6/30/2026
Scalar Representations of Neural Network Training Dynamics
Application of scalar embeddings to analyze neural network training trajectories as temporal networks for understanding optimization dynamics.
6/30/2026
Predict, Reuse, and Repair: Accelerating Dynamic Sparse Attention for Long-Context LLM Decoding
Speculate-reuse-repair runtime optimizing dynamic sparse attention for long-context LLM decoding by exploiting temporal locality in block selections.
6/30/2026
Beyond IID: How General Are Tabular Foundation Models, Really?
Evaluation of tabular foundation models on diverse out-of-distribution tasks revealing limitations of current benchmark protocols and model generalization.
6/30/2026
Diffusion Fine-tuning with Rewarded Moment Matching Distillation
Framework combining diffusion model distillation with RL fine-tuning via Rewarded Moment Matching Distillation to improve generative quality.
6/30/2026
Experience Augmented Policy Optimization for LLM Reasoning
Method for improving LLM reasoning via reinforcement learning with verifiable rewards by reusing accumulated experience rather than on-policy optimization from scratch.
6/30/2026
Proofs of Ownership for Machine Learning Models
Formal framework for proving ML model ownership through game-theoretic analysis between model owner, thief, and judge.
6/30/2026
Arko-T: A Foundation Model for Text-to-Structured 3D Generation
4B-parameter text-to-design model generating executable parametric CAD programs from natural language descriptions for mechanical part design.
6/30/2026
When Does Online Imitation Learning Help in LLM Post-Training? The Role of (Non-)Realizability Beyond Horizon
Research on when online imitation learning improves LLM post-training, showing benefits depend on realizability rather than error accumulation reduction.
6/30/2026
Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring
Study of internal-state probes for monitoring AI agents, finding they read situation context rather than enabling pre-action misalignment detection across model families.
6/30/2026
Exploring Differences Between Tabular Enterprise Data and Public Benchmarks
Analysis of statistical characteristics and performance measurements of enterprise tabular data versus public ML benchmarks for business applications.
6/30/2026
HSAP: A Hierachical Sequence-aware Parallelism for Hybrid-Context Generative Models
Proposes HSAP sequence parallelism framework for hybrid-context packed sequences in large language models, fixing cross-contamination in causal attention.
6/30/2026
MuonSSM: Orthogonalizing State Space Models for Sequence Modeling
Introduces MuonSSM framework stabilizing state space models for long-sequence modeling by conditioning update geometry rather than recurrent weights.
6/30/2026
Discovering Collaboration from Novelty: Random Network Distillation for Clustered Federated Learning
Proposes lightweight approach for clustered federated learning using random network distillation to discover client collaborations without coupling cluster assignment to training.
6/30/2026
Muon learns balanced solutions in matrix factorization without slow saddle-to-saddle dynamics
Studies Muon optimizer dynamics on matrix factorization problems, showing it avoids slow saddle-to-saddle transitions compared to gradient descent.
6/30/2026
Informational Frustration in Neural Manifolds: Shannon Bottlenecks and the Limits of Learnability
Theoretical framework linking information theory and topology to explain generalization in overparameterized deep networks, addressing theory-practice gap.
6/30/2026
ITSPACE: Monotone Gaussian Optimal Transport Updates
Proposes ITSPACE algorithm for optimal transport updates on covariance matrices using Bures-Wasserstein distance for domain adaptation.
6/30/2026
Convergence of Continual Learning in Homogeneous Deep Networks
Theoretical analysis of convergence properties in continual learning with deep networks, characterizing sequential projections onto task margin sets.
6/30/2026
TraceLab: Characterizing Coding Agent Workloads for LLM Serving
TraceLab characterizes real-world coding agent workloads and LLM serving patterns across multiple models for systems optimization.
6/30/2026
Attractor States Emerge in Multi-Turn LLM Conversations
Study analyzing attractor state emergence in multi-turn LLM conversations, showing topic-independent stable behaviors in debate interactions.
6/30/2026
SWE-INTERACT: Reimagining SWE Benchmarks as User-Driven Long-Horizon Coding Sessions
SWE-Interact testbed evaluating coding agents on multi-turn interactive tasks with progressive user requirements instead of complete upfront specifications.
6/30/2026
C$^{2}$R: Cross-sample Consistency Regularization Mitigates Feature Splitting and Absorption in Sparse Autoencoders
Cross-sample Consistency Regularization method addressing feature splitting and absorption problems in Sparse Autoencoders for LLM interpretation.
6/30/2026
Pessimism's Paradox: Conservative Offline Training Amplifies Reward Hacking During Online Adaptation in Reasoning Models
Study showing conservative offline training paradoxically amplifies reward hacking in reasoning models during online adaptation with DPO.
6/30/2026
One-Step Gradient Delay is Not a Barrier for Large-Scale Asynchronous Pipeline Parallel LLM Pretraining
Analysis showing one-step gradient delay doesn't hinder asynchronous pipeline parallelism for large-scale LLM pretraining with PipeDream-2BW.
6/30/2026
ReFreeKV: Towards Threshold-Free KV Cache Compression
ReFreeKV method for KV cache compression in LLM inference without requiring pre-determined domain-specific thresholds.
6/30/2026
TextClusterLab: An Integrated Framework for Reliable Text Clustering Studies
TextClusterLab framework for reliable evaluation of text clustering algorithms addressing dataset quality and semantic boundary challenges.
6/30/2026
PIXELRAG: Web Screenshots Beat Text for Retrieval-Augmented Generation
PixelRAG method for retrieval-augmented generation using website screenshots in pixel space instead of parsed text for improved context.
6/30/2026
Agentic Safety is an Epistemic Property, Not a Behavioral One
Argues AI agent safety is epistemic property dependent on system correctability during learning, not just current behavior snapshots.
6/30/2026
Generating in the Limit with Infinitely Many Hallucinations
Theoretical framework studying language generation in the limit and hallucinations as unavoidable consequence of learning.
6/30/2026
Zero-Label Driving Scenario Complexity Detection via Joint Embedding Predictive Architecture
Unsupervised method for detecting complex driving scenarios using Joint Embedding Predictive Architecture without labels.
6/30/2026
RADIANT-PET: Reasoning-Augmented PET/CT Lesion Segmentation with Large Language Models and Reinforcement Learning
RADIANT-PET framework combining segmentation models with LLM adjudication for improved lesion segmentation in PET/CT medical imaging.
6/30/2026
RadarTwin: Scene-Specific mmWave Radar Simulation and Learning for Mobile Indoor Perception
RadarTwin framework for generating synthetic mmWave radar training data using 3D reconstruction and vision-language models for mobile perception.
6/30/2026
Meta-learning as a principle for human-like visual representations
Research proposing meta-learning as principle for human-like visual representations in neural networks to support open-ended task flexibility.
6/30/2026
Vision-driven Preference Synthesis for Mitigating Hallucinations in VLMs
Method to reduce hallucinations in Vision-Language Models using preference alignment constructed from vision-driven synthesis rather than intervention-based approaches.
6/30/2026
Reinforcement Learning for Software Vulnerability Analysis: A Systematic Review with Emphasis on C/C++ Source Code and Static Analysis
Systematic review of reinforcement learning techniques for C/C++ vulnerability detection and static analysis following PRISMA guidelines.
6/30/2026
LoRA-Tuned Large Language Models for Dementia Detection via Multi-View Speech-Derived Features
LoRA fine-tuning of LLMs for dementia detection using multi-modal speech features with automatic speech recognition transcripts.
6/30/2026
Event-Conditioned Diagnostics of Kinematic, Contact, and Object-Permanence Fields in Passive Object-State World Models
Research on how world models organize physical information in latent representations using diagnostic protocols for passive object-state prediction.
6/30/2026
Spectral phase transitions and trainability in neural network learning dynamics
Theoretical analysis of spectral phase transitions in neural network weight matrices during SGD training.
6/30/2026
Turn-Averaged SAEs for Feature Discovery and Long-Context Attribution
Turn-averaged sparse autoencoders for interpretable feature extraction in language models with long context.
6/30/2026
DataComp-VLM: Improved Open Datasets for Vision-Language Models
DataComp benchmark for evaluating vision-language model dataset curation strategies with 160 open datasets.
6/30/2026
Depth-Staggered Fibonacci Spacing for Sparse Attention: Static Schedules Beat Learned Dilation and Extrapolate Where Dense Attention Fails
Study of sparse attention mechanisms using Fibonacci-spaced offsets in language models with depth-based scheduling.
6/30/2026
Neuromorphic Energy-Aware Learning for Adaptive Deep Brain Stimulation
Energy-aware learning approach for neuromorphic computing in closed-loop deep brain stimulation systems.
6/30/2026
Reproducing FACTER: Fairness via Conformal Thresholding and Prompt Repair
Reproducibility study of FACTER framework for fairness and coverage in LLM-based recommendation systems.
6/30/2026
Transformer-Based Active Learning for Data-Efficient Vaccine Epitope Selection in PRRS
Transformer-based active learning approach for efficient vaccine epitope selection using molecular docking simulations.
6/30/2026
AnTenA: Actionable and Explainable Tensor Analysis System with Large Language Models
LLM-based system for explaining patterns in multi-aspect tensor data without requiring labels or metadata.
6/30/2026
Self-Supervised Theorem Discovery in a Formal Axiomatic System
AI agent that autonomously discovers mathematical theorems in formal axiomatic systems without human priors, advancing machine reasoning capabilities.
6/30/2026
Structure-Preserving Document Translation via Multi-Stage LLM Pipeline: A Case Study in Marathi
Multi-stage LLM pipeline for preserving document structure and formatting in machine translation of Indian government documents to Marathi.
6/30/2026
Memory-Managed Long-Context Attention: A Preliminary Study of Editable Request-Local Memory
Memory-managed attention mechanism separating long-context compression from explicit memory write/edit operations in language models.
6/30/2026
A Task-Driven and Quality-Assured Agent Framework for SAR Data Generation
SAGA: Agent framework for synthetic aperture radar data augmentation with task-driven generation and quality validation.
6/30/2026
A Theoretical Interpretation of In-Context Learning via Probabilistic Modeling
Probabilistic theoretical framework explaining in-context learning in LLMs with rigorous performance analysis.
6/30/2026
Cybersecurity is the True Frontier for Generative AI Success or Failure
Position paper on cybersecurity as testbed for LLM-based autonomous agents handling heterogeneous tools and data.