6/1/2026
Gap-K%: Measuring Top-1 Prediction Gap for Detecting Pretraining Data
Gap-K% metric detects pretraining data in LLMs by measuring prediction gaps and token correlations for privacy/copyright concerns.
6/1/2026
Decouple Searching from Training: Scaling Data Mixing via Model Merging for Large Language Model Pre-training
Model merging approach decouples data mixture search from training, enabling scalable optimization for LLM pre-training.
6/1/2026
Multi-Agent Teams Hold Experts Back
Study shows self-organizing multi-agent LLM teams underperform expert models; examines coordination emergence in autonomous collaborators.
6/1/2026
The Gaussian-Head OFL Family: One-Shot Federated Learning from Client Global Statistics
One-shot federated learning approach reducing communication overhead through single-round server-client exchange.
6/1/2026
An Odd Estimator for Shapley Values
Novel estimator for computing Shapley values with theoretical analysis of paired sampling heuristic for attribution.
6/1/2026
Plain Transformers are Surprisingly Powerful Link Predictors
Transformer-based approach for link prediction in graphs outperforms GNNs without explicit heuristics or embeddings.
6/1/2026
Mixture of Concept Bottleneck Experts
Mixture of Concept Bottleneck Experts framework improves interpretability and accuracy of concept-based models with multiple task predictors.
6/1/2026
CVE-Factory: Scaling Expert-Level Agentic Tasks for Code Security Vulnerability
Multi-agent framework automatically transforms CVE metadata into executable security vulnerability tasks for evaluating code agent capabilities.
6/1/2026
Stop the Flip-Flop: Context-Preserving Verification for Fast Revocable Diffusion Decoding
Context-preserving verification method reduces flip-flop oscillations in parallel diffusion decoding for faster language model inference.
6/1/2026
Pull Requests as a Training Signal for Repo-Level Code Editing
Clean-PR training paradigm leverages GitHub pull requests to improve repository-level code editing models for multi-file modifications.
6/1/2026
A Kinetic Energy Perspective of Flow Matching
Physics-inspired diagnostic for flow-based generative models measuring kinetic effort along ODE trajectories.
6/1/2026
Inverting Data Transformations via Diffusion Sampling
Probabilistic approach to inverting unknown group transformations using diffusion sampling for transformation recovery on Lie groups.
6/1/2026
Breaking the Simplification Bottleneck in Amortized Neural Symbolic Regression
Addresses scalability bottleneck in amortized neural symbolic regression by improving expression normalization for discovering interpretable analytical expressions.
6/1/2026
A Behavioural and Representational Evaluation of Goal-Directedness in Language Model Agents
Framework for evaluating goal-directedness in LLM agents combining behavioral evaluation with representational interpretability analysis of internal model states.
6/1/2026
Biases in the Blind Spot: Detecting What LLMs Fail to Mention
Proposes automated pipeline for detecting unverbalized biases in LLM chain-of-thought reasoning without requiring predefined categories or hand-crafted datasets.
6/1/2026
Less is Enough: Synthesizing Diverse Data in LLM Feature Space with Sparse Autoencoders
Introduces Feature Activation Coverage metric using sparse autoencoders to measure task-relevant diversity in LLM post-training data synthesis.
6/1/2026
Weight Decay Improves Language Model Plasticity
Studies how weight decay during pretraining improves LLM plasticity and downstream adaptability beyond validation loss optimization.
6/1/2026
SCOPE: Selective Conformal Optimized Pairwise LLM Judging
SCOPE framework calibrates LLM-based pairwise evaluation judges using conformal prediction to control error rates and reduce miscalibration and bias in assessments.
6/1/2026
DTBench: A Synthetic Benchmark for Document-to-Table Extraction
DTBench introduces a synthetic benchmark for evaluating LLM performance on document-to-table extraction tasks requiring complex reasoning and structured output generation.
6/1/2026
The Information Geometry of Softmax: Probing and Steering
Studies information geometry of softmax representations in AI systems, focusing on how models encode semantic structure into representation spaces for behavior production.
6/1/2026
HiPER: Hierarchical Reinforcement Learning with Explicit Credit Assignment for Large Language Model Agents
HiPER proposes hierarchical reinforcement learning with explicit credit assignment for training LLM agents on long-horizon tasks with sparse rewards, addressing multi-turn decision-making challenges.
6/1/2026
Position: Evaluation of ECG Representations Must Be Fixed
Position paper arguing current ECG representation learning benchmarks need improvement to align with clinically meaningful objectives.
6/1/2026
HistCAD: A Constraint-Aware Parametric History-Based CAD Representation, Dataset, and Benchmark with Industrial Complexity
HistCAD benchmark for parametric CAD generation preserving design intent through constraint-aware representations and parametric history.
6/1/2026
From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves
Large reasoning models leak sensitive information in reasoning traces; paper addresses controllability of private information in model outputs.
6/1/2026
The Global Landscape of Environmental AI Regulation: From the Cost of Reasoning to a Right to Green AI
Analysis of environmental costs of AI systems, comparing generative search and reasoning models to previous approaches.
6/1/2026
MASPOB: Bandit-Based Prompt Optimization for Multi-Agent Systems with Graph Neural Networks
MASPOB uses bandit-based optimization with Graph Neural Networks to improve prompts for multi-agent LLM systems in complex workflows.
6/1/2026
NGDBench: Towards Neural Graph Data Management
NGDBench proposes next-generation neural data management systems supporting heterogeneous evolving data with implicit reasoning.
6/1/2026
Rank-Factorized Implicit Neural Bias: Scaling Super-Resolution Transformer with FlashAttention
Rank-Factorized Implicit Neural Bias improves Super-Resolution transformers by enabling FlashAttention without relative positional bias.
6/1/2026
Targeted Speaker Poisoning Framework in Zero-Shot Text-to-Speech
Speech Generation Speaker Poisoning framework removes specific speaker identities from zero-shot TTS models via machine unlearning.
6/1/2026
Variational Routing: A Scalable Bayesian Framework for Calibrated Mixture-of-Experts Transformers
Variational Routing applies Bayesian uncertainty quantification to Mixture-of-Experts transformers at foundation model scale.
6/1/2026
G-STAR: End-to-End Global Speaker-Tracking Attributed Recognition
G-STAR performs end-to-end speaker-attributed speech recognition for long-form multi-party overlapping speech with temporal consistency.
6/1/2026
Prompt Injection as Role Confusion
Research on prompt injection attacks as role confusion, showing LLMs identify text source by style rather than role labels, with measurement techniques.
6/1/2026
Surprised by Attention: Predictable Query Dynamics for Time Series Anomaly Detection
AxonAD detects multivariate time series anomalies by identifying shifts in cross-channel dependencies using attention mechanism analysis.
6/1/2026
REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge
REAL: Novel RL method for training LLMs-as-judges that respects ordinal structure of regression tasks instead of binary rewards.
6/1/2026
Empirical Characterization of Inference-Time Elicited Probability Transformations in Large Language Models
Empirical characterization of inference-time probability transformations in LLMs under chain-of-thought, refinement, and retrieval procedures.
6/1/2026
SimulCost: A Cost-Aware Benchmark and Toolkit for Automating Physics Simulations with LLMs
SimulCost benchmark for cost-aware evaluation of LLM agents on physics simulation tuning, addressing tool-use costs beyond tokens.
6/1/2026
Graph Energy Matching: Transport-Aligned Energy-Based Modeling for Graph Generation
Energy-based generative model for discrete graph data using transport-aligned energy matching for molecular and materials design.
6/1/2026
Beyond Static Uncertainty: Modeling Temporal Uncertainty Dynamics for Probabilistic Time Series Forecasting
Temporal uncertainty dynamics modeling for probabilistic time series forecasting capturing volatility clustering and regime shifts.
6/1/2026
Multi-Level Barriers to Generative AI Adoption Across Disciplines and Professional Roles in Higher Education
Survey study examining structural barriers to generative AI adoption across academic disciplines and professional roles in higher education.
6/1/2026
World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry
World Action Verifier framework improving robustness of world models for planning via forward-inverse asymmetry exploitation.
6/1/2026
Rays as Pixels: Learning A Joint Distribution of Videos and Camera Trajectories
Video Diffusion Model learning joint distribution over videos and camera trajectories simultaneously for novel view synthesis.
6/1/2026
Symmetry Reveals Layerwise Dynamics: How Transformers Perform In-Context Classification
Interpretable transformer analysis for in-context classification via permutation equivariance constraints revealing layerwise dynamics.
6/1/2026
SVL: Goal-Conditioned Reinforcement Learning as Survival Learning
Survival Value Learning probabilistic approach to goal-conditioned RL improving stability and sample efficiency over temporal-difference methods.
6/1/2026
Compile to Compress: Boosting Formal Theorem Provers by Compiler Outputs
Compiler-assisted optimization for LLM-based formal theorem proving, exploiting compiler outputs to reduce test-time compute requirements.
6/1/2026
Aligning Dense Retrievers with LLM Utility via Distillation
Utility-Aligned Embeddings framework improving dense retrieval for RAG by distilling LLM utility signals without expensive re-ranking.
6/1/2026
Robust Lightweight Crack Classification for Real-Time UAV Bridge Inspection
Deep learning approach for real-time UAV-based bridge crack detection addressing weak features, degradation, imbalance, and computational constraints.
6/1/2026
FreeTimeGS++: Secrets of Dynamic Gaussian Splatting and Their Principles
Systematic analysis of 4D Gaussian Splatting methods identifying key factors driving performance in dynamic scene reconstruction.
6/1/2026
Autoregressive Visual Generation Needs a Prologue
Prologue approach bridging reconstruction-generation gap in autoregressive image generation by using separate tokens for each objective.
6/1/2026
OBLIQ-Bench: Exposing Overlooked Bottlenecks in Modern Retrievers with Latent and Implicit Queries
OBLIQ-Bench benchmark exposing limitations in modern retrievers for latent and implicit queries, advancing information retrieval evaluation.
6/1/2026
Why DDIM Hallucinates More Than DDPM: A Theoretical Analysis of Reverse Dynamics
Theoretical comparison of DDPM and DDIM diffusion samplers, analyzing why DDIM produces more hallucinations through reverse dynamics.