4/6/2026
Reinforcement Learning-based Knowledge Distillation with LLM-as-a-Judge
RL framework using LLM-as-judge for label-free knowledge distillation and reasoning improvement in small and large language models without ground truth labels.
4/6/2026
Evaluating the Formal Reasoning Capabilities of Large Language Models through Chomsky Hierarchy
Systematic evaluation of LLM formal reasoning capabilities using Chomsky hierarchy and computation theory complexity metrics.
4/6/2026
Transfer Learning for Loan Recovery Prediction under Distribution Shifts with Heterogeneous Feature Spaces
Transfer learning approach for loan recovery prediction addressing data scarcity and distribution shifts across portfolio domains.
4/6/2026
Toward an Artificial General Teacher: Procedural Geometry Data Generation and Visual Grounding with Vision-Language Models
Procedural geometry data generation and visual grounding with vision-language models for Referring Image Segmentation in education.
4/6/2026
Learning from Synthetic Data via Provenance-Based Input Gradient Guidance
Learning from synthetic data using provenance-based input gradient guidance to teach models discriminative regions.
4/6/2026
SkillRT: Compiling Skills for Efficient Execution Everywhere
SkillRT: Compiler-inspired system treating LLM agent skills as code for consistent, efficient execution across diverse platforms.
4/6/2026
The Compression Gap: Why Discrete Tokenization Limits Vision-Language-Action Model Scaling
Analysis of why discrete tokenization limits Vision-Language-Action model scaling; introduces Compression Gap information-theoretic principle.
4/6/2026
Learning the Signature of Memorization in Autoregressive Language Models
First transferable learned membership inference attack on fine-tuned LLMs using unlimited labeled data from fine-tuning process itself.
4/6/2026
Efficient Causal Graph Discovery Using Large Language Models
LLM framework for causal graph discovery using breadth-first search queries, reducing complexity from quadratic to linear query requirements.
4/6/2026
Supplementary Materials to Graph Convolutional Branch and Bound
Integration of neural networks into combinatorial optimization for NP-hard problems, learning heuristics and optimality scores via graph convolutional networks.
4/6/2026
Amortized Inference of Causal Models via Conditional Fixed-Point Iterations
Amortized inference framework training single model to predict causal mechanisms across multiple datasets for out-of-distribution generalization.
4/6/2026
Distributional Statistics Restore Training Data Auditability in One-step Distilled Diffusion Models
Method for detecting unauthorized training data in one-step distilled diffusion models using distributional statistics instead of memorization detection.
4/6/2026
Zero-shot Concept Bottleneck Models
Zero-shot concept bottleneck models enabling interpretable predictions without target task training by leveraging pre-trained vision-language models.
4/6/2026
Accelerated Learning with Linear Temporal Logic using Differentiable Simulation
Integration of linear temporal logic specifications into RL using differentiable simulation for safe, correct-by-construction controller synthesis.
4/6/2026
Beyond Noisy-TVs: Noise-Robust Exploration Via Learning Progress Monitoring
Noise-robust exploration method for RL using learning progress monitoring to escape unlearnable noise sources with improved sample efficiency.
4/6/2026
ARMOR: High-Performance Semi-Structured Pruning via Adaptive Matrix Factorization
ARMOR: one-shot post-training pruning algorithm for LLMs achieving 2:4 semi-structured sparsity with minimal performance degradation for efficient deployment.
4/6/2026
High-probability Convergence Guarantees of Decentralized SGD
Theoretical analysis of convergence guarantees for decentralized stochastic gradient descent with high-probability bounds and reduced assumptions.
4/6/2026
Local Reinforcement Learning with Action-Conditioned Root Mean Squared Q-Functions
Extension of Forward-Forward algorithm to reinforcement learning with action-conditioned Q-functions, replacing backpropagation with local learning.
4/6/2026
f-INE: A Hypothesis Testing Framework for Estimating Influence under Training Randomness
f-INE framework for stable influence estimation under training randomness, addressing instability in sample-level impact estimation for data curation.
4/6/2026
Diffusion Models as Dataset Distillation Priors
Dataset distillation method leveraging diffusion models as priors to synthesize compact, representative datasets with improved diversity and generalization.
4/6/2026
Steering Autoregressive Music Generation with Recursive Feature Machines
MusicRFM framework enabling fine-grained control over pre-trained music generation models by steering internal activations via Recursive Feature Machines.
4/6/2026
Fast and Robust Simulation-Based Inference With Optimization Monte Carlo
Bayesian parameter inference method for complex stochastic simulators using differentiable approaches to reduce simulation costs in high-dimensional spaces.
4/6/2026
Goal-Driven Reward by Video Diffusion Models for Reinforcement Learning
Goal-driven reward signals from pretrained video diffusion models for reinforcement learning agent training.
4/6/2026
Pushing the Limits of Distillation-Based Continual Learning via Classifier-Proximal Lightweight Plugins
Distillation-based continual learning with classifier-proximal plugins addressing stability-plasticity tradeoff.
4/6/2026
Resting Neurons, Active Insights: Robustify Activation Sparsity for Large Language Models
Method to robustify activation sparsity in LLMs by addressing representational instability during inference acceleration.
4/6/2026
Forget Many, Forget Right: Scalable and Precise Concept Unlearning in Diffusion Models
Scalable multi-concept unlearning in text-to-image diffusion models addressing weight conflicts and collateral damage.
4/6/2026
Textual Equilibrium Propagation for Deep Compound AI Systems
Textual Equilibrium Propagation for optimizing compound AI systems with multiple modules in long-horizon agentic workflows.
4/6/2026
When RL Meets Adaptive Speculative Training: A Unified Training-Serving System
Unified training-serving system combining RL with adaptive speculative decoding for accelerated LLM inference.
4/6/2026
Infusion: Shaping Model Behavior by Editing Training Data via Influence Functions
Infusion: Framework using influence functions to craft training data perturbations that induce targeted model behavior changes.
4/6/2026
CeRA: Overcoming the Linear Ceiling of Low-Rank Adaptation via Capacity Expansion
CeRA: Parameter-efficient fine-tuning method extending LoRA with non-linear capacity expansion via gating and dropout.
4/6/2026
SafeSci: Safety Evaluation of Large Language Models in Science Domains and Beyond
SafeSci: Framework for evaluating safety of large language models in scientific domains with comprehensive benchmarks.
4/6/2026
CRISP: Compressed Reasoning via Iterative Self-Policy Distillation
CRISP: Method for teaching LLMs to reason more concisely via self-distillation with 'be concise' conditioning.
4/6/2026
Pretrained Video Models as Differentiable Physics Simulators for Urban Wind Flows
WinDiNet uses pretrained video diffusion model as differentiable physics simulator for urban wind flow prediction, replacing expensive CFD simulations.
4/6/2026
$\lambda$-GELU: Learning Gating Hardness for Controlled ReLU-ization in Deep Networks
λ-GELU parameterized gating function enabling controlled ReLU conversion while maintaining smooth activation properties for deployment.
4/6/2026
ERPO: Token-Level Entropy-Regulated Policy Optimization for Large Reasoning Models
ERPO method for token-level credit assignment in LLM reasoning models, addressing entropy collapse in GRPO through information heterogeneity.
4/6/2026
Temporal Credit Is Free
Recurrent network training without Jacobian propagation using hidden state temporal credit. Studies gradient normalization and online adaptation.
4/6/2026
The Spectral Edge Thesis: A Mathematical Framework for Intra-Signal Phase Transitions in Neural Network Training
Mathematical framework explaining phase transitions in neural network training via spectral gap of parameter update Gram matrices. Grokking and capability gains analysis.
4/6/2026
Transfer learning for nonparametric Bayesian networks
Transfer learning for nonparametric Bayesian networks under scarce data. Proposes PC-stable-transfer and hill climbing transfer learning methods.
4/6/2026
Efficient and Principled Scientific Discovery through Bayesian Optimization: A Tutorial
Tutorial on Bayesian Optimization for automating scientific discovery using surrogate models and probability-driven frameworks.
4/6/2026
annbatch unlocks terabyte-scale training of biological data in anndata
annbatch: mini-batch loader for terabyte-scale biological data in AnnData format, addressing memory bottlenecks in ML training on large datasets.
4/6/2026
Central Limit Theorems for Stochastic Gradient Descent Quantile Estimators
arXiv paper developing asymptotic theory for quantile estimation via stochastic gradient descent with constant learning rate.
4/6/2026
gen2seg: Generative Models Enable Generalizable Instance Segmentation
arXiv paper on gen2seg: using generative models (Stable Diffusion, MAE) for category-agnostic instance segmentation.
4/6/2026
LMask: Learn to Solve Constrained Routing Problems with Lazy Masking
arXiv paper proposing LMask, a learning framework using dynamic masking for constrained routing problems optimization.
4/6/2026
Are Statistical Methods Obsolete in the Era of Deep Learning? A Study of ODE Inverse Problems
arXiv paper comparing deep learning neural networks against statistical methods for solving ODE inverse problems.
4/6/2026
LLM Analysis of 150+ years of German Parliamentary Debates on Migration Reveals Shift from Post-War Solidarity to Anti-Solidarity in the Last Decade
arXiv paper analyzing 150+ years of German parliamentary migration debates using LLMs, revealing shift from post-war solidarity to anti-solidarity.
4/6/2026
ROPA: Synthetic Robot Pose Generation for RGB-D Bimanual Data Augmentation
arXiv paper on ROPA: synthetic robot pose generation for RGB-D bimanual data augmentation to improve imitation learning policies.
4/6/2026
Patterns behind Chaos: Forecasting Data Movement for Efficient Large-Scale MoE LLM Inference
Forecasting data movement patterns in MoE LLM inference to reduce bottlenecks in multi-unit serving systems.
4/6/2026
Seer: Online Context Learning for Fast Synchronous LLM Reinforcement Learning
Seer: context learning RL system for fast synchronous LLM training, addressing rollout latency and resource utilization.
4/6/2026
Investigating Test Overfitting on SWE-bench
Investigation of test overfitting in SWE-bench for code resolution, where models pass tests but miss important cases.
4/6/2026
Reward-Forcing: Autoregressive Video Generation with Reward Feedback
Autoregressive video generation using reward feedback to improve performance without strong teacher models.