Search arXiv⌕ Search

SEARCH · Search arXiv

Search Search arXiv

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

5,200 records · Page 3Linked to original sources
arXiv2026

DWFF-Net: A Multi-Scale Farmland System Habitat Identification Method with Adaptive Dynamic Weight Feature Fusion

To address insufficient accuracy in multi-scale segmentation for agricultural habitat recognition, this study proposes a Dynamic Weighted Feature Fusion Network (DWFF-Net). Its encoder uses frozen DINOv3 to extract basic features and introduces a data-level adaptive dynamic weighting strategy based on relationships between image categories and feature maps. The decoder employs a dynamic weight calculation network for deep fusion of multi-level features and a hybrid loss for optimization. Statistical analysis shows that weight entropy tends to decrease as habitat category count increases, indicating adaptive adjustment of fusion strategy according to scene complexity. Experiments on a previously constructed agricultural habitat dataset validate DWFF-Net. Ablations yield mIoU 0.6979 and mF1 0.8049, exceeding the Static Weighted Feature Fusion Network by 1.82% and 1.54%, respectively, confirming that dynamic weighting improves multi-level feature utilization. Compared with U-Net, DeepLabv3+, SegFormer, and DPT, DWFF-Net improves mIoU by 16.32%, 6.49%, 4.17%, and 3.18%, respectively. For tiny features like scattered trees, IoU reaches 0.2707, outperforming those models by 99.85%, 11.45%, 22.24%, and 19.32%, verifying effectiveness in tiny habitat segmentation. This framework enables low-cost, high-precision habitat mapping and supports refined monitoring in agricultural landscapes.

cs.CV↗
arXiv2026

Redundancy rules for MaxSAT

The concept of redundancy in SAT leads to more expressive and powerful proof search techniques, e.g., able to express various inprocessing techniques, and originates interesting hierarchies of proof systems [Heule et$.$al'20, Buss-Thapen'19]. Redundancy has also been integrated in MaxSAT [Ihalainen et$.$al'22, Berg et$.$al'23, Bonacina et$.$al'24]. In this paper, we define a structured hierarchy of redundancy proof systems for MaxSAT, with the goal of studying its proof complexity. We obtain MaxSAT variants of proof systems such as SPR, PR, SR, and others, previously defined for SAT. All our rules are polynomially checkable, unlike [Ihalainen et$.$al'22]. Moreover, they are simpler and weaker than [Berg et$.$al'23], and possibly amenable to lower bounds. This work also complements the approach of [Bonacina et$.$al'24]. Their proof systems use different rule sets for soft and hard clauses, while here we propose a system using only hard clauses and blocking variables. This is easier to integrate with current solvers and proof checkers. We discuss the strength of the systems introduced, we show some limitations of them, and we give a short cost-SR proof that any assignment for the weak pigeonhole principle $PHP^{m}_{n}$ falsifies at least $m-n$ clauses. We conclude by discussing the integration of our rules with the MaxSAT resolution proof system, which is a commonly studied proof system for MaxSAT.

cs.LO↗
arXiv2026

Freeze, Diffuse, Decode: Task-Aware Adaptation of Transformer Embeddings for Antimicrobial Peptide Design

Pretrained transformers provide rich, general-purpose embeddings, which are transferred to downstream tasks. However, current transfer strategies: fine-tuning and probing, either distort the pretrained geometric structure of the embeddings or lack sufficient expressivity to capture task-relevant signals. These issues become even more pronounced when supervised data are scarce. Here, we introduce Freeze, Diffuse, Decode (FDD), a novel diffusion-based framework that adapts pre-trained embeddings to downstream tasks while preserving their underlying geometric structure. FDD propagates supervised signal along the intrinsic manifold of frozen embeddings, enabling a geometry-aware adaptation of the embedding space. Applied to antimicrobial peptide design, FDD yields low-dimensional, predictive, and interpretable representations that support property prediction, retrieval, and latent-space interpolation.

cs.LG↗
arXiv2026

An Energy-Based Mechanism for Compositional Behavior

Flexible intelligence relies on the ability to reuse previously acquired behaviors and combine them differently as circumstances change. In biological and artificial systems, this ability is often attributed to gating mechanisms that determine how much each available behavior should contribute at a given time. Yet these gating rules, the dynamics that compute them, and the neural circuits that may implement them are usually introduced separately, leaving unclear whether they reflect a common underlying principle. Here, we show that they can all be derived from a single variational principle for behavioral composition. The resulting mechanism naturally gives rise to softmax gating, evolves as an energy-based dynamical system with guaranteed convergence, and admits a recurrent neural network instantiation featuring context-dependent and local interactions. Across collective behavior, human decision-making, and layered control, the same mechanism reproduces characteristic behavioral patterns, provides interpretable accounts of how different behaviors are combined, and matches or outperforms established approaches. These results provide a unified account of how behavioral composition can emerge from a common principle, with implications for understanding flexible behavior in natural systems and for designing artificial agents that can adapt by recombining existing capabilities.

math.OC↗
arXiv2026

Stepwise Think-Critique: Interleaved Reasoning and Self-Critique in a Single LLM

Human beings solve complex problems through critical thinking, where reasoning and evaluation are intertwined to converge toward correct solutions. However, most existing large language models (LLMs) treat the reasoning and verification as separate processes: they either generate reasoning without explicit self-checking or rely on external verifiers to detect errors post hoc. The former lacks immediate feedback, while the latter increases system complexity and hinders synchronized learning. Motivated by human critical thinking, we propose Stepwise Think-Critique (STC), an end-to-end trainable framework in which a single LLM emits a structured, step-level critique inline with each reasoning step. STC is trained with reinforcement learning that complements reasoning rewards with a critique-consistency reward derived from final-answer correctness, jointly optimizing reasoning correctness and critique reliability. On five mathematical reasoning benchmarks, STC improves Pass@1 by 7.2% over the 1.5B base model and attains 67.4% step-level critique F1, surpassing seven external process reward models evaluated at their per-dataset oracle thresholds---a step toward LLMs with built-in critical thinking.

cs.AI↗
arXiv2026

On Cost-Aware Designs for Sequential Hypothesis Testing

We introduce Cost-Aware (CA) Sequential Hypothesis Testing (CASHT), in which an active decision-maker selects sensing actions with differing, random costs to identify the true hypothesis under an average-error constraint $δ$ while minimizing the expected total cost rather than the number of samples. For fixed costs, we prove that the optimal expected total cost scales as $Θ(\log(1/δ))$, and is achievable by Multihypothesis Sequential Probability Ratio Test-based procedures. We show that the CA design principle is to maximize the ratio of expected information gain to expected cost under the policy-induced action distribution. Guided by this principle, we adapt two classic policies to the CA setting and establish their asymptotic optimality. We then treat random costs under two revelation models: ex-post, where costs are disclosed only after a sample is obtained, and the cost-error tradeoff coincides with the fixed-cost case, and ex-ante, where costs accrue before acquisition, and the decision maker may cancel an action mid-operation. For the ex-ante model, we characterize when cancellation lowers the total cost and analyze several cost distributions in detail. Simulations confirm our findings that the CA variants consistently reduce total cost relative to their classical counterparts, and when action cancellation helps or hurts.

cs.IT↗
arXiv2026

FitControler: Toward Fit-Aware Virtual Try-On

Realistic virtual try-on (VTON) concerns not only faithful rendering of garment details but also coordination of the style. Prior art typically pursues the former, but neglects a key factor that shapes the holistic style -- garment fit. Garment fit delineates how a garment aligns with the body of a wearer and is a fundamental element in fashion design. In this work, we introduce fit-aware VTON and present FitControler, a learnable plug-in that can seamlessly integrate into modern VTON models to enable customized fit control. To achieve this, we highlight two challenges: i) how to delineate layouts of different fits and ii) how to render the garment that matches the layout. FitControler first features a fit-aware layout generator to redraw the body-garment layout conditioned on a set of delicately processed garment-agnostic representations, and a multi-scale fit injector is then used to deliver layout cues to enable layout-driven VTON. In particular, we build a fit-aware VTON dataset termed Fit4Men, including 13,000 body-garment pairs of different fits, covering both tops and bottoms, and featuring varying camera distances and body poses. Two fit consistency metrics are also introduced to assess the fitness of generations. Extensive experiments show that FitControler can work with various VTON models and achieve accurate fit control. Code and data will be released.

cs.CV↗
arXiv2026

What Drives Success in Physical Planning with Joint-Embedding Predictive World Models?

A long-standing challenge in AI is to develop agents capable of solving a wide range of physical tasks and generalizing to new, unseen tasks and environments. A popular recent approach involves training a world model from state-action trajectories and subsequently use it with a planning algorithm to solve new tasks. Planning is commonly performed in the input space, but a recent family of methods has introduced planning algorithms that optimize in the learned representation space of the world model, with the promise that abstracting irrelevant details yields more efficient planning. In this work, we characterize models from this family as JEPA-WMs and investigate the technical choices that make algorithms from this class work. We propose a comprehensive study of several key components with the objective of finding the optimal approach within the family. We conducted experiments using both simulated environments and real-world robotic data, and studied how the model architecture, the training objective, and the planning algorithm affect planning success. We combine our findings to propose a model that outperforms two established baselines, DINO-WM and V-JEPA-2-AC, in both navigation and manipulation tasks. Code, data and checkpoints are available at https://github.com/facebookresearch/jepa-wms.

cs.AI↗
arXiv2026

Constrained dynamics for searching saddle points on embedded Riemannian submanifolds of Euclidean space

Finding constrained saddle points on embedded Riemannian submanifolds of Euclidean space is significant for analyzing energy landscapes arising in physics and chemistry. Existing works exploit explicit global/local regular level-set representations of manifolds, which may be unavailable or computationally inconvenient for manifolds represented through, e.g., projectors, factorizations, or rank constraints. In this paper, we develop a constrained saddle dynamic based on embedded-submanifold geometric primitives, completely avoiding the use of explicit representations. In particular, our dynamic is formulated compactly on the Grassmann bundle of the tangent bundle. By analyzing the Grassmann bundle geometry, we rigorously establish the local linear stability of the dynamic and the local linear convergence of the resulting algorithms. Remarkably, our analysis provides the first iterate convergence result for discretized algorithms to saddle points of prescribed indices in embedded-submanifold settings. Moreover, by virtue of the Grassmann bundle formulation, we remove unnecessary nondegeneracy assumptions on the eigenvalues of the Riemannian Hessian that are present in existing works. We also point out that locating saddle points can be more ill-conditioned than finding local minimizers, and requires using nonredundant parametrizations. Finally, numerical experiments on linear eigenvalue problems and electronic excited-state calculations showcase the effectiveness of the proposed algorithms and corroborate the established local theory.

math.NA↗
arXiv2026

Exposía: Teaching and Assessment of Academic Writing Skills for Research Project Proposals and Peer Feedback

We present Exposía, the first public dataset that connects writing and feedback in higher education, enabling research on educationally grounded computational approaches to teaching and evaluating academic writing. Exposía includes student research project proposals and peer and instructor feedback consisting of comments and free-text reviews. The dataset was collected in the "Introduction to Scientific Work" course of the Computer Science. Exposía reflects the multi-stage nature of the academic writing process that includes drafting, receiving feedback, and revising the writing based on the feedback received. Both the project proposals and peer feedback are accompanied by human assessment scores based on a fine-grained, pedagogically-grounded schema for writing and feedback assessment that we develop. We use Exposía to benchmark state-of-the-art large language models (LLMs) on two tasks: automated scoring of (1) the proposals and (2) the student reviews. We find that the two tasks are best served by different LLMs. Furthermore, closed-source models consistently outperform open-weight models, motivating further research on improving the performance of open-weight models preferred in classroom settings. Finally, we establish that a prompting strategy that scores multiple aspects of the writing together is the most effective, paving the way for more effective classroom deployment of modern LLMs.

cs.CL↗
arXiv2026

LLMs Can't Play Hangman: On the Necessity of a Private Working Memory for Language Agents

As LLMs move from text completion toward autonomous agents, they remain constrained by the standard chat interface, which lacks private working memory. This raises a fundamental question: can agents reliably perform interactive tasks that depend on hidden state? We define Private State Interactive Tasks (PSITs), which require agents to generate and maintain hidden information while producing public responses consistent with a fixed hidden state. We show theoretically that any agent restricted to the public conversation history cannot both keep the secret unresolved in the transcript and respond consistently with a fixed hidden state in PSITs, yielding an architectural impossibility theorem. To empirically validate this limitation, we introduce a self-consistency testing protocol that evaluates whether agents can maintain a hidden secret across forked dialogue branches. Standard chat-based LLMs and retrieval-based memory baselines fail this test regardless of scale, demonstrating that semantic retrieval does not enable true state maintenance. To address this, we propose a novel architecture incorporating an explicit private working memory; we demonstrate that this mechanism restores consistency with a fixed hidden state, establishing private state as a necessary component for PSIT-capable language agents. Our code is available at https://github.com/chandar-lab/Hangman

cs.CL↗
arXiv2026

Beyond Dialogue Time: Temporal Semantic Memory for Personalized LLM Agents

Memory enables Large Language Model (LLM) agents to perceive, store, and use information from past dialogues, which is essential for personalization. However, existing methods fail to properly model the temporal dimension of memory in two aspects: 1) Temporal inaccuracy: memories are organized by dialogue time rather than their actual occurrence time; 2) Temporal fragmentation: existing methods focus on point-wise memory, losing durative information that captures persistent states and evolving patterns. To address these limitations, we propose Temporal Semantic Memory (TSM), a memory framework that models semantic time for point-wise memory and supports the construction and utilization of durative memory. During memory construction, it first builds a semantic timeline rather than a dialogue one. Then, it consolidates temporally continuous and semantically related information into a durative memory. During memory utilization, it incorporates the query's temporal intent on the semantic timeline, enabling the retrieval of temporally appropriate durative memories and providing time-valid, duration-consistent context to support response generation. Experiments on LongMemEval and LoCoMo show that TSM consistently outperforms existing methods and achieves up to 12.2% absolute improvement in accuracy, demonstrating the effectiveness of the proposed method.

cs.AI↗
arXiv2026

Vision-Language Model for Accurate Crater Detection

The European Space Agency (ESA), driven by its ambitions on planned lunar missions with the Argonaut lander, has a profound interest in reliable crater detection, since craters pose a risk to safe lunar landings. This task is usually addressed with automated crater detection algorithms (CDA) based on deep learning techniques. It is non-trivial due to the vast amount of craters of various sizes and shapes, as well as challenging conditions such as varying illumination and rugged terrain. Therefore, we propose a deep-learning CDA based on the OWLv2 model, which is built on a Vision Transformer, that has proven highly effective in various computer vision tasks. For fine-tuning, we utilize a manually labeled dataset fom the IMPACT project, that provides crater annotations on high-resolution Lunar Reconnaissance Orbiter Camera Calibrated Data Record images. We insert trainable parameters using a parameter-efficient fine-tuning strategy with Low-Rank Adaptation, and optimize a combined loss function consisting of Complete Intersection over Union (CIoU) for localization and a contrastive loss for classification. We achieve satisfactory visual results, along with a maximum recall of 92.6% and a maximum precision of 71.4% on a test dataset from IMPACT. Our method achieves reliable crater detection across challenging lunar imaging conditions, paving the way for robust crater analysis in future lunar exploration.

cs.CV↗
arXiv2026

Beyond Transfer Accuracy: Mechanism-Guided Controlled Adaptation for Low-Resource Languages

Existing circuit discovery methods rely on templated tasks with clean counterfactuals, limiting their use on diverse natural text. We adapt Contextual Decomposition for Transformers (CD-T) for unstructured settings via label-balanced activation means and task-directional relevance scoring, enabling counterfactual-free circuit discovery. We leverage the discovered circuits for Circuit-Targeted Supervised Fine-Tuning (CT-SFT), restricting parameter updates to task-relevant heads and LayerNorm. Experiments on NusaX cross-lingual sentiment transfer show that CT-SFT is highly competitive for low-resource adaptation. While non-circuit sparse updates and full fine-tuning sometimes match target accuracy through capacity recruitment, CT-SFT most consistently avoids catastrophic forgetting, preserving source-language and related-task performance. Extensions to XNLI support the source-retention and intervention findings on a harder task and two model families, showing that circuit-targeted adaptation provides a more controlled, intervention-supported alternative to global fine-tuning.

cs.CL↗
arXiv2026

Learning and extrapolating scale-invariant processes

Machine Learning (ML) has deeply changed some fields recently, like Language and Vision and we may expect it to be relevant also to the analysis of of complex systems. Here we want to tackle the question of how and to which extent can one regress scale-free processes, i.e. processes displaying power law behavior, like earthquakes or avalanches? We are interested in predicting the large ones, i.e. rare events in the training set which therefore require extrapolation capabilities of the model. For this we consider two paradigmatic problems that are statistically self-similar. The first one is a 2-dimensional fractional Gaussian field obeying linear dynamics, self-similar by construction and amenable to exact analysis. The second one is the Abelian sandpile model, exhibiting self-organized criticality. The emerging paradigm of Geometric Deep Learning shows that including known symmetries into the model's architecture is key to success. Here one may hope to extrapolate only by leveraging scale invariance. This is however a peculiar symmetry, as it involves possibly non-trivial coarse-graining operations and anomalous scaling. We perform experiments on various existing architectures like U-net, Riesz network (scale invariant by construction), or our own proposals: a wavelet-decomposition based Graph Neural Network (with discrete scale symmetry), a Fourier embedding layer and a Fourier-Mellin Neural Operator. Based on these experiments and a complete characterization of the linear case, we identify the main issues relative to spectral biases and coarse-grained representations, and discuss how to alleviate them with the relevant inductive biases.

cond-mat.dis-nn↗
arXiv2026

CALIBURN: Self-Calibrated LLM Unlearning Alignment

LLM unlearning aims to remove the influence of undesirable knowledge from pretrained language models, which offers a practical mechanism for addressing safety and privacy concerns. Existing unlearning approaches, such as Gradient Ascent, are prone to catastrophic forgetting. Alignment-based approaches provide an alternative direction, yet their effectiveness is limited by the quality of the reference model. In realistic settings, both methods still require large retention datasets to preserve general knowledge. We propose a principled method that quantifies the target LLM's confidence in undesirable knowledge and uses it to calibrate the model's unlearning gradient updates more precisely. It enables fine-grained control over forgetting while better preserving model utility, thus reducing the dependence on retention data or prohibitive unlearning training data. Extensive evaluations on multiple benchmarks, including MUSE and WMDP, show that our method achieves effective unlearning and improves the trade-off between knowledge removal and utility preservation compared with state-of-the-art methods.

cs.CL↗
arXiv2026

Learning Query-Specific Rubrics from Human Preferences for DeepResearch Report Generation

Nowadays, developing reliable DeepResearch-style long-form report generation remains challenging, as training and evaluation lack verifiable reward signals. Accordingly, rubric-based evaluation has become a common practice. However, existing approaches either rely on coarse, pre-defined rubrics that lack sufficient granularity or depend on manually constructed query-specific rubrics that are costly and difficult to scale. In this paper, we propose a pipeline to train preference-grounded query-specific rubric generators tailored for DeepResearch report generation. We first construct a dataset of DeepResearch-style queries annotated with human preferences over paired reports, and train rubric generators via reinforcement learning with a hybrid reward combining preference consistency, format validity, and LLM-based rubric evaluation. We evaluate the resulting rubric generators in two stages. First, on a held-out human-preference test set, the learned rubrics discriminate preferred from rejected reports more effectively than generic, prompted, or SFT-trained rubric alternatives. Second, when used as reward signals to train DeepResearch systems, our rubric generators yield substantial performance gains under both a simple single-agent ReAct framework and a complex multi-agent workflow on the DeepResearch Bench.

cs.CL↗
arXiv2026

PoC-Gym: Towards More Reliable LLM-Assisted Proof-of-Concept Exploit Generation

Recently Large Language Models (LLMs) have been used in security-related tasks, including generating proof-of-concept (PoC) exploits. Several LLM-assisted approaches have been proposed; they typically generate PoCs from vulnerability descriptions and use additional guidance. But, such approaches are often ineffective because the signals-such as printed markers, generated files, or runtime side effects-that they use for validation may not imply that the vulnerability is triggered. Research for more reliable PoC generation is in need but yet remains challenging. We propose PoC-Gym, a pipeline for LLM-based PoC generation for Java security vulnerabilities. PoC-Gym uses both static and dynamic information, e.g., CVE-tailored prompts, static traces, and coverage-based feedback, and iteratively generates PoC candidates. Each candidate goes through a series of validations: whether the execution is complete, manifests a success signal, and reaches the sink of the target trace. We evaluate PoC-Gym using 20 Java CVEs. Across 338 runs, 116 candidates pass PoC-Gym's runtime validation and 65 candidates pass post-hoc validation against the ground-truth vulnerable locations, covering 12 of the 20 CVEs. On the 14-CVE overlap with FaultLine, the strongest PoC-Gym configuration is post-hoc valid for 8 CVEs, while FaultLine reports success for 5 CVEs under its original evaluation criterion. But, given the complexity of PoC generation, PoC-Gym also generates many runtime-valid but post-hoc-invalid PoCs. To better understand how to achieve more reliable PoC generation, we present an in-depth analysis of such PoCs and identify common sources of failures. We believe that our work provides insights for future research.

cs.SE↗