformer architecture, paired with a straightforward training protocol, can match the zero-shot forecasting performance of state-of-the-art TSFMs. We conduct a comprehensive ablation study covering model scaling, data composition, and training techniques to examine the ingredients behind this performance. Our findings show that pretraining data composition, contiguous patch masking, and context length have a substantial effect on forecasting accuracy, and that the generic architecture itself can deliver high performance with good scalability. By strictly controlling these experimental variables, we provide comprehensive empirical results on model scaling across multiple dimensions. Furthermore, we establish that the effectiveness of this training protocol generalizes robustly across benchmarks and alternative architectures. We release our pretrained models and document the detailed training protocol and findings to establish a transparent, reproducible baseline for future research.

## Change One Monty Hall Rule and Five LLMs Still Say Switch

DevFeed: [Change One Monty Hall Rule and Five LLMs Still Say Switch](<https://devfeed.tech/articles/change-one-monty-hall-rule-and-five-llms-still-say-switch-122523.md>)

Original publisher: [Read original article](<https://towardsdatascience.com/change-one-monty-hall-rule-and-five-llms-still-say-switch/>)

Author: Utkarsh Mangal

Published: 2026-11-08T13:00:00Z

Content type: article

Language: en

Sources: [Towards Data Science](<https://devfeed.tech/sources/towards-data-science.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Chain-of-thought](<https://devfeed.tech/topics/chain-of-thought.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [bayesian-statistics](<https://devfeed.tech/tags/bayesian-statistics.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [experiment](<https://devfeed.tech/tags/experiment.md>), [game](<https://devfeed.tech/tags/game.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llms](<https://devfeed.tech/tags/llms.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [models](<https://devfeed.tech/tags/models.md>), [monty-hall-problem](<https://devfeed.tech/tags/monty-hall-problem.md>), [probability](<https://devfeed.tech/tags/probability.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [switch](<https://devfeed.tech/tags/switch.md>)

### AI overview

In a small test of five models across six Monty Hall variants, changing the host's rule often failed to change the models' answer: they said "switch" even when switching could never win. The author explains the correct calculation by weighting each car location by how likely the host was to open a goat door or offer a swap under that scenario. On four variants where switching was wrong, models assigned the correct answer just 0-13% probability on average; step-by-step sampled reasoning produced only 30-37% accuracy, close to always switching. Controls showed that some models could answer trivial questions or use a directly stated clue, yet still ignored evidence in the host's behavior. The study is limited to five small models, six stories and short chains; the author's conclusion is that getting a familiar puzzle right is weak evidence of reasoning, and changing a rule to make the answer flip is a useful probe.

### Source excerpt

Even with step-by-step reasoning, the models landed near 33% accuracy, and one quoted the host's rule correctly before ignoring it. The post Change One Monty Hall Rule and Five LLMs Still Say Switch appeared first on Towards Data Science.

[Next page](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md?cursor=WyIyMDI2LTExLTA4VDEzOjAwOjAwKzAwOjAwIiwgIjQ2OTgyOGFkLTQwZTUtNDQ2NC04NWY0LWMwNDUxZGI2ODgzYSJd>)