The paper introduces ThermoField, a framework that unifies thermal scene reconstruction and thermophysical parameter estimation via differentiable heat-transfer simulation. It uses neural fields to represent spatially varying properties like thermal diffusivity, constrained by scene geometry and physics, enabling joint reconstruction of geometry, estimation of diffusivity, and prediction of thermal evolution under unseen conditions.

Researchers propose adversarial decoys, independently optimized image patches that redirect attention away from adversarial regions, bypassing attention-based defenses in Vision Transformers. The approach decouples misclassification and defense evasion, is attack-agnostic, and preserves attack effectiveness. Experiments on ImageNet reveal a fundamental limitation of using attention magnitude as an indicator of adversarial relevance.

Researchers developed a low-cost method using UAV and crane-based photogrammetry to reconstruct deciduous trees in 3D for monitoring shoot elongation (primary growth). Achieving 5-6 mm accuracy and 92-98% completeness, the approach addresses a gap in climate change impact studies on tree growth.

GIRAF is a text-conditioned diffusion model for generating realistic full-body interactions with articulated objects. It addresses limitations of prior works by jointly reasoning about locomotion, contact, and articulation, using an object-centric representation, mixed-domain training, and contact-based augmentation, achieving strong generalization to unseen object configurations.

DreamCharacter-1 is a lightweight post-adaptation framework that calibrates pretrained 3D foundation models for high-fidelity, production-ready 3D character generation. It includes geometry post-training, texture post-training, and inference acceleration, consistently outperforming state-of-the-art methods.

Identifying faithfulness hallucinations in LLM-generated outputs remains challenging due to the scarcity of high-quality annotated data. This paper introduces Hallucination Self-Play (HSP), a framework where a detector and generator bootstrap each other. The detector is fine-tuned on human labels, then used as a reward model to train the generator via RLAIF to produce harder-to-detect hallucinations. The evolved generator's outputs further optimize the detector via rule-based RL. Experiments on RAGTruth and two model families show a small LLM can match or outperform advanced LLMs without external supervision.

A new study evaluates the reliability of Gemini models as audio judges for full-duplex voice agent conversations. Using 209 stereo sessions scored on 8 dimensions, Gemini 2.5 Flash shows high agreement with human raters on most dimensions, with cost savings of roughly two orders of magnitude. The paper also cautions that model swaps require re-validation on calibration data.

This paper identifies a failure mode called Positive-Credit Contamination in RL for LLMs, where low-probability erroneous tokens receive identical positive credit as plausible ones. The proposed TACO method computes a tail-risk score to calibrate credit assignment, outperforming GRPO baselines across three LLMs and eight benchmarks while improving training stability in long-horizon RL.

This paper proposes a multi-cluster boundary learning method using MiniLM embedding for out-of-scope (OOS) intent detection. It addresses the accuracy drop of traditional multi-class classification and the large parameter issue of LLM embeddings, achieving state-of-the-art performance on three public datasets.

Preprocessing-based debiasing methods in NLP, while reducing stereotypes for targeted groups, can cause unintended shifts that increase stereotyping or counter-stereotyping for other demographics, including unrelated categories. The study demonstrates these side effects across model families and preprocessing strategies, and argues for side-effect-aware mitigation practices.

This research proposes a cost-efficient human-LLM collaborative annotation framework to construct multilingual stereotype datasets. Applied to Spanish, it yields EspanStereo, covering multiple Spanish-speaking countries. Evaluations show significant variation in LLM stereotypical behavior across countries, highlighting the need for culturally grounded assessments.

This paper argues that Barenholtz's autogenerative theory of language enriches Harrisean integrationism by providing a structural mechanism for prospective openness, a computational correlate for semiotic continuity, and a theory of the archive. It offers insights for NLP and LLM design.

DeepSearch-Evolve is a self-distillation framework for web agents built on DeepSearch-World, a deterministic and verifiable environment. It contains 420K multi-hop QA tasks and supports cognitive behaviors like progress verification and failure recovery. Without teacher distillation, DeepSearch-World-9B achieves competitive results on BrowseComp, GAIA, and HotpotQA, showing that verifiable environments enable scalable self-evolution for long-horizon web agents.

This position paper reviews recent advances in AI for Mathematics (AI4Math), particularly LLM-driven theorem provers for formal proof generation. It argues that current systems are limited to well-defined problems and cannot handle open-ended frontier research like discovering new theorems. The authors advocate shifting from problem solvers to research agents capable of rigorous formal reasoning, and identify key limitations across datasets, relational structure, exploration, tools, and human-AI collaboration, outlining a roadmap for the future of AI4Math.

This paper compares LSTM with traditional machine learning models for Twitter sentiment analysis. The LSTM model achieved 90.98% training accuracy, 80% testing accuracy, and a ROC-AUC score of 0.92, outperforming other methods.

LiST (Lipschitz Scaling Training) is a novel training paradigm that automatically adjusts the global Lipschitz constant to achieve calibrated neural networks while balancing accuracy and robustness. It reveals a theoretical link between Lipschitz constraints and Temperature Scaling, using calibration as a principled criterion to select an optimal operating point. Experiments on CIFAR-10/100 and Tiny-ImageNet show competitive accuracy and robustness while remaining calibrated out of the box.

MetaNCA learns local update rules to self-organize neural network weights, enabling weight generation for diverse architectures without backpropagation and generalizing to unseen architectures.

Jet-Long introduces a tuning-free zero-shot method for extending LLM context windows by using dynamic bifocal RoPE, which adapts the rescaling factor to sequence length, achieving high efficiency and strong performance on multiple benchmarks.

SHIFT is a missingness-aware survival model that directly predicts from incomplete genomic inputs without test-time imputation, using masked self-attention and a feature-availability mask. It introduces variable-rate feature masking during training for robustness to heterogeneous missingness. Evaluated on glioblastoma and lung squamous cell carcinoma across multiple cohorts, SHIFT shows strong generalization and outperforms baselines and imputation-based methods, supporting missingness-aware modeling for multi-center survival prediction in precision oncology.

In block-sparse attention for long-context LMs, fixed top-k cutoff can drop crucial blocks when scores are tied. A value-of-information router doubles kept blocks for uncertain queries, achieving significant recall gains across models.

Omni-Sleep is a new sleep foundation model that leverages the physiological partition of the central nervous system (CNS) and autonomic nervous system (ANS) as a prior for topology-constrained representation learning. It learns structured representations through three objectives: intra-system consistency, inter-system synchronization, and latent-space masked temporal modeling. Pre-trained on over 100,000 hours of multi-center multimodal PSG data, Omni-Sleep outperforms strong baselines on sleep staging and multi-disease classification, demonstrating improved label efficiency, cross-dataset generalization, and robustness to missing modalities.

ReCoLoRA addresses catastrophic forgetting in continual fine-tuning by recursively consolidating low-rank adapters. It outperforms LoRA, PiSSA, AdaLoRA, and DoRA on three out of four 7-8B backbones on a six-task GLUE sequence while using fewer parameters.

Introduces Local Linear Transformer (LLT), a novel neural operator architecture combining linear global attention with local spatial mixing to address quadratic scaling and lack of local bias in standard attention for PDEs. Achieves competitive errors and 1.8-2.5x training speedup over Transolver on multiple PDE problems.

In chest X-ray classification, acceptable ranking performance can still miss rare-positive patients below threshold, especially within subgroups. Using a diagnostic ladder, group-tail weighting followed by tail-aware thresholding reduces tail FNR from 0.665 to 0.269, but residual missed rates remain high. Fairness depends on finding, subgroup, and threshold jointly.

This paper introduces MemExplainer, a method that explains Temporal Graph Network (TGN) predictions using a topology attribution tree and a memory backtracking tree. It accounts for the memory module, quantifies the influence of historical events on node memory vectors, and ensures the sum of event contributions equals model logits via Layer-wise Relevance Propagation (LRP). Experiments on nine datasets covering node property prediction, link prediction, and graph classification show faithful explanations outperforming state-of-the-art baselines. The paper is a Spotlight at ICML 2026.

This paper examines how AI architectures, from single LLMs to multi-agent agentic systems, can support straight-through underwriting by comparing three pipelines in a synthetic small commercial Business Owner Policies (BOPs) environment. The agentic RAG pipeline, combining retrieval, third-party data checks, and multi-step rule evaluation, outperforms others, especially in complex scenarios requiring multi-step reasoning and handling missing information.

Researchers propose a graph neural network approach for real-time hand gesture recognition using sEMG signals, achieving 99% accuracy and 48ms processing time on a Myoband with 8 subjects, outperforming state-of-the-art methods.

VectorizationLLM is a specialized large language model based on Google open-weight LLMs, designed to help students learn smart vectorization and related topics in MATLAB for the course CTEC 247 at New York Institute of Technology. It uses a RAG knowledge base and system prompts to provide explanations and examples without giving direct answers.

We present Infinity-Parser2, a large multimodal model that couples a controllable data-synthesis pipeline with multi-task reinforcement learning for end-to-end document parsing. It introduces Infinity-Doc2-5M, a 5-million-sample bilingual corpus, and a multi-task reward system for joint RL across eight objectives. Two variants are released: Flash (low-latency) and Pro (precision), with Pro achieving SOTA results on benchmarks.

A new paper introduces 'idiobionics' as an interdisciplinary field to systematically investigate privacy risks in intelligent prosthetics. With advances in sensors and AI, bionic limbs become more capable but also introduce threat vectors. The paper defines the field, demonstrates potential adversarial attacks, and curates open research questions to advance wearable robotics.