Saturday, 31 Oct 2026

Video Generation Meets Agent Harness: Agentic Video Generation, Agentic World Modeling and Recursive Self-Improvement

Maps how agent harnesses coordinate video generation, world models, and self-improvement.

Training Foundation Models with a Single Algorithm

Shows pretraining, fine-tuning, and RL are one likelihood-based training algorithm.

Don't Lose the Thread: Empowering Long-Horizon LLM Agents with Cognitive Resource Self-Allocation

Agent checkpoints verified facts and resets cluttered context between episodes.

When Requirements Interact: Scaling Long-Horizon Coding via Interdependent Specifications

Synthesizes coding tasks with interdependent constraints and audits their tests.

SCARL: Sequence Confidence-Aligned Reinforcement Learning

Aligns each RLVR advantage with the policy confidence in the response.

VeriCUA: Verification as Agentic Exploration for Computer-Using Agents

Verifier that explores the interface to gather evidence of task completion.

L3G-World: Real-Time Infinite World Generation with Latent 3D Gaussian Memory

Latent 3D Gaussian memory keeps world generation real-time over long exploration.

Reward-Adaptive Co-Evolution of Policy and Rubric

Routes training groups and rubric phases by reward informativeness.

VideoMemBench: How Well Do MLLMs Remember in Video Conversations?

Tests whether multimodal models remember earlier turns in long video chats.

FIRME: Registry-Anchored Memory Graphs for Evolving, Heterogeneous Investor Knowledge

Memory graph that keeps investor knowledge tied to identity and source.

Can LLMs Ignore What They Know? Measuring Knowledge Leakage in Persona-Based User Simulation

Tests whether persona simulations leak the model own knowledge.

How Much Should a VLA Trust Its Reasoning? Step-Level Guidance as a Measured Action-Value Function

Guidance weight controls how strongly VLA reasoning steers actions.

HEMO: Harness Self-Evolution for Multi-Objective Combinatorial Optimization

Evolution also adapts the process that discovers optimization operators.

AI Cosmos: An Agentic Infra for Situated Learning

Scenario-based learning infrastructure keeps practice work as reusable assets.

Hybrid AI Patterns: Thirteen Ways to Couple Language Models with Rules, and How to Choose Among Them

Catalogue of thirteen patterns coupling language models with rule engines.

UPPAAi: Towards LLM-Assisted Modeling, Verification, Analysis and Synthesis

LLM workflow for modeling and verifying real-time systems in UPPAAL.

Rethinking Pre-pretraining

Compute-matched tests find synthetic pre-pretraining never beats natural text.

One Vector or Many? Clustered Optimal-Transport Steering under Quantization

Cluster-level optimal transport steering beats one-vector activation edits.

Learning Safe Flow Matching Robot Policies from Binary Feasibility Oracles

Learns robot policies that respect constraints known only through an oracle.

Upcycling Language Models for Multiple Specialized Domains

Upcycled adapters route domain knowledge without separate models.

The generative paradox: the Anti-Prompt paradigm for human capability

Names the gap between team performance and what people still understand.

A Negotiation-Aware Hybrid LLM Framework for Intent Interpretation and Service Discovery

Hybrid LLM framework resolves conflicting natural-language service intents.

Locating and Closing Information Gaps in Multi-Agent Systems

Adds one edge to deliver missing evidence in multi-agent workflows.

Understanding axial attention in TSFMs through single location regression

Analyzes when axial attention is statistically optimal in time series models.

Bridging Semantics and Structure: A Theoretically Grounded Framework for Link Prediction on TAGs

Theory-backed fusion of graph and language models on text graphs.

Towards Full-Text Scholarly Argument Mining using Entity-Anchored Claim Extraction

Extracts claims from full articles using scientific entity anchors.

EpiAgent: Epistemic Self-Play for LLM Agents

Learns explicit beliefs about other agents through epistemic self-play.

Mobilizing the Archive: Bitext Extraction at Scale from Archival Documents for Extremely Low-resource Languages

Extracts parallel text from archival grammars for low-resource languages.

ScienceJudge: Pattern Perturbation Augmentation for Sentence-Level Scientific Claim Verification

Pattern perturbation trains detectors for subtle scientific claim mismatches.

Judged in Company: In-Context Batching Makes LLM Verdicts Depend on Other Items

Shows LLM verdicts shift with the other items in the prompt.

LLM-guided Planar Loco-Manipulation for Quadruped Robots

LLM plans push-and-slide maneuvers for quadrupeds facing blocked routes.

TaskFrame: Connecting Task-Level Planning with Frame-Level Control for Game Agents

Planner-controller pair joins task reasoning with frame-level game input.

Neighborhood Elites: Quality-Diversity Search for Uncertainty Quantification in Question Answering

Quality-diversity search builds better uncertainty probes for question answering.

Latent Path Exploration via Potential-Guided Subtrajectory Balance

Potential-guided exploration steers continuous latent reasoning toward useful paths.

Only Accepted Tokens Count: An Energy Accounting of Speculative Decoding Under Latency Targets

Measures speculative decoding energy by the tokens the target accepts.

DolphinRouting: Cluster-Level Binary Routing for Disaggregated LLM Inference Serving

Learns request routing for disaggregated LLM serving from latency feedback.

Rimuru: Unified Memory Management for Modern LLM Serving via Elastic Expert Offloading

Unified page pool shares memory between model weights and attention states.

Factorized Representations of Concurrent Temporal Variables in Pretrained Language Models

Maps how language models represent independent temporal variables in documents.

RFW-A: Agentic Procedural Generation of 400K Articulated Assets over 500 Categories

Agents build category generators that produce 400K articulated assets cheaply.

A Unified Hessian-Guided Framework for Post-Training Quantization

Hessian-guided objective jointly tunes quantization ranges, transforms, and rounding.

The Athlete as a Scientist of One: Agentic Bayesian Experimental Design for Individualized Physiology from Consumer Wearables

Agent runs falsifiable N-of-1 training experiments for one athlete.

Agentic LLM reasoning in pediatric diabetes outcome prediction from multimodal medical records

Evidence-citing agent predicts pediatric diabetes outcomes better than trained baselines.

Beyond Selected Breakthroughs: Evaluating Language Models on Randomly Sampled Mathematical Conjectures

Benchmarks language models on randomly sampled conjectures, not famous ones.

Understanding and Mitigating Role Spoofing in LLMs

Studies role confusion that lets injected text pose as trusted roles.

Retrieving from Your Own Mistakes: On-Policy Training of Embedding Models for Multi-Hop Retrieval-Augmented Generation

Retrieves multi-hop evidence by training on the model own failed traces.

Prefix Executability: Evaluating Tool-Using Agents Beyond Final Success

Scores tool agents by whether every step of the trajectory stays executable.

Training language models in simulation for psychiatric intake

Multi-turn RL in simulated interviews improves psychiatric intake performance.

Preserving Utility in Continual LLM Unlearning via Orthogonal Projections

Orthogonal projections keep LLM utility stable during repeated unlearning.

rTaste: Do Research Agents Have Taste? Evaluating Early Stage Research Judgment

Tests whether research agents rank experiments as well as executors judge them.

Agentic University: Collaborative Agents for Personalized Learning

Multi-agent organization where learners co-run a revisable capability program.

AI in the EGI e-infrastructure -- Past experiences and future strategies

Reviews AI use in European research infrastructure and future strategy.

A Low-Rank Space Captures Linear Alignment Between LLM and Human Neural Activity During Naturalistic Language Listening

A tiny subspace of LLM embeddings predicts human neural responses.

How I Aligned Your Model: Scaling Up Multilingual Safety Training with Templates

Template pipeline builds multilingual safety corpora in sixty languages.

IndicIPR-QA: A Benchmark for Extractive Question Answering on Indian IPR Laws

Extractive question answering benchmark over seven Indian intellectual property laws.

NEOPIC: A Neural Operator Framework for Particle-based Kinetic Plasma Simulations

Neural operator that learns plasma dynamics directly at particle positions.

BURGER: Benchmarking and Understanding Robotic Generalization via Embodied Drawing

Drawing benchmark that tests how vision-language-action models scale with data.

Homebrewed SLMs: Autoregressive and Masked Diffusion Language Models Built at Home for Portuguese

Tutorial that trains Portuguese language models on one consumer GPU.

Efficient Scalable Oversight via Rewinding

Rewinding gives weak verifiers cheaper checks on powerful model outputs.

TRACER, a time-to-event EHR foundation model for disease detection across health systems

EHR foundation model that forecasts incident diseases across four health systems.

On the Fundamental Limits of Prefill-Based Decoding Length Prediction

Shows decoding-length benchmarks conflate predictor quality with task hardness.

Retrieval as Traversal: Value-Indexed Memory for Stateful Long-Context Agents

Persistent local traversal indexes memories by the value they reveal.

Ramanujan Challenge Results: AI-Assisted Evaluation Pipeline for Multi-Format Proofs

Evaluates AI proof pipelines on research-level number theory problems.

Advancing Clinical Genomic Analysis with a Skill-Centred Agent

Clinical genomics agent that composes curated executable skills for case analysis.

Comparing Human Role-Players and LLM-Simulated Clients in Online Counselling Training: An Analysis of Counselling Patterns

Compares human role-players with LLM clients in counselling training sessions.

Every Token Needs Its Own Teacher: Multi-teacher Adaptive On-policy Self-distillation

Adapts the self-distillation teacher token by token during on-policy training.

PIVOT: Progressive Intervention via Online Teacher Correction

Adds teacher guidance to RLVR when all rollouts in a group fail.

Collaborative AI agents for autonomous experimental research with solid-state quantum spins

Agents run a quantum-spin study, check each other, and revise drifting experiments.

TextSeal2: Distortion-Free LLM Watermarking with Copula Mixing and Memoryless Races

Watermark that survives speculative decoding without hurting diversity or latency.

Testimony from Timelines: Evidence-Grounded Significance Testing of Temporal Clinical Hypotheses

Significance tests that check temporal clinical claims against real records.

ShallowDefend: Harnessing Shallow Safety Alignment for Multi-Turn Defender

Trains defenders to commit a safety decision before writing each reply.

From extreme weather to agricultural consequences with agentic AI

Agentic system that turns weather forecasts into crop hazard judgments.

Rethinking Neural Thickets Without Weight Perturbations

Shows reported gains from weight perturbation come from one block, not dense experts.

Elucidating the Design Space of LM Data Attribution

Tests language model data attribution methods under controlled dataset changes.

Pre-Crime for Agents, at the Last Second: Probing When a Recruited Agent Gives In

A linear probe flags the moment a recruited agent is about to leak.

EnvOS: Auditing Computer-Use Agents When the World Moves On : Phenomenon-Based Task Construction, Certification, and Evidence-Based Evaluation

Certifies changing environments so agent failures can be blamed correctly.

MLLM-GS: Multimodal Large Language Model Prior-Guided Fast Novel View Synthesis in Extreme Environments

Multimodal model priors speed novel view synthesis in degraded scenes.

BiHashFormer: Hash-Driven Dual-Branch Transformer for Efficient Object Detection in HRW Shots

Hash-based dual-branch detector finds small objects in gigapixel images.

ElasticFormer: Detecting Objects in HRW Shots via Elastic Computing Vision Transformer

Elastic sparse backbone detects objects in gigapixel wide shots.

UniRA: Unified Representation Alignment and Auto-Encoding for Efficient Diffusion Generation

One tokenizer serves as diffusion target and alignment anchor.

Mind The Gap (MTG-AD): Dataset and Fine-grained Evaluation for End-to-End Inline Audio Descriptions

Dataset and metrics for audio descriptions that fit video silences.

Blind Video Denoising for Vocal Tract rtMRI using a Self-Supervised Ensemble

Self-supervised ensemble denoises real-time vocal tract MRI without clean references.

Greenscapes: Optimized Vegetation Configurations in Cities

Optimizes urban vegetation for shade, microclimate, and style.

DINOv2 for PET: A Pre-Training Domain Benchmark Across Classification, Segmentation, and Prognosis

Benchmarks DINO pretraining across PET classification, segmentation, and prognosis.

Segment Anything Model Based Context Guided Mamba Unet for Medical Image Segmentation

SAM context and Mamba features combine for medical segmentation.

EPC-RawDC: Equivariant Physics Constrained 3D Diffusion with Raw Data Consistency for CBCT to CT Image Synthesis

Physics constraints from raw scanner projections improve CT synthesis.

Localizing Concepts in Visual Autoregressive Models

Locates where and when concepts form in visual autoregressive models.

BRACHIO: Broad Representations Across Centres for Head-and-neck Imaging Oncology (HECKTOR 2026)

Report on HECKTOR head-and-neck cancer challenge submissions.

Uncertainty-aware Coarse-to-Fine Main Pancreatic Duct Segmentation for Width Measurement in 3D MRCP

Coarse-to-fine model measures pancreatic duct width with uncertainty.

Morphology-Aware Channel-Decoupled Mamba Network for Medical Image Segmentation

Mamba network separates organ shape and contrast for CT segmentation.

Neural Blind Deconvolution for Image Scanning Microscopy

Blind deconvolution estimates microscope response functions from the scans themselves.

RepConTUR: A Report-Conditioned Volumetric Multimodal Model for Abdominal Tumor Segmentation

Radiology reports condition native 3D tumor segmentation.

Coupled Guidance for Flow Matching

Coupled guidance brings classifier-free guidance to flow matching models.

D-MOD: Diffusion mode estimation for accelerated MRI

Algorithm returns a posterior mode instead of averaging many MRI samples.

SelfThinker: Self-Evolving Reasoning via Reliability-Verified Intrinsic Visual Thoughts

Self-generated visual thoughts improve understanding without human image labels.

SynthFaces: Balanced Large Scale Human Dataset

Synthetic balanced face dataset reduces identity bias and privacy risk.

Learning a Motion Manifold to Constrain Ambiguity in Unsupervised 4D Medical Image Interpolation

Motion manifold constrains ambiguous intermediate anatomy in 4D interpolation.

It Takes Two to Probe, Thousands to Retrieve: Putting Vision--Language Probes to the Retrieval Test

Shows fine-grained probes overstate how well models retrieve.

NeurEO: dissecting Earth observation embeddings with computational neuroscience

Neuroscience methods decode what Earth observation embeddings represent.

Mode Dynamics: A Framework for Neural Network Learning

Links collective parameter motion to changes in model outputs.

Hierarchical Step Decomposition for Training-free Online Video Step Grounding

Groups video into semantic events to ground procedural steps online.

Fisher Contrastive Learning to Mitigate Feature Suppression: A Case Study in Acute Promyelocytic Leukemia Diagnosis

Fisher regularized contrastive learning stops dominant factors suppressing others.

EventSeeColor: Learning to Reconstruct Color from Monochrome Events

Reconstructs scene color from monochrome event camera streams.

Beyond CCT: Multi-channel SPD recovery for per-shot camera color correction

Recovers full light spectra for per-shot color correction.

AdMuse: Skill-Guided Creative Advertising Image Generation

Multi-agent system reuses creative skills to plan advertising images.

Decision-Level Error Attribution and Depth-Aware Safety Evaluation of Monocular RGB Human Tracking for Autonomous Robot Navigation

Separates safety-rule error from perception error in robot human tracking.

Benchmarking color judgment in multimodal LLMs with CIE-computable ground truth

Benchmarks multimodal model color judgment against colorimetric ground truth.

Confidence-Aware Speed-over-Ground Estimation from Doppler Spectra via Reconstruction-Based Representation Learning

Estimates rail speed from Doppler spectra with frame-level confidence.

From MS/MS evidence to adaptive spectroscopic characterization: Reinforcement learning guided by candidate molecular structures

RL policy decides which extra measurements to request for molecules.

Latent Prediction Yields Cross-Subject Alignable Representations in Sensory-Evoked fMRI

Self-supervised fMRI encoders align brain representations across subjects.

LoRe-MIM: Lost-or-Replaced Masked Image Modeling for Self-Supervised Representation Learning in Whole-Body PET/CT

Masked image modeling learns PET/CT representations without dense labels.

Self-training for cross-centre coronary vessel segmentation in X-ray angiography using pre-contrast frame subtraction

Self-training with pseudo-labels improves coronary segmentation across hospitals.

Causal-prior-constrained deep learning for rapid-intensification warning of western North Pacific tropical cyclones

Causal priors help deep models warn of cyclone rapid intensification.

Textual Response-Aware Guidance for Zero-Shot Adversarial Robustness of Vision-Language Models

Text encoder responses guide robust fine-tuning of vision-language models.

Radiology Guideline-derived Decision Graph for Efficient Clinical Decision Support

Compiles radiology guidelines into a decision graph for case evaluation.

Multistep and Multivariate Time Series Explainability with GradientSHAP: A Wind Power Case Study

GradientSHAP explains multivariate wind power forecasts step by step.

Wizard: Manipulate Any World in Real Time via Agentic Control

Agentic game world model gives real-time control of character motion.

HiFOR: Hierarchical Foresight with Action-Guided World Grounding for Vision-Language-Action Model

Links short-term foresight with long-term goals in VLA planning.

Composing and Steering Diffusion Models with Generalized Feynman-Kac Correctors

Correctors sample from combinations of diffusion models, including reward-tilted ones.

Scaling Mathematical Proof Supervision from Structured Textbooks

Converts mathematics textbooks into structured proof supervision at scale.

Continuous-Action World Models: Learning Cross-Embodiment Spatiotemporal Dynamics Priors

Shared latent functions let world models transfer actions across robots.

Bridging Neural Representations and Symbolic Semantics in Audio Classification

Induces human-understandable acoustic concepts from audio classifier neurons.

OCCAMS: A Multi-Task Learning-Based Investigation of the CMB Power Spectra

Multi-task autoencoders recover cosmological parameters from CMB spectra.

IMPACT: Inverter-Informed Multimodal Learning for Photovoltaic Generation Forecasting

Physics-informed multimodal model forecasts photovoltaic power minutes ahead.

Soil Moisture Estimation in Precision Agriculture: A Knowledge Distillation Approach

Knowledge distillation estimates soil moisture without expensive aerial imagery.

PhysInsert: Physically Plausible Object Insertion with Geometric Priors and Bidirectional Object-Scene Interactions

Inserts objects into scenes with geometry priors and two-way interactions.

A Unified Benchmark for Aerial-Enhanced Autonomous Driving: The KITScenes Aerial Dataset

Aerial imagery benchmark unifies mapping and cross-view localization for driving.

InfantNet: A Large Scale Dataset for Infant Body Pose and Shape Estimation

Large infant pose and shape dataset for childcare and diagnosis.

Accurate MRI-free PET Quantification and Reliability Assessment via CT-to-MRI Synthesis and Conformal Prediction

CT-to-MRI synthesis with conformal prediction enables MRI-free amyloid PET quantification.

AVPS-360: A Spatial Audio-Visual Benchmark for Instance Segmentation and Tracking of Sound-Producing Persons

Spatial audio-visual benchmark for segmenting and tracking sound-producing people.

The Precision Wall: An Optimizer-Dependent Boundary for Optimization under Power-Law Measurement Spectra

Explains an optimizer precision wall under power-law measurement spectra.

MetaRefine: Score Refinement with Multi-Level Features for Image Assessment

Image assessment model that keeps original geometry and composition cues.

LucidLyra: Feed-Forward Generative 3D Scene Reconstruction and Joint Understanding via Multi-Task Distillation

Reconstructed 3D scenes stay queryable for semantics without per-scene optimization.

Numerical Integration on Bounded Domains via Learned Reflected Diffusions

Learned reflected diffusions propose points for integration on bounded domains.

Algebraic World Model: From Exploration to Proofs

Turns exploration in unknown environments into explicit algebraic models and proofs.

Discovering Interpretable Acoustic Features in Autoregressive Voice-Cloning TTS Models

Probes which acoustic features voice-cloning models actually use.

CrossWeave: Cross-Modal Scene and Instance Retrieval from Sparse 2D-3D Alignment

One model handles scene and instance retrieval from sparse 2D-3D matches.

PhysCount: Do MLLMs Count Visual Appearances or Physical Entities?

Benchmark that separates counting visible appearances from counting physical entities.

RADIOlogy: A Single-Pass Approach to Training-Free Zero-Shot Anomaly Detection

Single-pass zero-shot anomaly detection from pooled vision-language features.

Beyond Ground Truth: A Heuristic Unsupervised Quality Assurance Engine for Diagnosing Calibrational Blindness in Dental Segmentation

Quality engine that detects confident-but-wrong dental segmentations without ground truth.

HairDeck: Animation-Ready Stylized Hair Cards from a Single Image

Generates animation-ready stylized hair cards from a single character image.

Fight World Model: Customizable Combat through Visual Interaction Proxies

Game engine rules drive combat video generation through aligned interaction proxies.

Difficulty Calibration Across Time for Membership Inference in Diffusion Models

Separates image difficulty from memorization to improve membership inference auditing.

An ODE-based Sampler for Discrete Diffusion Models

Exponential residual flows give discrete diffusion models an ODE-style sampler.

SymAnt: Grounding Long-Horizon Action Anticipation with Adaptable Symbolic Priors

Neuro-symbolic model that grounds long-horizon action prediction in symbolic priors.

Probing Spatial State Representations in Visual Foundation Models

Tests whether visual foundation models hold a persistent spatial state.

world2vec: Mechanistic world models through latent dynamics learning

Learns interpretable world model dynamics from observations, without reward signals.

SonoWorld4D: From One Video to a 4D Audio-Visual Scene via Diffusion Synchronization

Turns one video into an explorable 4D scene with matched spatial sound.

Luggage re-identification: A study of strategies towards a real-world application

Studies strategies for re-identifying luggage across camera views in real settings.

Read, Allocate, Route: Role-Coupled Tokenization for Pixel-Space Generative Transformers

Pixel-space transformer that keeps global attention small and models fine detail locally.

Micro-Edits: Computational Support for Fine-Scale Edits in Film

Tools that speed up fine-scale frame edits in film editing.

Propose-Filter-Release: Scaling Noise to Typical Sensitivity via Private Filtering

Keeps private noise small by first filtering the few records that dominate sensitivity.

Human--AI Interactions as First-Class Data: A Formal Interaction Model and Runtime for Explainable AI Collaboration

Makes human-AI interactions durable, replayable data instead of throwaway prompts.

Agentic process evidence

Calls for provenance records when AI agents change production software.

DynaRev: Benchmarking Web Agents under Dynamic Goal Revision

Benchmark tests web agents when users revise goals mid-task.

Hammering with Vampire in Lean

Vampire prover discharges Lean goals with kernel-checked proof reconstruction.

Fast and accurate neural triage decoding of ultra-high-rate quantum LDPC codes

Neural decoder triages ultra-high-rate quantum error correction codes.

Gym-Anything for Robotics: Scaling physical simulations with digital agents

Digital agents build diverse physics simulations for robot training at scale.

When to Assist and What to Suggest: Joint Gating and Memory-Grounded Hint Generation in One Compact Decoder

One compact decoder decides when to give proactive hints and what to say.

A Deep Learning Framework for Quantile Demand Forecasting in E-commerce with Doubly Censored Data

Quantile demand forecasts handle doubly censored e-commerce sales data.

FARE: Factor-Aware Retrieval with Error-Guided Fusion for Retail Forecasting

Retrieves partial factor matches to forecast retail demand spikes.

Leveraging Recourse and Size Bounds for OnlineKnapsack with Removable Items

Online knapsack gains competitive ratios when accepted items can be removed.

Machine Learning-Based Detection of Mobile Money Scan Messages in Ghana: A Comparative Evaluation of Naive Bayes, Support Vector Machine, and Randon Forest Classifiers

Compares three classifiers for detecting mobile money fraud messages in Ghana.

H$^2$R: Hierarchical Hint-Based Reward for Mathematical Reasoning via Reinforcement Learning

Sub-question hints give RLVR denser reward for math reasoning.

Cheap Retrieval and Observation Credit Assignment from One Quantity: The Value of Information

Value of information guides both memory retrieval and its credit.

Reinforcement Learning Under Runtime Confounding: Policy Learning and Off-Policy Evaluation

Policy learning and off-policy evaluation when deployment hides key variables.

Escaping Path Mirages in Offline Goal-Conditioned Reinforcement Learning

Offline goal-reaching escapes deceptive apparently successful trajectories.

Closing the Log Gap: Minimax-Optimal Regret for Episodic RL

First regret bound matching the minimax lower bound in episodic RL.

DYNAS: A Dynamic Network Penetration Testing Framework Using Deep Hierarchical Reinforcement Learning

Simulator trains penetration testing agents under noisy, changing defenses.

Distributional Reward Calibration and Budget-Neutral Credit Assignment for Multimodal RLVR

Factorizes GRPO advantages into baselining and token credit assignment.

Learned Terminal Costs for Short-Horizon Model Predictive Control of Autonomous Trucks in Interactive Highway Traffic

Learned terminal costs improve short-horizon truck control in traffic.

Is It the Curriculum or Just the Easy Start? Benchmarking Unsupervised Environment Design on Quadrupeds

Benchmark compares unsupervised curricula on five quadruped robots in MuJoCo.

Uncertainty-aware Feedback Loop Diagnosis: integrating reinforcement learning and conformal prediction for reliable wafer defect classification

RL plus conformal prediction gives reliable wafer defect classification.

Adaptive search with path-constrained mobile sensors using multi-arm bandits

Bandits pick sensor paths under energy and obstacle constraints.

Reinforcement learning for control of partial differential equations

Survey of reinforcement learning for controlling partial differential equations.

When to spend privacy: optimal allocation in private learning

Control theory sets the noise schedule for private training budgets.

Observability-Aligned Evidential Masked Pretraining for Reliable VRU Motion Prediction

Masked pretraining ties prediction confidence to what the model can observe.

Accumulating Pruning Constraints: New Exact Polynomial-Space Algorithms for Max 2-SAT

Pruning constraints give faster exact Max 2-SAT algorithms.

A physics-informed uncertainty-aware surrogate for $p$-$n$ junction diodes

Uncertainty-aware surrogate predicts p-n junction diode behavior from device parameters.

Enhancing Atrial Fibrillation Detection Through Hyperparameter Optimization

Hyperparameter tuning lifts atrial fibrillation detection across model families.

Efficient Learning of Mechanism Designs and Compositionality

Reusable constraint modules improve learning of mechanism behavior from few examples.

FIDE: Fiber-Informed Differentiable Experimental Design for Reference-Relative Law Dynamics

Differentiable design resolves which measurement system best identifies a law.

Does System Size Matter in Uncertainty Quantification for Machine Learning Interatomic Potentials?

Shows uncertainty scores in interatomic potentials shift with system size.

Hidden Disparities in Ranking: Category-Aware Fairness via Distributional Stability

Formal fairness framework for rankings with category-aware utility parity.

Reaction precedents extend retrosynthetic prediction beyond training coverage

Retrieval over reaction precedents recovers accuracy outside training coverage.

Conformal Annotation with Randomized Expert Review: Application to Online Cell-Type Labeling

Conformal calibration decides which AI cell labels skip expert review.

Statistical Analysis of Synthetic Data with Verification: A Linear Regression Framework

Statistical framework studies verified synthetic data and model collapse.

Decision and System One models in rheumatology: accuracy and calibration on specialty-training examination questions and diagnostic case vignettes

Measures accuracy and calibration of reasoning models on rheumatology exams.

How to Loop Your Transformer: Loop Scaling, Specialization, and Parallel Inference

Maps when weight looping wins, and how to fix its slowdowns.

Your LR Schedule May Not Be Principled: Effective Learning Rate Scheduling in Weight-Normalized Training

Shows weight-normalization gains come from the effective learning rate trajectory.

Deep Learning Based Behavioral Drift Modeling for Continuous Biometric Authentication

Two-level keystroke system monitors behavioral drift during continuous authentication.

NECA: Neighbor Event Curve Augmentation for Multi-Horizon ICU Deterioration Prediction

Neighbor event curves improve multi-horizon ICU deterioration prediction.

A Pretrained Model for Bayesian Filtering in Heterogeneous Dynamical Systems

Foundation model for Bayesian filtering across heterogeneous dynamical systems.

Framework-Conditional Open Materials Generation

Generates stable novel materials conditioned on the chosen framework.

Learning developmental dynamics from lineage context

Generative model predicts cell state dynamics from lineage context.

Glance: The Generator Is Its Own Reranker

The generator reranks its own retrieved passages with a lightweight head.

Rural Readiness: Domain Generalization with Adversarially Chosen Domains

Domain generalization benchmark where the test domains are chosen adversarially.

FARMS: A Verifiable Foundation for Automated Computational Science

Turns computational mathematics into AI-ready knowledge with explicit guarantees.

Multi-View MoE: Letting Every Router Shape Expert Computation

Lets every router shape expert computation in sparse mixture-of-experts models.

Multistatic ISAC with Deep Learning-Based Clutter Suppression for 5G and Beyond

Deep learning suppresses clutter for multistatic sensing with 5G signals.

Adaptive Scenario Generation for KL-DRO Around a Learned Density Model

Builds robust scenario sets around a learned density model for stochastic programs.