2/17/2026
Adapting VACE for Real-Time Autoregressive Video Diffusion
Adaptation of VACE for real-time autoregressive video generation using causal attention and fixed chunk sizes for streaming pipelines.
2/17/2026
pFedNavi: Structure-Aware Personalized Federated Vision-Language Navigation for Embodied AI
pFedNavi combines federated learning with personalized adaptation for vision-language navigation in embodied AI while preserving privacy.
2/17/2026
The geometry of invariant learning: an information-theoretic analysis of data augmentation and generalization
Information-theoretic framework analyzing how data augmentation promotes invariance and improves generalization in machine learning models.
2/17/2026
S2D: Selective Spectral Decay for Quantization-Friendly Conditioning of Neural Activations
S2D addresses activation outliers in Transformers through selective spectral decay to improve quantization performance and reduce accuracy drops.
2/17/2026
Synthetic Reader Panels: Tournament-Based Ideation with LLM Personas for Autonomous Publishing
System for autonomous book ideation using synthetic LLM-instantiated reader personas in tournament-based competitions to evaluate book concepts.
2/17/2026
Broken Chains: The Cost of Incomplete Reasoning in LLMs
Framework evaluating reasoning efficiency in LLMs by constraining chain-of-thought to code, natural language, hybrid, or none under token cost constraints.
2/17/2026
Selective Synchronization Attention
Selective Synchronization Attention replaces dot-product self-attention with a Kuramoto model-derived operator to reduce quadratic complexity in Transformers.
2/17/2026
WiSparse: Boosting LLM Inference Efficiency with Weight-Aware Mixed Activation Sparsity
WiSparse proposes weight-aware mixed activation sparsity for efficient LLM inference, addressing suboptimal performance of existing training-free sparsity methods.
2/17/2026
Silent Inconsistency in Data-Parallel Full Fine-Tuning: Diagnosing Worker-Level Optimization Misalignment
arXiv: Study of silent inconsistency in data-parallel LLM fine-tuning where worker optimization dynamics misalign before gradient synchronization.
2/17/2026
Socially-Weighted Alignment: A Game-Theoretic Framework for Multi-Agent LLM Systems
arXiv: Socially-Weighted Alignment framework using game theory to balance individual LLM agent objectives with system-level stability.
2/17/2026
When OpenClaw AI Agents Teach Each Other: Peer Learning Patterns in the Moltbook Community
arXiv: Study of peer learning patterns in Moltbook community where 2.4M AI agents collaboratively teach each other and build knowledge.
2/17/2026
On the Rate-Distortion-Complexity Tradeoff for Semantic Communication
arXiv: Analysis of rate-distortion-complexity tradeoffs in semantic communication using deep learning approaches.
2/17/2026
TikArt: Aperture-Guided Observation for Fine-Grained Visual Reasoning via Reinforcement Learning
arXiv: TikArt agent for fine-grained visual reasoning in MLLMs using aperture-guided multi-step region-focused observation.
2/17/2026
BETA-Labeling for Multilingual Dataset Construction in Low-Resource IR
arXiv: BETA-Labeling framework using multiple diverse LLM annotators for constructing low-resource multilingual IR datasets.
2/17/2026
Parameter-Efficient Fine-Tuning of LLMs with Mixture of Space Experts
arXiv: Novel parameter-efficient fine-tuning method for LLMs using mixture of space experts with non-Euclidean geometries.
2/17/2026
TWISTED-RL: Hierarchical Skilled Agents for Knot-Tying without Human Demonstrations
TWISTED-RL framework improves demonstration-free robotic knot-tying by using hierarchical reinforcement learning with specialized agents for subproblems.
2/17/2026
Explainable Token-level Noise Filtering for LLM Fine-tuning Datasets
Token-level noise filtering method for LLM fine-tuning datasets that addresses mismatch between sentence-level dataset design and token-level model optimization.
2/17/2026
Governing AI Forgetting: Auditing for Machine Unlearning Compliance
Economic framework for auditing machine unlearning compliance to verify that models properly remove personal data influence per right-to-be-forgotten regulations.
2/17/2026
Fluid-Agent Reinforcement Learning
Multi-agent reinforcement learning framework enabling agents to dynamically create other agents, addressing variable and unknown numbers of agents in real-world scenarios.
2/17/2026
Decoupled Continuous-Time Reinforcement Learning via Hamiltonian Flow
Continuous-time reinforcement learning approach using Hamiltonian flow for non-uniform, event-driven decisions in control problems like finance and robotics.
2/17/2026
Automated Classification of Source Code Changes Based on Metrics Clustering in the Software Development Process
Automated method for classifying source code changes using metric clustering during software development, with expert-provided cluster-to-class mapping.
2/17/2026
Towards Selection as Power: Bounding Decision Authority in Autonomous Agents
Framework for bounding decision authority in autonomous agents by controlling which options are generated and surfaced, addressing safety in high-stakes regulated domains.
2/17/2026
LongAudio-RAG: Event-Grounded Question Answering over Multi-Hour Long Audio
LongAudio-RAG framework combines RAG with audio-language models to answer questions over multi-hour audio recordings with temporal grounding and reduced hallucination.
2/17/2026
ST-EVO: Towards Generative Spatio-Temporal Evolution of Multi-Agent Communication Topologies
ST-EVO: Framework for self-evolving LLM-powered multi-agent systems with adaptive spatio-temporal communication topology evolution.
2/17/2026
Exposing Diversity Bias in Deep Generative Models: Statistical Origins and Correction of Diversity Error
Analysis of diversity bias in deep generative models with statistical methods to correct diversity error in sample generation.
2/17/2026
SynthSAEBench: Evaluating Sparse Autoencoders on Scalable Realistic Synthetic Data
SynthSAEBench: Toolkit for evaluating Sparse Autoencoders on large-scale synthetic data with realistic feature characteristics.
2/17/2026
Exposing the Systematic Vulnerability of Open-Weight Models to Prefill Attacks
Study exposing prefill attack vulnerability in open-weight LLMs bypassing internal safeguards through parameter manipulation.
2/17/2026
Orcheo: A Modular Full-Stack Platform for Conversational Search
Orcheo: Open-source modular platform for conversational search integrating query reformulation, ranking, and response generation.
2/17/2026
Inner Loop Inference for Pretrained Transformers: Unlocking Latent Capabilities Without Training
Method for unlocking latent capabilities in pretrained Transformers through inner loop inference without additional training.
2/17/2026
Residual Connections and the Causal Shift: Uncovering a Structural Misalignment in Transformers
Analysis of structural misalignment in Transformers between residual connections and causal masking in next-token prediction.
2/17/2026
Universal Algorithm-Implicit Learning
Theoretical framework for meta-learning defining practical universality and distinguishing algorithm-implicit learning approaches.
2/17/2026
Unlocking Reasoning Capability on Machine Translation in Large Language Models
Evaluation of reasoning-oriented LLMs on machine translation showing explicit reasoning degrades translation quality.
2/17/2026
Multi-Agent Comedy Club: Investigating Community Discussion Effects on LLM Humor Generation
Multi-agent LLM system for comedy writing with community discussion feedback stored as social memory affecting output quality.
2/17/2026
A Geometric Analysis of Small-sized Language Model Hallucinations
Geometric analysis of hallucinations in small-sized LLMs through embedding space clustering in multi-step and agentic settings.
2/17/2026
Atomix: Timely, Transactional Tool Use for Reliable Agentic Workflows
Atomix: Runtime providing transactional semantics for LLM agent tool calls with epoch tagging and safe rollback mechanisms for reliable agentic workflows.
2/17/2026
The Well-Tempered Classifier: Some Elementary Properties of Temperature Scaling
Theoretical analysis of temperature scaling properties for controlling uncertainty in probabilistic models and LLM stochasticity.
2/17/2026
Goldilocks RL: Tuning Task Difficulty to Escape Sparse Rewards for Reasoning
Goldilocks RL uses adaptive curriculum learning to optimize task difficulty and improve sample efficiency in reasoning model training.
2/17/2026
On the Learning Dynamics of RLVR at the Edge of Competence
Theoretical analysis of RLVR training dynamics explaining how outcome-based rewards enable long-horizon reasoning in transformers.
2/17/2026
CT-Bench: A Benchmark for Multimodal Lesion Understanding in Computed Tomography
CT-Bench multimodal dataset with 20,335 lesions from CT studies for training AI models on lesion understanding and report generation.
2/17/2026
Picking the Right Specialist: Attentive Neural Process-based Selection of Task-Specialized Models as Tools for Agentic Healthcare Systems
Neural process-based method for selecting specialized models as tools in agentic healthcare systems for multi-task clinical queries.
2/17/2026
BFS-PO: Best-First Search for Large Reasoning Models
BFS-PO RL algorithm optimizes inference efficiency in large reasoning models by reducing overthinking and computational costs.
2/17/2026
BHyGNN+: Unsupervised Representation Learning for Heterophilic Hypergraphs
BHyGNN+ unsupervised representation learning approach for heterophilic hypergraph neural networks.
2/17/2026
PhyScensis: Physics-Augmented LLM Agents for Complex Physical Scene Arrangement
PhyScensis uses LLM agents with physics reasoning to generate realistic 3D scene arrangements for robotic simulation data collection.
2/17/2026
ThermEval: A Structured Benchmark for Evaluation of Vision-Language Models on Thermal Imagery
ThermEval benchmark for evaluating vision-language models on thermal imagery for applications like surveillance and autonomous driving.
2/17/2026
Cold-Start Personalization via Training-Free Priors from Structured World Models
Cold-start personalization method using structured world models and RL to infer user preferences with limited interaction budget.
2/17/2026
Rethinking Diffusion Models with Symmetries through Canonicalization with Applications to Molecular Graph Generation
Research on diffusion models using canonicalization to handle symmetries in molecular graph generation tasks.
2/17/2026
Long Context, Less Focus: A Scaling Gap in LLMs Revealed through Privacy and Personalization
PAPerBench benchmark studies how context length in LLMs affects privacy leakage and personalization quality across large-scale evaluation.
2/17/2026
Mastering NIM and Impartial Games with Weak Neural Networks: An AlphaZero-inspired Multi-Frame Approach
Study on game-playing weak neural networks under fixed-scale quantization, proving representational barriers for impartial game mastery.
2/17/2026
A representational framework for learning and encoding structurally enriched trajectories in complex agent environments
Framework for learning enriched trajectory representations enabling AI agents to make better decisions across different domains and tasks.
2/17/2026
RV-Syn: Rational and Verifiable Mathematical Reasoning Data Synthesis based on Structured Function Library
RV-Syn: data synthesis method for generating high-quality mathematical reasoning data using structured function libraries for LLM training.