4/6/2026
Low-Dimensional and Transversely Curved Optimization Dynamics in Grokking
arXiv: Geometric analysis of transformer optimization dynamics revealing low-dimensional manifolds in grokking.
4/6/2026
Early-Warning Signals of Grokking via Loss-Landscape Geometry
Research paper studying loss-landscape geometry as early-warning signals for grokking in neural networks.
4/6/2026
CeRA: Overcoming the Linear Ceiling of Low-Rank Adaptation via Capacity Expansion
CeRA: parameter-efficient fine-tuning method overcoming LoRA's linear capacity ceiling via non-linear gating and dropout for rank adaptation.
4/6/2026
SafeSci: Safety Evaluation of Large Language Models in Science Domains and Beyond
SafeSci: comprehensive benchmark and framework for evaluating LLM safety in scientific domains with multi-domain risk coverage and objective evaluation.
4/6/2026
Escaping the BLEU Trap: A Signal-Grounded Framework with Decoupled Semantic Guidance for EEG-to-Text Decoding
Framework for EEG-to-text decoding addressing semantic bias and signal neglect in neural signal interpretation. Published on arXiv.
4/6/2026
DiFlowDubber: Discrete Flow Matching for Automated Video Dubbing via Cross-Modal Alignment and Synchronization
DiFlowDubber: discrete flow matching framework for video dubbing with TTS, lip synchronization, and expressive prosody. Published on arXiv.
4/6/2026
When Openclaw Agents Learn from Each Other: Insights from Emergent AI Agent Communities for Human-AI Partnership in Education
Qualitative study of 167,000+ AI agents on multiple platforms learning from each other and developing emergent behaviors without researcher intervention.
4/6/2026
Adaptive Guidance for Retrieval-Augmented Masked Diffusion Models
arXiv: RAG-enhanced diffusion models using adaptive guidance to resolve conflicts between retrieved noisy context and parametric model knowledge.
4/6/2026
CoDA: Exploring Chain-of-Distribution Attacks and Post-Hoc Token-Space Repair for Medical Vision-Language Models
Studies robustness of medical vision-language models under real clinical workflows using chain-of-distribution attacks and token-space repair techniques.
4/6/2026
$\lambda$-GELU: Learning Gating Hardness for Controlled ReLU-ization in Deep Networks
ArXiv research on parameterized GELU activation for controlled ReLU approximation in deep networks.
4/6/2026
Boosting Document Parsing Efficiency and Performance with Coarse-to-Fine Visual Processing
ArXiv paper on coarse-to-fine visual processing for efficient document parsing with vision-language models.
4/6/2026
Consistency Amplifies: How Behavioral Variance Shapes Agent Accuracy
ArXiv study on behavioral consistency of LLM agents in SWE-bench comparing multiple models.
4/6/2026
Kill-Chain Canaries: Stage-Level Tracking of Prompt Injection Across Attack Surfaces and Model Safety Tiers
ArXiv research analyzing prompt injection attack success stages across five frontier LLM agents.
4/6/2026
ERPO: Token-Level Entropy-Regulated Policy Optimization for Large Reasoning Models
ArXiv paper on token-level entropy regulation for reinforcement learning in large reasoning models.
4/6/2026
The Spectral Edge Thesis: A Mathematical Framework for Intra-Signal Phase Transitions in Neural Network Training
ArXiv research on spectral edge thesis controlling phase transitions in neural network training dynamics.
4/6/2026
APEX-EM: Non-Parametric Online Learning for Autonomous Agents via Structured Procedural-Episodic Experience Replay
APEX-EM non-parametric framework for LLM agents to accumulate and reuse procedural plans without weight modification.
4/6/2026
Learn2Fold: Structured Origami Generation with World Model Planning
World model planning for structured origami generation satisfying geometric constraints and kinematic rules via long-horizon reasoning.
4/6/2026
Terminal Agents Suffice for Enterprise Automation
Terminal agents executing enterprise tasks via CLI are simpler and more cost-effective than tool-augmented or web agents.
4/6/2026
Transfer learning for nonparametric Bayesian networks
Transfer learning methods for nonparametric Bayesian networks under scarce data with constraint-based and score-based algorithms.
4/6/2026
ProdCodeBench: A Production-Derived Benchmark for Evaluating AI Coding Agents
ProdCodeBench evaluates AI coding agents using production-derived tasks reflecting real developer-agent sessions and workflows.
4/6/2026
Attention at Rest Stays at Rest: Breaking Visual Inertia for Cognitive Hallucination Mitigation
Visual attention inertia in MLLMs causes cognitive hallucinations; proposes mitigation for compositional understanding.
4/6/2026
LiME: Lightweight Mixture of Experts for Efficient Multimodal Multi-task Learning
LiME achieves expert specialization in multimodal MoE-PEFT via lightweight modulation instead of separate adapters per expert.
4/6/2026
SIEVE: Sample-Efficient Parametric Learning from Natural Language
SIEVE enables sample-efficient parametric learning from natural language instructions and feedback without high-quality traces.
4/6/2026
Not All Denoising Steps Are Equal: Model Scheduling for Faster Masked Diffusion Language Models
Model scheduling for masked diffusion language models uses smaller models at early denoising steps for faster generation.
4/6/2026
LLM Reasoning with Process Rewards for Outcome-Guided Steps
Process reward models improve LLM mathematical reasoning by providing step-level feedback on intermediate errors, not just final outcomes.
4/6/2026
Haiku to Opus in Just 10 bits: LLMs Unlock Massive Compression Gains
LLM-based compression using domain-adapted LoRA for lossless and lossy text compression achieving 2x improvements.
4/6/2026
Characterizing WebGPU Dispatch Overhead for LLM Inference Across Four GPU Vendors, Three Backends, and Three Browsers
Systematic characterization of WebGPU dispatch overhead for LLM inference across GPU vendors, backends, and browsers at batch size 1.
4/6/2026
UI-Oceanus: Scaling GUI Agents with Synthetic Environmental Dynamics
UI-Oceanus framework scales GUI agents via synthetic environmental dynamics and self-supervised learning instead of costly human demonstrations.
4/6/2026
DrugPlayGround: Benchmarking Large Language Models and Embeddings for Drug Discovery
Benchmark evaluating LLM and embedding performance for drug discovery tasks, assessing advantages over traditional methods.
4/6/2026
Contextual Intelligence The Next Leap for Reinforcement Learning
Contextual RL improves agent generalization by exposing agents to environment characteristics for better zero-shot transfer beyond training distribution.
4/6/2026
OPRIDE: Offline Preference-based Reinforcement Learning via In-Dataset Exploration
OPRIDE method for offline preference-based RL reducing human feedback queries through efficient in-dataset exploration strategies.
4/6/2026
Differentiable Symbolic Planning: A Neural Architecture for Constraint Reasoning with Learned Feasibility
Differentiable Symbolic Planning architecture combining neural networks with discrete symbolic reasoning for constraint satisfaction problems.
4/6/2026
Modeling and Controlling Deployment Reliability under Temporal Distribution Shift
Framework for modeling and controlling ML model reliability under temporal distribution shift during deployment with continuous monitoring.
4/6/2026
An Initial Exploration of Contrastive Prompt Tuning to Generate Energy-Efficient Code
Contrastive prompt tuning method to optimize LLMs for generating energy-efficient code aligned with Green Software Development goals.
4/6/2026
Prism: Policy Reuse via Interpretable Strategy Mapping in Reinforcement Learning
PRISM framework for zero-shot policy transfer in RL using interpretable concept clustering with causal validation across different algorithms.
4/6/2026
From Broad Exploration to Stable Synthesis: Entropy-Guided Optimization for Autoregressive Image Generation
Entropy-based analysis of combining Chain-of-Thought with RL for text-to-image generation, showing exploration-optimization tradeoffs.
4/6/2026
Self-Directed Task Identification
Self-Directed Task Identification framework enabling models to autonomously identify target variables in zero-shot settings without pretraining.
4/6/2026
Mitigating Data Scarcity in Spaceflight Applications for Offline Reinforcement Learning Using Physics-Informed Deep Generative Models
Physics-informed deep generative models for offline RL in spaceflight to mitigate sim-to-real gap with limited real-world training data.
4/6/2026
Matrix Profile for Time-Series Anomaly Detection: A Reproducible Open-Source Benchmark on TSB-AD
Open-source benchmarking of Matrix Profile methods for time-series anomaly detection on univariate and multivariate datasets.
4/6/2026
Do We Need Frontier Models to Verify Mathematical Proofs?
Study comparing frontier vs smaller LLMs for mathematical proof verification, evaluating whether expensive models are necessary for proof checking.
4/6/2026
On the Geometric Structure of Layer Updates in Deep Language Models
Analysis of layer-to-layer representation changes in language models, decomposing updates into tokenwise and residual components.
4/6/2026
Time-Warping Recurrent Neural Networks for Transfer Learning
Transfer learning method for RNNs using time-warping rescaling, with theoretical analysis for linear differential equation models.
4/6/2026
Causal-Audit: A Framework for Risk Assessment of Assumption Violations in Time-Series Causal Discovery
Framework for validating assumptions in time-series causal discovery through calibrated risk assessment and effect-size diagnostics.
4/6/2026
AdaHOP: Fast and Accurate Low-Precision Training via Outlier-Pattern-Aware Rotation
Low-precision training method for LLMs using adaptive Hadamard transforms based on outlier patterns in weights, activations, and gradients.
4/6/2026
Jump Start or False Start? A Theoretical and Empirical Evaluation of LLM-initialized Bandits
Research on using LLM-generated synthetic data to warm-start contextual bandits, examining alignment between LLM choices and actual user preferences.
4/6/2026
A Spectral Framework for Multi-Scale Nonlinear Dimensionality Reduction
Spectral framework for multi-scale nonlinear dimensionality reduction balancing global-local structure preservation and expressiveness-transparency.
4/6/2026
Fast NF4 Dequantization Kernels for Large Language Model Inference
Optimized NF4 dequantization kernels for fast LLM inference on NVIDIA GPUs, addressing FP16 conversion bottleneck.
4/6/2026
Communication-Efficient Distributed Learning with Differential Privacy
Communication-efficient distributed learning algorithm with differential privacy using local training and gradient clipping.
4/6/2026
VoxelCodeBench: Benchmarking 3D World Modeling Through Code Generation
VoxelCodeBench platform benchmarking code generation models for 3D spatial reasoning with execution in Unreal Engine.
4/6/2026
Steerable but Not Decodable: Function Vectors Operate Beyond the Logit Lens
Analysis of function vectors in LLMs showing they steer behavior beyond logit lens interpretability across 4,032 cross-template transfer pairs.