# objectives Published articles for objectives. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## How I learned to stop worrying and love StopGrads: Stationarity, Convergence, and a case study on Flow Map Learning DevFeed: [How I learned to stop worrying and love StopGrads: Stationarity, Convergence, and a case study on Flow Map Learning]() Original publisher: [Read original article]() Author: Max W. Shen, Mark Goldstein, Zichu Wang, Aahlad Puli, Rajesh Ranganath Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Reinforcement learning]() Tags: [case-study](), [characterization](), [convergence](), [cs-lg](), [diffusion](), [expression](), [flow](), [functional](), [gradient](), [grounding](), [guarantees](), [learning](), [machine](), [machine-learning](), [map](), [maps](), [memory](), [models](), [objectives](), [optimizing](), [point]() ### AI overview The paper proposes a stopgrad regression principle that characterizes stationary points and their uniqueness across objectives for flow maps, reinforcement learning, and diffusion samplers. For flow map learning, it reports that the unique stationary point is the true flow map and establishes positive convergence results for Eulerian and Lagrangian objectives, including MeanFlow and improved MeanFlow. Under functional semi-gradient flow, it derives a closed-form learned flow map by composing the initial and true flow maps. It also proposes changed stopgrad placements for flow map objectives that the authors report reduce training memory by 2x. ### Source excerpt arXiv:2609.16222v2 Announce Type: replace Abstract: Stopgrads are widely used in training machine learning models, but stopgrads can alter the gradient, stationary points and convergence guarantees of the original objective, which can make stopgrad training theoretically ungrounded. We introduce a stopgrad regression principle, which identifies a general template for stopgrad objectives with a closed-form characterization of stationary points and their uniqueness, unifying stopgrad objectives for flow maps, reinforcement learning, and diffusion samplers. We provide theoretical grounding for optimizing stopgrad flow map objectives by showing their unique stationary point is the true flow map, and showing positive convergence results for Eulerian and Lagrangian objectives, including MeanFlow and improved MeanFlow. Remarkably, we show that under functional semi-gradient flow, the learned flow map has a closed-form expression composing the initial flow map and the true flow map. We additionally use our stopgrad regression principle to propose modified stopgrad placements for flow map objectives which reduce training memory by 2x. ## Coresets for Capacitated Clustering via Dual Concentration DevFeed: [Coresets for Capacitated Clustering via Dual Concentration]() Original publisher: [Read original article]() Author: Milind Prabhu, Chris Schwiegelshohn, Sudarshan Shyam, Erik Waingarten Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [clustering](), [Algorithms]() Tags: [algorithms](), [clustering](), [cost](), [cs-cg](), [cs-ds](), [cs-lg](), [data](), [datasets](), [dimension](), [distribution](), [flow](), [input](), [large-datasets](), [linear](), [means](), [median](), [metric](), [objectives](), [problems](), [sampling](), [scaling](), [size](), [small]() ### AI overview The paper develops coresets for capacitated k-median and k-means, where assignments must satisfy prescribed cluster-size constraints. It reports optimal-size coresets for finite metric spaces, matching the best-known bounds for unconstrained clustering. For Euclidean space, it gives improved bounds, including the first coresets with near-linear dependence on k that are independent of dimension and input size. Its analyses use concentration bounds for dual solutions of min-cost flow problems to simplify and tighten the sampling-distribution analysis. ### Source excerpt arXiv:2611.09600v1 Announce Type: cross Abstract: We study coresets for clustering problems, focusing on the classical $k$-median and $k$-means objectives. A coreset is a small weighted summary of the input data that approximately preserves the clustering cost for every choice of centers. Such summaries are a central tool for scaling clustering algorithms to large datasets. In this work, we consider the capacitated setting, where the clustering is subject not only to choosing centers but also to respecting prescribed cluster-size constraints. This model is a natural generalization of standard clustering and has been widely studied, but its coreset complexity is less well understood. For finite metric spaces, we obtain optimal-size coresets; surprisingly, these match the best-known bounds for unconstrained clustering. We also give improved coreset bounds in Euclidean space, notably the first coresets with near-linear dependence in $k$ while being independent of the dimension and input size. The main new technical ingredients are concentration bounds for the dual solutions of min-cost flow problems, which allow for both significantly simpler and tighter analyses of the underlying sampling distribution used in the coreset construction. ## Steerable Multi-Objective Group-Relative Preference Optimization with Jointly Trained Low-Rank Adapters DevFeed: [Steerable Multi-Objective Group-Relative Preference Optimization with Jointly Trained Low-Rank Adapters]() Original publisher: [Read original article]() Author: Constantinos Karouzos, Nikolaos Aletras Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [grpo](), [post-training](), [Reinforcement learning](), [lora](), [Fine-tuning](), [Large Language Model]() Tags: [adapters](), [cs-ai](), [cs-cl](), [cs-lg](), [grpo](), [inference](), [instructions](), [language-model](), [math](), [objectives](), [optimization](), [reinforcement-learning](), [tasks]() ### AI overview LambdaMix learns a preference-conditioned family of policies in one training run: it jointly trains one low-rank adapter per objective over a frozen base model, then uses a preference vector to select the adapter mixture and weight separately normalized rewards during group-relative online reinforcement learning. Preferences can change at inference without further optimization. Across four models, it reports higher hypervolume than evaluated conditioned and policy-merging baselines on helpfulness/harmlessness and math-solving/instruction-following tasks. It also uses fewer training completions and stores fewer adapter parameters than a set of GRPO specialist policies. On two larger models, it evaluates comment rewriting to preserve wording while reducing personal-attribute clues. With architecture and initialization fixed, varying adapter mixtures and objective weights together yields wider score ranges and higher coverage than varying either alone. ### Source excerpt arXiv:2611.09634v1 Announce Type: cross Abstract: Language model deployment often requires different trade-offs between competing objectives as user preferences change, yet standard post-training methods typically produce a single policy for a fixed objective weighting. We introduce LambdaMix, a new parameter efficient method for learning a preference conditioned family of policies with group-relative online reinforcement learning. LambdaMix jointly trains one low-rank adapter per objective over a frozen base model. One preference vector selects the adapter mixture that generates responses and weights the separately normalized rewards used to update the adapters. A single training run supports changing preferences at inference without further optimization. We experiment with four models on helpfulness and harmlessness and on solving math problems while following instructions. We also evaluate the two largest models on rewriting comments to preserve wording while reducing clues about personal attributes. On the first two tasks, LambdaMix achieves higher hypervolume than all evaluated conditioned and policy merging baselines across all four models. LambdaMix uses fewer training completions and stores fewer adapter parameters than a set of GRPO specialist policies. With architecture and initialization fixed, varying adapter mixtures and objective weights together gives wider score ranges and higher coverage than varying either alone. ## Flow-Direct: Feedback-Efficient and Reusable Guidance for Flow Models via Non-Parametric Guidance Field DevFeed: [Flow-Direct: Feedback-Efficient and Reusable Guidance for Flow Models via Non-Parametric Guidance Field]() Original publisher: [Read original article]() Author: Kim Yong Tan, Yueming Lyu, Ivor Tsang, Yew-Soon Ong Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Electrical engineering and computer science (EECS)]() Tags: [cs-ai](), [cs-lg](), [dataset](), [diffusion](), [distribution](), [estimator](), [feedback](), [flow](), [guidance](), [machine](), [objectives](), [optimization](), [optimize](), [ratio](), [reward](), [sample](), [training]() ### AI overview Flow-Direct uses a persistent guidance field to steer pretrained diffusion and flow models toward objectives evaluated by black-box reward functions. The field is theoretically derived from the log-density ratio between the base and reward-weighted target distributions, and is estimated in practice from all accumulated reward-evaluated samples. As samples accrue, the estimate is intended to improve, reusing feedback that other methods may discard after localized gradient approximations or discrete search decisions. After optimization, the collected dataset defines a field that can generate new target samples without further reward evaluations; separate fields can also be combined for multiple objectives. ### Source excerpt arXiv:2605.16348v2 Announce Type: replace Abstract: Training-free guidance enables pre-trained diffusion and flow models to optimize application-specific objectives using feedback from external black-box reward functions. However, existing methods are feedback-inefficient because reward feedback is used only transiently to inform a localized gradient approximation or a discrete search decision, and is subsequently discarded. To address this limitation, we propose Flow-Direct, a framework that guides the generation process via a persistent guidance field. Theoretically, this guidance field is analytically derived from the log-density ratio between the base and reward-weighted target distributions; it transports the pre-trained distribution to the target distribution. In practice, the field is implemented as a non-parametric estimator constructed from all accumulated reward-evaluated samples. As more samples are collected during optimization, this empirical guidance field becomes increasingly accurate. This persistent formulation yields two major advantages. First, Flow-Direct is highly feedback-efficient: because every evaluated sample is used to refine the global guidance field, no reward information is wasted. Second, the framework is naturally reusable: once optimization is complete, the collected dataset defines a reusable guidance field for generating novel target samples without additional reward evaluations, and distinct guidance fields can be combined to generate samples that simultaneously satisfy multiple objectives. ## Multi-objective Hyperparameter Optimization with Expert Priors DevFeed: [Multi-objective Hyperparameter Optimization with Expert Priors]() Original publisher: [Read original article]() Author: Soham Basu, Nastaran Alipour, Frank Hutter, Danny Stoll Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Hyperparameter optimization]() Tags: [algorithm](), [benchmarks](), [cs-lg](), [deep-learning](), [experts](), [hyperparameter-optimization](), [objectives](), [optimization](), [user](), [vision-language-model]() ### AI overview PriMO is an HPO algorithm designed to incorporate expert beliefs across multiple objectives, addressing a gap in methods that use prior knowledge. The abstract reports state-of-the-art performance in both multi-objective and single-objective settings across eight deep-learning benchmarks, and at larger scale when optimizing a vision-language model over three objectives. The supplied text does not describe the method's implementation or benchmark details. ### Source excerpt arXiv:2511.08371v2 Announce Type: replace Abstract: While Deep Learning (DL) experts often have prior knowledge about which hyperparameter settings yield strong performance, only few Hyperparameter Optimization (HPO) algorithms can leverage such prior knowledge and none incorporate priors over multiple objectives. As DL practitioners often need to optimize not just one but many objectives, this is a blind spot in the algorithmic landscape of HPO. To address this shortcoming, we propose the novel problem setting of multi-objective HPO with expert priors and introduce PriMO, the first HPO algorithm that can integrate user beliefs over multiple objectives. We show PriMO achieves state-of-the-art performance across 8 DL benchmarks in the multi-objective and single-objective setting. Further, we demonstrate that PriMO remains state-of-the-art at large scales by optimizing a vision-language model over three objectives. ## Interactive Backdoor Attacks in LLM-Based Proactive Dialogue Systems DevFeed: [Interactive Backdoor Attacks in LLM-Based Proactive Dialogue Systems]() Original publisher: [Read original article]() Author: Zhiyang Xia, Yuming Shang, Xi Zhang Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [LLM security](), [Adversarial attacks](), [ai security](), [Fine-tuning]() Tags: [attacks](), [backdoor](), [cs-ai](), [cs-cl](), [defense](), [experiments](), [fine-tuning](), [framework](), [guidance](), [interaction](), [interactive](), [llm](), [models](), [objectives](), [open](), [policy](), [post](), [proactive](), [prompt](), [steering](), [systems](), [utility]() ### AI overview Poisoned fine-tuning dialogues can implant a trigger-conditioned bias that steers an LLM's later multi-turn dialogue toward an attacker's objective, activated by ordinary user utterances without deployment-time prompt changes or further attacker action. The framework covers an immediate Direct Attack, a Constrained Progressive Attack along a predefined trajectory, and an Unconstrained Progressive Attack across open-ended user continuations. In psychological counseling experiments using two model backbones and two dialogue datasets, all three reportedly achieved their objectives while largely preserving normal utility. The tested training-data filters retained nearly all poisoned dialogues, and three post-admission defense baselines left substantial attack-specific success. ### Source excerpt arXiv:2611.06969v1 Announce Type: cross Abstract: LLM-based proactive dialogue systems provide goal-directed support through sustained interaction. Fine-tuning their underlying models on externally supplied dialogue data introduces backdoor risks. Existing LLM backdoor studies primarily focus on eliciting an attacker-specified response upon trigger activation. We present an interactive backdoor attack framework that uses training-data poisoning to steer post-trigger, multi-turn interactions toward an attacker-specified objective. Poisoned fine-tuning dialogues implant a trigger-conditioned guidance-policy bias that ordinary user utterances can activate, without deployment-time prompt modification or further attacker intervention. Specifically, we instantiate three complementary attacks under progressively expanded post-trigger interaction settings: (1) Direct Attack, which targets an immediate response deviation upon trigger activation, retaining the conventional trigger--response setting; (2) Constrained Progressive Attack, which progressively advances the attack objective along a predefined multi-turn trajectory; and (3) Unconstrained Progressive Attack, which extends target-directed steering to open-ended interactions with varied user continuations. Using psychological counseling as a representative setting, experiments on two LLM backbones and two dialogue datasets show that all three attacks effectively achieve their objectives under their respective interaction settings while largely preserving normal utility. Separate defense evaluations show that the tested training-data filters retain nearly all poisoned dialogues, while three post-admission baselines leave substantial residual attack-specific success. ## The Alignment Tax Is Graded: Task-Selective Capability Loss Under Length-Normalized DPO with SFT Retention DevFeed: [The Alignment Tax Is Graded: Task-Selective Capability Loss Under Length-Normalized DPO with SFT Retention]() Original publisher: [Read original article]() Author: Ye Yuan Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [dpo](), [ai alignment](), [Fine-tuning](), [post-training](), [lora]() Tags: [alignment](), [characterization](), [cs-lg](), [dpo](), [financial](), [format](), [json](), [lora](), [margin](), [measurement](), [model](), [objectives](), [qwen3-5](), [retention](), [sft](), [synthetic](), [tasks](), [tax](), [testing]() ### AI overview In a study of Qwen3.5-4B on three structured financial-filing tasks, increasing the implicit reward margin under length-normalized DPO with an SFT retention term produced task-selective capability loss. Across 8 seeds, extraction was preserved within a ±5-point equivalence margin (−0.55 points); JSON performance fell 3.52 points, with equivalence not established; and router performance fell 30 points, mainly in a margin 1-3 damage window that included a transient format-break spike. At least 96% of margin growth came from suppressing rejected completions while chosen reward stayed approximately flat; the study does not assign the behavioral damage to a single gradient channel. Measured reward margins differed by up to 3.4x between nf4 training and fp16 evaluation on the unquantized base. These findings are limited to one 4B model, sparse LoRA, easily separable synthetic preferences, and the tested objective; the authors leave a no-retention control and measured format-alignment metric for future work. ### Source excerpt arXiv:2611.06476v2 Announce Type: replace Abstract: Setting. This is an empirical characterization study; we propose no new method and run no comparisons against alternative objectives. We fine-tune Qwen3.5-4B with sparse LoRA and length-normalized DPO plus an SFT retention term (the RPO objective) on structured financial-filing tasks, and track downstream capability as the implicit reward margin grows to $\sim 7$ $\beta$-scaled units (nf4 training; $\sim 3$ under fp16 measurement on the unquantized base). Question. Does margin growth degrade capability uniformly across tasks, or selectively? Findings. In a pre-registered study (8 seeds, 6 unseen at freeze) with artifact-aware scoring and TOST equivalence testing, the alignment tax is graded and task-selective: extraction is preserved ($\Delta = -0.55$ pts, equivalent at $\pm5$ pts), json loses little ($\Delta = -3.52$ pts, equivalence fails on a clustered item-level interval), and router degrades severely ($\Delta = -30$ pts). Router's loss concentrates in a margin 1-3 damage window containing a transient format-break spike. The margin is accounted for almost entirely ($\geq 96\%$) by suppression of rejected completions, with chosen reward approximately flat, consistent with the retention term anchoring it; we do not attribute the behavioral damage to any single gradient channel. Gold-logp deflation is task-selective (extraction $-0.15$, json $-1.04$, router $-1.47$ nats/token under fp16), with gold selectively protected relative to incorrect controls on extraction (8/8 seeds). QLoRA reward margins differ up to $3.4\times$ between nf4 training and unquantized fp16 measurement, with an axis-dependent gap. Scope. Length-normalized rewards with SFT retention, sparse LoRA on 8/32 layers, one 4B model, easily-separable synthetic preferences, three structured tasks. A no-retention control and a measured format-alignment metric are left to future work. ## Mirror Descent-Ascent for mean-field min-max problems DevFeed: [Mirror Descent-Ascent for mean-field min-max problems]() Original publisher: [Read original article]() Author: Razvan-Andrei Lascu, Mateusz B. Majka, {\\L}ukasz Szpruch Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Operations research and optimization]() Tags: [algorithm](), [analysis](), [commutator](), [convergence](), [convex](), [cs-lg](), [math-oc](), [math-pr](), [mathematics](), [objectives](), [optimization](), [space](), [spaces]() ### AI overview The paper analyzes simultaneous and alternating mirror descent-ascent for convex-concave min-max problems over measures, assuming the payoff is relatively smooth with respect to a suitable Bregman divergence defined via flat derivatives. It establishes convergence to mixed Nash equilibria measured by Nikaidô-Isoda error: O(N⁻¹/²) for simultaneous updates and an improved O(N⁻²/³) for alternating updates. The analysis relates Bregman divergences on measures to dual divergences on bounded continuous functions, enabling control of asymmetric terms from alternating updates. The results extend prior analyses limited to bilinear objectives to nonlinear convex-concave problems. This overview is based on the supplied abstract. ### Source excerpt arXiv:2402.08106v4 Announce Type: replace-cross Abstract: We study two variants of the mirror descent-ascent (MDA) algorithm for solving min-max problems on the space of measures: simultaneous and alternating. We work under assumptions of convexity-concavity and relative smoothness of the payoff function with respect to a suitable Bregman divergence, defined on the space of measures via flat derivatives. We establish non-asymptotic convergence rates to mixed Nash equilibria, measured in the Nikaid\^{o}-Isoda error, proving an $\mathcal{O}(N^{-1/2})$ rate for simultaneous MDA and an improved $\mathcal{O}(N^{-2/3})$ rate for alternating MDA. The main technical contribution is an infinite-dimensional dual space analysis that relates Bregman divergences on measures to dual Bregman divergences on spaces of bounded continuous functions, allowing us to control asymmetric commutator terms created by alternating updates. The results substantially generalize prior analyses restricted to bilinear objectives and also apply to nonlinear convex-concave problems on measure spaces, thereby providing a unified theoretical foundation for MDA in mean-field min-max optimization. ## Indirect Prompt Injection in the Wild: An Empirical Study of Prevalence, Techniques, and Objectives DevFeed: [Indirect Prompt Injection in the Wild: An Empirical Study of Prevalence, Techniques, and Objectives]() Original publisher: [Read original article]() Author: Soheil Khodayari, Xuenan Zhang, Bhupendra Acharya, Giancarlo Pellegrino Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [LLM security](), [prompt injection](), [ai security](), [AI Bots]() Tags: [automation](), [bot-detection](), [compliance](), [content-protection](), [crawlers](), [cs-cr](), [customer-support-agents](), [experiments](), [headers](), [hidden](), [hiring](), [html](), [impact](), [indirect-prompt-injection](), [llms](), [metadata](), [objectives](), [persistence](), [prompt-injection](), [study](), [techniques]() ### AI overview The study finds indirect prompt injections in ordinary web content at measurable scale: among 1.2 billion URLs across 24.8 million hosts, it validated 15,300 instances on 11,700 pages, often following recurring templates. The instructions serve varied aims, including disruption, reputation manipulation, content protection, and AI-bot detection, and may target crawlers, search systems, support agents, or hiring workflows. About 70% appeared in non-rendered HTML, while some visible cases were concealed through rendering techniques. In 5,200 controlled experiments across 13 models and four webpage representations, compliance reached up to 8% for smaller models given plain text; structured representations reduced compliance, which the authors attribute to preserved structural cues. The abstract reports limited but non-negligible compliance and does not establish broader real-world impact. ### Source excerpt arXiv:2604.27202v2 Announce Type: replace Abstract: As LLMs are increasingly integrated into systems that browse, retrieve, summarize, and act on web content, webpages have become an untrusted input vector for downstream model behavior. This enables site owners, contributors, and adversaries to embed instructions directly in web resources, i.e., indirect prompt injections. While prior work demonstrates such attacks in controlled settings, their prevalence, deployment, and real-world impact remain unclear. We present one of the first large-scale empirical analyses of indirect prompt injections in webpages and HTTP responses. Analyzing 1.2B URLs from 24.8M hosts, we identify 15.3K validated instances across 11.7K pages. These are not isolated cases: a small number of recurring templates account for most cases. We characterize their objectives, delivery mechanisms, visibility, persistence, and impact, revealing a heterogeneous ecosystem spanning disruptive prompts, reputation manipulation, content-protection directives, and AI-bot detection, targeting systems such as crawlers, search pipelines, customer-support agents, and hiring workflows. A key finding is that most instructions target machines rather than humans: about 70% appear in non-rendered HTML (e.g., headers, comments, metadata), and many visible cases are hidden via rendering techniques. To assess practical risk, we run 5,200 controlled experiments across 13 models and four webpage representations. Our results show compliance is limited but non-negligible, reaching up to 8% for smaller models on plain-text inputs, while structured representations reduce compliance by preserving structural cues. Overall, prompt-based interference is already present in the web ecosystem and represents a growing source of tension between LLM-driven automation and the sites it consumes. ## A Definition of Catastrophic Performance DevFeed: [A Definition of Catastrophic Performance]() Original publisher: [Read original article]() Author: Henrik Marklund, Alex Infanger, Benjamin Van Roy Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [ai alignment](), [ai evaluation]() Tags: [agent](), [ai](), [ai-misalignment](), [align](), [analysis](), [benefit](), [cs-ai](), [cs-lg](), [definition](), [environment](), [examples](), [formal](), [misalignment](), [objectives](), [performance](), [progress](), [reference](), [risk](), [state](), [theoretical](), [unrelated](), [value]() ### AI overview The paper proposes a formal definition to distinguish catastrophic outcomes of AI misalignment from merely suboptimal performance. It compares an agent's value with two references: a baseline for what would be attained if no agent were deployed, and an orthogonal value for a trajectory unrelated to human preferences. The authors say orthogonal value is typically much worse than baseline. In examples where prior human activity has improved the environment, catastrophe means an agent deployed to accelerate progress undoing those human-created benefits. This overview is based on the supplied abstract. ### Source excerpt arXiv:2611.06219v1 Announce Type: new Abstract: A central challenge in AI is specifying objectives that align with human preferences. While existing theoretical results elucidate how misspecification impacts performance, they typically do not distinguish between merely suboptimal and catastrophic outcomes. To enable rigorous analysis of catastrophic risk from AI misalignment, we introduce a formal definition of catastrophic performance. Our definition is based on two reference values. The first, baseline value, captures the value attained if no agent is deployed. The second, orthogonal value, captures the value of a trajectory that is unrelated to human preferences, in a sense we describe. Orthogonal value is typically far worse than baseline value. We interpret these reference values through examples in which past human activity has altered the state of the environment to human benefit. Starting at the resulting state, an AI agent is deployed with the goal of accelerating further progress. Catastrophic performance represents this agent's undoing of benefits afforded by human activity. ## EviSeek: An Evidence-Centric Agentic Framework for Embodied Active Information Seeking DevFeed: [EviSeek: An Evidence-Centric Agentic Framework for Embodied Active Information Seeking]() Original publisher: [Read original article]() Author: Jintao Li, Zhiyong Cao, Baihua Xiao, Yu Bai, B\\"orje F. Karlsson Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Machine Perception]() Tags: [acquisition](), [actions](), [agent](), [agentic](), [arxiv](), [bench](), [cs-ai](), [framework](), [gemini](), [gemini-3-1-pro](), [information](), [interaction](), [linking](), [model](), [needs](), [objectives](), [points](), [question](), [questions](), [qwen3](), [retention]() ### AI overview EviSeek is a training-free framework for embodied question answering that turns question-derived evidence needs into explicit acquisition objectives. It links actions conditioned on observations and selectively retained evidence to what the agent needs to learn, addressing interactions whose informational purpose may otherwise remain implicit. Across 3,436 ESI-Bench questions, it improved accuracy over the Single-model agent by 7.1 percentage points with Qwen3-VL-32B and 3.7 points with Gemini 3.1 Pro, while also improving visual-support-weighted accuracy and reducing mean token usage. ### Source excerpt arXiv:2611.06434v1 Announce Type: new Abstract: Active embodied question answering requires agents to acquire evidence through viewpoint changes and physical interaction. The central challenge is to avoid undirected interaction: taking plausible actions without specifying what evidence they should reveal for answering the question. Action plans and accumulated histories alone can leave the informational purpose of an interaction implicit. We introduce EviSeek, a training-free framework that makes question-derived evidence demands explicit acquisition objectives, linking observation-conditioned actions and selectively retained evidence to the information the agent seeks. On 3,436 ESI-Bench questions, EviSeek improves accuracy over the Single-model agent by 7.1 percentage points with Qwen3-VL-32B and 3.7 points with Gemini 3.1 Pro, while improving visual-support-weighted accuracy and reducing mean token usage. These findings support making the informational purpose of interaction explicit, with action selection and evidence retention organized around what the agent needs to learn to answer the question. ## DenseTwins: Dense Modality Decoupling for Multimodal Self-Supervision DevFeed: [DenseTwins: Dense Modality Decoupling for Multimodal Self-Supervision]() Original publisher: [Read original article]() Author: Carlo Sgaravatti, Daniele Kopyshevskiy, Davide Secco, Giacomo Boracchi Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Representation learning](), [multimodal-ai]() Tags: [alignment](), [applications](), [autonomous-driving](), [benchmarks](), [cs-ai](), [cs-cv](), [cs-lg](), [decoupling](), [embedding](), [features](), [framework](), [multimodal](), [objectives](), [performance](), [reduction](), [redundancy](), [robotics](), [self-supervised-learning](), [semantics](), [sensing](), [space](), [spatial](), [supervision]() ### AI overview DenseTwins learns dense multimodal representations by separating shared features from modality-specific ones, aiming to preserve both cross-modal semantics and spatial structure. It combines dense cross-modal correspondences with redundancy-reduction objectives: an intra-modal objective connects augmented views, while a cross-modal objective aligns shared features and decorrelates modality-specific information. The abstract reports state-of-the-art or competitive performance against multimodal self-supervised baselines across multiple benchmarks and sensing setups; it provides no benchmark-specific results or further implementation details. ### Source excerpt arXiv:2611.04710v1 Announce Type: cross Abstract: Multimodal perception is central to many computer vision applications, including robotics and autonomous driving, where different sensing modalities provide complementary cues. Self-supervised learning offers a scalable solution for exploiting large amounts of unlabeled multimodal data. However, existing methods either neglect modality-specific information by enforcing feature-space alignment between modalities or learn only global representations without preserving spatial structure. We propose DenseTwins, a multimodal self-supervised framework that learns dense representations by explicitly decoupling the dense feature space into shared and modality-specific components. The proposed approach combines dense cross-modal correspondences with redundancy-reduction objectives, enabling the model to capture both shared semantics and modality-specific cues within a structured dense embedding space. DenseTwins jointly optimizes an intra-modal objective across augmented views and a cross-modal objective that aligns shared features while decorrelating modality-specific information. We evaluate DenseTwins across multiple multimodal benchmarks and sensing setups, achieving state-of-the-art or competitive performance across multimodal self-supervised baselines. Code available at: https://github.com/CarloSgaravatti/DenseTwins. ## SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction DevFeed: [SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction]() Original publisher: [Read original article]() Author: Minsu Kim, Ye-Sung Kim, Hyeseong Jeon, Wooseok Hyung, Joshua Lee, Chang-Hwan Im Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [AI Foundation Models](), [Machine learning](), [Representation learning]() Tags: [architecture](), [arxiv](), [attention](), [bci](), [cs-lg](), [datasets](), [eeg](), [efficiency](), [foundation-model](), [machine-learning](), [neural](), [objectives](), [parameter](), [prediction](), [recording](), [robustness](), [signal](), [similarity](), [spatial](), [structure](), [tasks]() ### AI overview SPERA is an EEG foundation model that predicts in latent space using a joint-embedding predictive architecture (JEPA), aiming to avoid reconstructing observed signals that mix neural and non-neural components. It adds a Legendre-polynomial spatial prior in attention to represent differing scalp electrode geometries, interleaves factorized temporal and spatial attention with periodic full-attention blocks, and uses a relational spectral regularizer to align latent similarities with spectral views. Pretrained on about 80,000 hours of EEG from 29,048 subjects across 106 datasets, it reports the highest average balanced accuracy across nine clinical, cognitive, and BCI tasks, along with parameter efficiency under linear probing and robustness across recording conditions. These findings are from the supplied abstract. ### Source excerpt arXiv:2610.10571v2 Announce Type: replace Abstract: Electroencephalography (EEG) provides a non-invasive measure of ongoing neural activity, but building general-purpose EEG models remains challenging due to the heterogeneity of subjects, devices, and electrode montages. Existing EEG foundation models predominantly rely on reconstruction-based objectives defined on the observed signal, which contains both neural and non-neural components. We introduce the Spherical Prior EEG Representation Architecture (SPERA), an EEG foundation model that adopts the joint-embedding predictive architecture (JEPA) to predict in latent space. SPERA introduces a Legendre-polynomial spatial prior, incorporated into attention to encode varying scalp electrode geometries. Two further components adapt the model to EEG: factorized temporal and spatial attention interleaved with periodic full-attention blocks, and a relational spectral regularizer aligning latent similarity structure with spectral views. Pretrained on approximately 80,000 hours of EEG from 29,048 subjects across 106 datasets, SPERA achieves the highest average balanced accuracy across nine downstream tasks spanning clinical, cognitive, and BCI applications. SPERA further exhibits strong parameter efficiency under linear probing and robustness across varying recording conditions, suggesting its potential as a general-purpose backbone for diverse EEG analyses. ## TopoDPO: Dense Preference Alignment for Structure-Sensitive Aerial Referring Segmentation DevFeed: [TopoDPO: Dense Preference Alignment for Structure-Sensitive Aerial Referring Segmentation]() Original publisher: [Read original article]() Author: Huaiyuan Peng, Jisheng Dang, Wencan Zhang, Jingwen Zhao, Hong Peng, Wenjun Ma, Bin Hu, Qi Tian, Tat-Seng Chua Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [dpo]() Tags: [alignment](), [arxiv](), [bootstrap](), [code](), [component](), [cs-ai](), [cs-cv](), [direct-preference-optimization](), [dpo](), [error](), [framework](), [image](), [intersection](), [local](), [margins](), [masks](), [metrics](), [mining](), [multimodal](), [objectives](), [optimization](), [structure]() ### AI overview TopoDPO addresses the mismatch between spatially varied mask preferences and image-level preference objectives for aerial referring segmentation. It selects preference pairs using connected-component discrepancy and macro-corner recall, then applies preference penalties to local margins before spatial aggregation against a frozen supervised reference. On the 4,590-instance Topo-RS-Hard set, it raises IoU from 50.91% to 51.24% over image-level DPO and lowers Betti-0 error from 1.34 to 0.90. Reported paired bootstrap intervals exclude zero for all five metrics. Analysis across 39,244 validation records finds the clearest structural gains where pre-alignment component-count errors are moderate or severe. ### Source excerpt arXiv:2611.01551v1 Announce Type: cross Abstract: Preference optimization for multimodal segmentation faces a granularity mismatch: dense masks contain spatially heterogeneous preference signals, whereas image-level objectives assign one shared nonlinear weight to all spatial margins. We introduce TopoDPO, a topology-aware dense preference alignment framework for structure-sensitive aerial referring segmentation. Topo-Prior mines informative preference pairs using connected-component discrepancy and macro-corner recall, while the dense objective applies the preference penalty to each local margin before spatial aggregation against a frozen supervised reference. On Topo-RS-Hard, a 4590-instance structural stress set, TopoDPO improves intersection over union (IoU) from 50.91% to 51.24% over image-level direct preference optimization (DPO) and reduces Betti-0 error from 1.34 to 0.90; paired image-cluster bootstrap intervals against image-level DPO exclude zero for all five reported metrics. Matched supervised controls and a complete factorial analysis separate the effects of topology-aware pair mining, localized correction, and dense reference-relative preference weighting. An unfiltered stratification over all 39,244 validation records further shows the clearest structural gains on samples with moderate or severe pre-alignment component-count errors. The code is publicly available at: https://github.com/phy709/TopoDPO. ## Controlling Sharpening in LLM Post-Training DevFeed: [Controlling Sharpening in LLM Post-Training]() Original publisher: [Read original article]() Author: Simon Matrenok, Nicolas Menet, Nicolas Flammarion, Andreas Krause Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [llm-reasoning](), [post-training](), [Large Language Model]() Tags: [adaptive](), [algorithms](), [code-generation](), [completion](), [control](), [coverage](), [cs-lg](), [llm](), [mathematical-reasoning](), [objectives](), [optimization](), [policies](), [policy](), [post-training](), [prompt](), [reinforcement-learning](), [reward](), [rl](), [sampling](), [training](), [transformation]() ### AI overview The paper derives closed-form optimal policies for KL-regularized pass@k and max@k, showing that these objectives shift learning pressure according to prompt difficulty: they reduce sharpening on easier prompts and prioritize harder ones. It shows standard @1 objectives can reach the same policies through prompt-adaptive regularization or reward transformation, then introduces KAPPO to adapt existing @1 policy optimization methods after a one-time reference calibration without otherwise changing their training procedures. Experiments on mathematical reasoning and code generation report direct control over sharpening; the authors also report that single-completion training matches multi-completion @k optimizers at a fraction of the cost. ### Source excerpt arXiv:2611.03146v1 Announce Type: new Abstract: LLM post-training via reinforcement learning has shown consistent improvements in accuracy at the cost of reducing behavioral coverage - a phenomenon known as sharpening that can stall progress on difficult tasks. Repeated sampling objectives such as $\mathrm{pass@k}$ and $\mathrm{max@k}$ are known to mitigate sharpening, but the mechanism behind it remains poorly understood. Moreover, existing approaches for $\mathrm{@k}$ optimization use dedicated algorithms, and no general framework exists for adapting established policy optimization methods from $\mathrm{@}1$ to $\mathrm{@k}$. We derive the optimal policies for KL-regularized $\mathrm{pass@k}$ and $\mathrm{max@k}$ objectives in closed form. Our explicit solutions reveal that $\mathrm{@k}$ optimization naturally redistributes learning pressure based on prompt difficulty, directing it toward harder problems. Crucially, we show that standard $\mathrm{@}1$ objectives achieve these exact same optimal policies through prompt-adaptive regularization or reward transformation. Ultimately, this yields an optimizer-independent explanation for why $\mathrm{@k}$ objectives reduce sharpening on easy tasks while prioritizing success on harder ones. Building on these insights, we introduce $\mathrm{@k}$ Adaptive Pressure for Policy Optimization (KAPPO). Following a one-time reference calibration, KAPPO seamlessly adapts existing $\mathrm{@}1$ methods to $\mathrm{@k}$ via prompt-adaptive regularization or reward transformation, without otherwise altering their training procedures. Experiments with several RL methods on mathematical reasoning and code generation demonstrate that KAPPO provides direct control over sharpening. Crucially, we demonstrate that $\mathrm{@k}$ pressure can be fully realized with single completion training, matching the performance of multi-completion $\mathrm{@k}$ optimizers at a fraction of the cost. ## Benign population landscapes do not transfer to empirical risk minimization in nonsmooth optimization DevFeed: [Benign population landscapes do not transfer to empirical risk minimization in nonsmooth optimization]() Original publisher: [Read original article]() Author: Penghao Wang, Mateo D\\'iaz, Liwei Jiang Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Operations research and optimization](), [Data Science]() Tags: [arxiv](), [cross](), [cs-lg](), [data-science](), [gaussian](), [global](), [guarantees](), [local](), [lower-bound](), [math-oc](), [measurements](), [objectives](), [optimization](), [risk](), [sharp]() ### AI overview In robust phase retrieval with Gaussian measurements, benign geometry of the population objective does not ensure benign geometry for the empirical objective. The authors report that, despite the population objective being benign, the empirical objective has sharp spurious local minima with probability bounded away from zero as sample size tends to infinity; this lower bound is independent of dimension. The result challenges the population-to-empirical transfer strategy used for smooth optimization, where concentration of gradients and Hessians is used to transfer landscape properties, in the weakly convex nonsmooth setting. ### Source excerpt arXiv:2611.01384v1 Announce Type: cross Abstract: Benign landscapes provide a geometric explanation for the success of iterative methods on nonconvex problems arising in data science. The landscape is benign if every local minimizer is global and every other critical point is a strict saddle. For methods that avoid strict saddles, convergence to a critical point therefore implies convergence to a global minimizer. For smooth stochastic optimization, a common strategy for establishing such landscapes proceeds in two steps: $(i)$ establish benign geometry for the population objective---which is often easier to analyze---and $(ii)$ transfer this geometry to the empirical objective using concentration of gradients and Hessians. Recent work has extended saddle-point avoidance guarantees to weakly convex nonsmooth optimization through the notion of active strict saddles. We show, however, that benign population geometry need not transfer to empirical objectives in this setting. Specifically, for robust phase retrieval with Gaussian measurements---a well-studied and highly structured problem---the population objective is benign. Nevertheless, the empirical objective has sharp spurious local minima with probability bounded away from zero as the sample size tends to infinity. This asymptotic probability lower bound is independent of the dimension. ## OmniMechanism Design for Human-AI Collaboration with Impressionable Minds DevFeed: [OmniMechanism Design for Human-AI Collaboration with Impressionable Minds]() Original publisher: [Read original article]() Author: Tianwei Gao, Yifei Pang, Lunjia Hu, Kate Donahue, Lihua Lei, Vahid Tarokh, Zhun Deng Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Artificial Intelligence](), [Interactive AI Systems]() Tags: [ai](), [ai-collaboration](), [ai-systems](), [bandit](), [boosting](), [collaboration](), [cs-ai](), [cs-hc](), [cs-lg](), [decision-making](), [deployment](), [design](), [efficiency](), [evaluation](), [framework](), [healthcare](), [human-ai-collaboration](), [influence](), [input](), [novel](), [objectives](), [optimization](), [policies]() ### AI overview The paper proposes collaborative omnimechanism design to account for human judgments changing with the presentation of AI predictions and for deployment-time changes in the cost of eliciting human input. Its method uses a contextual-bandit variant of boosting for outcome indistinguishability, reducing a calibration-type condition to weighted calibration to improve time efficiency and address high-dimensional settings. A unified optimization scheme is intended to adapt decisions across query costs and evaluation objectives without retraining. In three human-AI collaboration settings, the authors report that resulting policies perform comparably to or better than the evaluated baselines. ### Source excerpt arXiv:2611.00441v1 Announce Type: cross Abstract: As AI systems are increasingly deployed in high-stakes domains such as healthcare, effectively integrating human expertise with AI outputs is critical for reliable decision-making. A substantial body of theoretical work studies optimal mechanisms for human-AI collaboration (HAC), focusing on when to involve humans and how to combine their expertise with AI predictions. However, these approaches largely overlook a key factor highlighted by empirical evidence: human judgments are impressionable and can vary depending on how AI outputs are presented. For example, when AI systems defer decisions to humans, revealing or withholding the AI prediction can significantly influence human judgments. Moreover, existing work neglects deployment-time variability, where conditions, particularly the cost of eliciting human input, may differ substantially from those observed during training. To address these challenges, we introduce a rigorous and theoretically grounded framework for human-AI collaboration, termed collaborative omnimechanism design, based on a novel and computationally efficient contextual bandit variant of boosting for outcome indistinguishability. In particular, we reduce a calibration-type outcome indistinguishability condition to weighted calibration, providing a novel time-efficiency result, bypassing the high-dimension curse induced by complicated collaboration methods and outcome space. Our framework provides a principled characterization of human impressionability and introduces a unified optimization scheme that enables robust and adaptable decision-making across diverse query costs and evaluation objectives without retraining. We evaluate the framework in three HAC settings, examining adaptation to query costs and prediction objectives without retraining, and show that its induced policies perform comparably to or better than the evaluated baselines. ## Surrogate Losses for Fairness-Aware Classification: Fisher Consistency and Calibration DevFeed: [Surrogate Losses for Fairness-Aware Classification: Fisher Consistency and Calibration]() Original publisher: [Read original article]() Author: Changcheng Song, Yi Yang, Xiangyu Chang, Xiaoling Lu, Yufeng Liu Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [responsible-ai](), [Machine learning](), [text-classification]() Tags: [application](), [characterization](), [classification](), [consistency](), [cs-lg](), [experiments](), [fairness](), [loss-functions](), [objectives](), [optimization](), [parameter](), [stat-ml]() ### AI overview The paper studies when surrogate-loss optimization yields decisions that are optimal under fairness-aware risk for demographic parity and equal opportunity, in both attribute-blind and attribute-aware settings. It defines fairness-aware Fisher consistency and characterizes it through derivative conditions, showing that consistency depends on how individual losses interact within the fairness-regularized objective. A calibration inequality connects excess surrogate risk to excess fairness-aware risk. Experiments examine how violations of the consistency conditions affect risk and how the regularization parameter affects whether optimization is well-posed. In a consumer credit default application, the abstract reports that appropriately calibrated fairness regularization substantially reduced demographic disparities while maintaining predictive performance. ### Source excerpt arXiv:2611.01613v1 Announce Type: cross Abstract: Fairness-aware classification seeks to balance predictive accuracy and fairness across demographic groups. Although surrogate losses facilitate the optimization of fairness-aware objectives, it remains unclear when minimizing these losses yields optimal decisions under the original fairness-aware risk. This paper investigates the statistical validity of surrogate-based fairness-aware classification under demographic parity and equal opportunity, considering both attribute-blind and attribute-aware settings. We introduce fairness-aware Fisher consistency and characterize it through conditions on the derivatives of the surrogate losses. The characterization shows that consistency depends not only on the individual loss functions but also on their interaction within the fairness-regularized objective. We further establish a calibration inequality that relates the excess fairness-aware risk to the excess surrogate risk, providing a quantitative connection between surrogate optimization and the original classification objective. Numerical experiments illustrate how deviations from the consistency conditions affect fairness-aware excess risk and how the regularization parameter influences the well-posedness of surrogate optimization. An application to consumer credit default prediction demonstrates that appropriately calibrated fairness regularization can substantially reduce demographic disparities while maintaining predictive performance. ## RaMMOBO: Random Max-Margin Multi-Objective Bayesian Optimization DevFeed: [RaMMOBO: Random Max-Margin Multi-Objective Bayesian Optimization]() Original publisher: [Read original article]() Author: Thomas Wedenig, Paul Plantosar, Robert Peharz Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Operations research and optimization](), [Hyperparameter optimization]() Tags: [acquisition](), [bayesian](), [benchmarks](), [box](), [cs-lg](), [improvement](), [linear](), [lower-bounds](), [metric](), [objectives](), [optimization](), [orders](), [probabilistic](), [problems](), [random](), [real-world](), [region](), [set](), [space](), [synthetic](), [utility]() ### AI overview RaMMOBO reduces the cost of multi-objective Bayesian optimization as the number of objectives grows. Standard expected hypervolume improvement (EHVI) scales exponentially with objective count; RaMMOBO instead uses the volume of a max-margin box within the hypervolume-improvement region, averaged over random objective-space reweightings, to lower-bound EHVI with linear evaluation cost. The method is described as strictly Pareto-monotone on hypervolume-improving outcomes, compatible with noisy observations and different probabilistic surrogates, and suitable for gradient-based acquisition optimization. On synthetic and real-world benchmarks, it reportedly matches or outperforms state-of-the-art methods, with larger gains at higher objective counts; acquisition optimization was up to 100 times faster than qLogNEHVI. ### Source excerpt arXiv:2611.00312v1 Announce Type: new Abstract: Multi-objective Bayesian optimization (MOBO) maximizes $M$ expensive black-box objectives with the goal of approximating the Pareto set, whose objective vectors cannot be improved in one component without sacrificing another. A common quality measure is hypervolume, the volume in objective space dominated by these vectors relative to a reference point. Expected hypervolume improvement (EHVI) directly targets this metric and is one-step Bayes-optimal, but its standard computation scales exponentially with $M$, making it prohibitively expensive for many-objective problems. We introduce random max-margin multi-objective Bayesian optimization (RaMMOBO), whose acquisition function lower-bounds EHVI while requiring only linear cost in $M$ to evaluate. RaMMOBO measures the volume of a max-margin box contained in a candidate's hypervolume-improvement region and averages over random reweightings of the objective space, yielding a utility that is strictly Pareto-monotone on hypervolume-improving outcomes. It naturally handles noisy observations and different probabilistic surrogates and supports efficient gradient-based optimization. Across synthetic and real-world benchmarks, RaMMOBO matches or outperforms state-of-the-art methods, with largest gains at higher $M$, with acquisition optimization up to two orders of magnitude faster than qLogNEHVI. Our implementation is available at https://anonymous.4open.science/r/rammobo. ## Unimodal Augmentation-based SSL Objectives Induce Cross-Modal Alignment DevFeed: [Unimodal Augmentation-based SSL Objectives Induce Cross-Modal Alignment]() Original publisher: [Read original article]() Author: Zohar Mashiah, Nir Ben-Ari, Ronen Talmon, Uri Shaham Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [multimodal-ai](), [data augmentation]() Tags: [alignment](), [audio](), [cross](), [cs-lg](), [machine-learning](), [modal](), [objectives](), [ssl](), [stat-ml](), [text](), [vision]() ### AI overview The paper argues that augmentation-based self-supervised objectives can make independently trained text, vision, and audio encoders align across modalities. It reports that low unimodal loss forces high cross-modal alignment, while high unimodal loss prevents it, and theoretically connects InfoNCE loss with local CKA. The authors report supporting evidence from controlled settings and 53 unconstrained frozen unimodal encoders. The supplied text is an abstract, so it does not provide details of the experiments or limitations. ### Source excerpt arXiv:2611.02917v1 Announce Type: new Abstract: Recent works have observed implicit cross-modal alignment between independently trained unimodal models. However, the mechanisms underlying the emergence of this phenomenon remain unresolved. In this paper, we demonstrate that unimodal augmentation-based self-supervised objectives implicitly induce cross-modal alignment. We show that low unimodal losses force high cross-modal alignment, whereas a high unimodal loss prevents such alignment. We theoretically connect the widely used InfoNCE loss to local CKA, the popular measure of cross-modal alignment. We further provide extensive empirical evidence supporting this claim, both in controlled settings and across 53 unconstrained frozen text, vision, and audio unimodal encoders. In both cases, unimodal augmentation-based objectives control cross-modal alignment. Overall, our results provide a theoretical and empirical foundation for understanding the emergence of implicit cross-modal alignment by establishing a direct connection between unimodal training objectives and cross-modal structure. ## Beyond Scarcity: When Encoder Adaptation Is Worthwhile DevFeed: [Beyond Scarcity: When Encoder Adaptation Is Worthwhile]() Original publisher: [Read original article]() Author: Zexing Zhang, Jichao Li, Tianyang Lei, Yiyin Zhang, Kewei Yang Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Neural Network](), [ImageNet](), [benchmark overfitting machine learning]() Tags: [benchmarks](), [cs-cv](), [cs-lg](), [evaluation](), [framework](), [incremental](), [inference](), [machine-learning](), [metrics](), [network](), [objectives](), [optimization](), [recognition](), [support]() ### AI overview In long-tailed recognition, rare-class support size alone does not determine whether encoder updates help after classifier correction: the benefit depends on how heavily rare-class examples shape the representation. A controlled framework compares matched classifier-only and full-network continuations from a common corrected model while separating unique example support from representation allocation. The same low-support regime shifts from classifier-only training being adequate to encoder adaptation yielding a practically meaningful benefit as representation allocation becomes more imbalanced. The reported transition persists across classifier corrections, representation objectives, normalization and augmentation controls, and ImageNet-scale evaluation, while frequent-class accuracy is preserved. Simultaneous confidence bounds distinguish the regimes under a prespecified one-percentage-point operating margin. The abstract does not provide specific implementation details or numerical results beyond that margin. ### Source excerpt arXiv:2610.41213v1 Announce Type: new Abstract: Long-tailed recognition often learns a shared representation before correcting the classifier for class imbalance. Once the classifier has been rebalanced, however, final accuracy does not reveal whether further encoder adaptation remains worthwhile. Standard imbalance benchmarks obscure this question because they jointly vary the number of distinct examples available to rare classes and the frequency with which those examples shape the representation. We introduce a controlled evaluation framework that separates unique support from representation allocation and estimates the incremental value of encoder updates through matched classifier-only and full-network continuations from a common corrected model. The framework combines stage-specific controls, multiple representation and adaptation objectives, whole-distribution metrics, and hierarchical inference over class assignments, support reconstructions, and optimization runs. Across the controlled study, the same low-support regime moves from one in which classifier-only training is adequate to one in which encoder adaptation yields a practically meaningful benefit solely as representation allocation becomes more imbalanced. The transition persists across classifier corrections, representation objectives, normalization and augmentation controls, and ImageNet-scale evaluation, while preserving frequent-class accuracy. Under the prespecified one-percentage-point operating margin, the two regimes are separated by simultaneous confidence bounds. These results show that benchmark construction can alter a post-correction training decision and that support size alone does not characterize the remaining value of representation adaptation. ## Specification-Aware Distribution Shaping for Robotics Foundation Models DevFeed: [Specification-Aware Distribution Shaping for Robotics Foundation Models]() Original publisher: [Read original article]() Author: Sad{\\i}k Bera Y\\"uksel, Derya Aksaray Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [AI Foundation Models](), [Artificial Intelligence](), [Robotics Simulation]() Tags: [constraint](), [cs-ai](), [cs-ro](), [data-driven](), [deployment](), [distribution](), [environments](), [execution](), [formal](), [foundation-models](), [guarantees](), [horizon](), [language](), [logic](), [model](), [natural-language](), [objectives](), [operational](), [optimization](), [reasoning](), [robotics](), [specification]() ### AI overview The framework enforces Signal Temporal Logic (STL) constraints during execution of a pretrained robotics foundation model without changing its parameters. At each decision step, it uses forward dynamics over the remaining horizon to find a minimally modified action distribution that meets a hard STL feasibility constraint. The targeted requirements include time-bounded goal visits, sequential objectives, and persistent safety conditions. The abstract reports simulation validation across multiple environments and complex specifications, but gives no quantitative results or details about the constraints' scope. ### Source excerpt arXiv:2603.17969v2 Announce Type: replace-cross Abstract: Robotics foundation models have demonstrated strong capabilities in executing natural language instructions across diverse tasks and environments. However, they remain largely data-driven and lack formal guarantees on constraint satisfaction during deployment. In practice, robots often need to comply with operational constraints involving rich spatio-temporal requirements such as time-bounded goal visits, sequential objectives, and persistent safety conditions. In this work, we propose a specification-aware action distribution optimization framework that enforces a broad class of Signal Temporal Logic (STL) constraints during execution of a pretrained robotics foundation model without modifying its parameters. At each decision step, the method computes a minimally modified action distribution that satisfies a hard STL feasibility constraint by reasoning over the remaining horizon using forward dynamics propagation. We validate the proposed framework in simulation using state-of-the-art robotics foundation models across multiple environments and complex specifications. ## Self-Building Labs DevFeed: [Self-Building Labs]() Original publisher: [Read original article]() Author: Jiaru Bai, Yuchi Zhao, Han Hao, Elia Savino, Miroslav Bogdanovic, Derk H. W. ten Klooster, Simone Pilon, Timothy No\\"el, Yang Cao, Varinia Bernales, Al\\'an Aspuru-Guzik Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [AI for science](), [Loop Engineering](), [lerobot]() Tags: [agents](), [ai](), [automation](), [building](), [cs-ai](), [cs-ma](), [cs-ro](), [designs](), [digital-twin](), [discovery](), [engineering](), [execution](), [experiment](), [flexibility](), [flexible](), [laboratory](), [labs](), [layout](), [objectives](), [optimization](), [programming](), [quantum](), [real-world]() ### AI overview Self-building laboratories (SBLs) aim to remove the manual engineering bottleneck that limits existing self-driving labs to procedures they were built to run. Given a high-level scientific goal, SBL starts from an empty room or an existing lab and builds missing capabilities, including the layout, digital twin, robot skills, and experimental workflows. The authors report operation across different robots and setups, from a single workstation to multi-robot laboratories, and validate the end-to-end system with a real-world quantum-dot synthesis experiment. The stated goal is to shift researchers from configuring automation to specifying scientific objectives. ### Source excerpt arXiv:2610.23826v2 Announce Type: replace Abstract: Self-driving laboratories (SDLs) accelerate scientific discovery by automating experimental execution and optimization. AI agents can increasingly propose new experiments, but existing SDLs can only run the procedures they were built for. Adapting them to a new experiment still requires extensive manual engineering, which makes laboratory flexibility the bottleneck for scaling discovery. We introduce the self-building laboratory (SBL), which designs, configures, and operates its own laboratory setup given a high-level scientific goal. SBL starts from whatever already exists, from an empty room to a working SDL, and builds the missing capabilities: the lab layout, its digital twin, robot skills, and experimental workflows. We demonstrate the system working across different robots and setups, scaling from a single workstation to laboratories with several robots working together. We validate SBL's end-to-end capabilities through a real-world quantum-dot synthesis experiment. By shifting human researchers from manually configuring and programming device automation to specifying scientific objectives, SBL makes scientific automation more accessible, flexible, and scalable. Project website: https://self-building-labs.github.io/ ## How to Score Experts for One-Shot MoE Expert Pruning: A Systematic Formulation and Selection Principle DevFeed: [How to Score Experts for One-Shot MoE Expert Pruning: A Systematic Formulation and Selection Principle]() Original publisher: [Read original article]() Author: Zongfang Liu, Jinghui Zhang, Zijian Ma, Guangyi Chen, Xin Yuan Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Mixture of Experts (MoE)](), [Post-training optimization]() Tags: [activation](), [benchmarks](), [cs-lg](), [deployment](), [experts](), [gate](), [language-models](), [machine-learning](), [memory](), [mixture-of-experts](), [mixture-of-experts-moe](), [moe](), [objectives](), [performance](), [routing](), [token](), [weight]() ### AI overview The paper formulates one-shot MoE expert pruning around routing frequency, gate weighting, and activation strength, then uses these factors to guide criterion choice. For task-agnostic pruning, it recommends routed-token-averaged activation criteria that omit gate weights; task-specific pruning may benefit from retaining routing-frequency and gate-weight information. It introduces two task-agnostic criteria, Mean Activation Norm (MAN) and Mean Squared Activation Norm (MSAN). Across four MoE models and 16 benchmarks, they rank in the top two on average and improve average performance by 4.0 points, up to 8.8 points, over the strongest baseline. ### Source excerpt arXiv:2606.15716v2 Announce Type: replace Abstract: Mixture-of-Experts (MoE) language models reduce per-token computation through sparse expert activation, yet deployment still requires storing the full expert pool, making one-shot expert pruning a practical approach for reducing memory usage. Although effective, existing criteria are largely heuristic, and no single criterion is universally optimal. Thus, establishing a principle for selecting pruning criteria suited to different deployment objectives remains an important yet largely underexplored problem in one-shot expert pruning. To this end, we introduce a systematic formulation for one-shot MoE expert pruning organized around three factors: routing frequency, gate weighting, and activation strength. The formulation yields a criteria selection principle: task-agnostic pruning should favor routed-token-averaged, gate-free activation-based criteria, whereas task-specific pruning can benefit from retaining routing-frequency and gate-weight information. Beyond this principle, the formulation also provides a systematic view of existing heuristic criteria and gives rise to two new task-agnostic criteria, Mean Activation Norm (MAN) and Mean Squared Activation Norm (MSAN). Across four representative MoE models and 16 diverse benchmarks, MAN and MSAN are consistently strong in the task-agnostic setting, obtain the top-two average ranks, and improve average performance by 4.0 points on average and up to 8.8 points over the strongest baseline. [Next page]()