2/19/2026
PolicyPad: Collaborative Prototyping of LLM Policies
PolicyPad system supporting collaborative policy design for LLMs in high-stakes domains via rapid prototyping and iteration.
2/19/2026
FeatBench: Towards More Realistic Evaluation of Feature-level Code Generation
FeatBench evaluates LLM code generation for realistic repository-level feature implementation with minimal data leakage.
2/19/2026
Predicting Training Re-evaluation Curves Enables Effective Data Curriculums for LLMs
Training re-evaluation curves diagnostic enabling better data curriculum design by characterizing batch retention across LLM training.
2/19/2026
Multilingual Routing in Mixture-of-Experts
Analysis of expert routing patterns in multilingual Mixture-of-Experts LLMs revealing language-specific dynamics across layers.
2/19/2026
StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars
StarEmbed benchmark for evaluating time series foundation models on irregular astronomical observations of variable stars.
2/19/2026
Lossless Vocabulary Reduction for Auto-Regressive Language Models
Technique for reducing LLM vocabulary size losslessly to improve auto-regressive text generation efficiency without performance loss.
2/19/2026
CreativityPrism: A Holistic Evaluation Framework for Large Language Model Creativity
CreativityPrism framework for holistic, scalable evaluation of LLM creativity across diverse scenarios without heavy human involvement.
2/19/2026
Transformers can do Bayesian Clustering
Cluster-PFN uses Transformers for unsupervised Bayesian clustering with uncertainty quantification, handling missing values.
2/19/2026
Q3R: Quadratic Reweighted Rank Regularizer for Effective Low-Rank Training
Q3R regularizer enabling parameter-efficient low-rank training and pre-training for large deep learning models.
2/19/2026
Reasoning Up the Instruction Ladder for Controllable Language Models
Method for enforcing instruction hierarchy in LLMs to handle competing directives from multiple sources for reliable decision-making.
2/19/2026
Mastering Olympiad-Level Physics with Artificial Intelligence
LOCA framework enabling AI agents to solve Olympiad-level physics problems via logical chain decomposition and verification.
2/19/2026
Language-Guided Invariance Probing of Vision-Language Models
Language-Guided Invariance Probing benchmark evaluating vision-language model robustness to paraphrases and semantic changes on 40k images.
2/19/2026
Mixture-of-Experts as Soft Clustering: A Dual Jacobian-PCA Spectral Geometry Perspective
Geometric analysis of Mixture-of-Experts architectures using Jacobian-PCA spectral methods to understand routing and function geometry.
2/19/2026
StableQAT: Stable Quantization-Aware Training at Ultra-Low Bitwidths
StableQAT framework for stable quantization-aware training of large models at ultra-low bitwidths for efficient deployment.
2/19/2026
Cardinality-Preserving Attention Channels for Graph Transformers in Molecular Property Prediction
Graph transformer architecture with cardinality-preserving attention for molecular property prediction in drug discovery.
2/19/2026
Do Vision-Language Models Respect Contextual Integrity in Location Disclosure?
Privacy risks of vision-language models inferring sensitive locations from photos with street-level precision.
2/19/2026
Vision and Language: Novel Representations and Artificial intelligence for Driving Scene Safety Assessment and Autonomous Vehicle Planning
Vision-language models for autonomous vehicle safety assessment and planning, integrating VLM representations into perception, prediction, and planning pipelines.
2/19/2026
When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential Processing
Analysis of whether LLM self-referential language reflects internal computation or confabulation via Pull Methodology tracking vocabulary-activation correspondence.
2/19/2026
Scaling Verification Can Be More Effective than Scaling Policy Learning for Vision-Language-Action Alignment
Vision-Language-Action models improved via test-time verification scaling to reduce intention-action gap in robot instruction following, offering alternative to policy learning scaling.
2/19/2026
Knowledge-Based Design Requirements for Generative Social Robots in Higher Education
Framework for designing generative social robots using LLMs for educational tutoring, addressing hallucinations, overreliance, and privacy risks in responsible AI deployment.
2/19/2026
Semantic Chunking and the Entropy of Natural Language
Statistical model capturing multi-scale structure of natural language relating entropy rate to semantic chunking in LLMs.
2/19/2026
Learning to Select Like Humans: Explainable Active Learning for Medical Imaging
Active learning method for medical imaging that explains selection decisions based on clinically meaningful features.
2/19/2026
Arming Data Agents with Tribal Knowledge
Framework arming NL2SQL agents with database-specific tribal knowledge to improve translation accuracy on real-world databases.
2/19/2026
Does Socialization Emerge in AI Agent Society? A Case Study of Moltbook
Case study analyzing emergence of social dynamics in AI agent societies through Moltbook's open-ended multi-agent environment.
2/19/2026
A Geometric Analysis of Small-sized Language Model Hallucinations
Geometric analysis showing hallucinations in small LLMs exhibit looser clustering than genuine responses in embedding space.
2/19/2026
Indic-TunedLens: Interpreting Multilingual Models in Indian Languages
Indic-TunedLens: interpretability framework for multilingual LLMs in Indian languages with shared affine transformations.
2/19/2026
Weight space Detection of Backdoors in LoRA Adapters
Weight-space detection method for backdoor attacks in LoRA adapters without requiring execution or knowing trigger patterns.
2/19/2026
Closing the Distribution Gap in Adversarial Training for LLMs
Method closing distribution gap in adversarial training for LLMs to improve robustness against simple in-distribution exploits.
2/19/2026
High-Fidelity Network Management for Federated AI-as-a-Service: Cross-Domain Orchestration
Cross-domain orchestration framework for managing federated AI-as-a-Service deployments with network-compute integration.
2/19/2026
AI-Paging: Lease-Based Execution Anchoring for Network-Exposed AI-as-a-Service
AI-Paging system for runtime selection and execution of AIaaS model instances via network-based intent matching.
2/19/2026
Far Out: Evaluating Language Models on Slang in Australian and Indian English
Evaluation of slang comprehension in state-of-the-art LLMs for Indian and Australian English varieties.
2/19/2026
SecCodeBench-V2 Technical Report
SecCodeBench-V2: benchmark of 98 scenarios evaluating LLM code generation security across 5 languages and 22 CWE categories.
2/19/2026
STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens
STAPO method stabilizing RL fine-tuning for LLMs by controlling rare spurious token gradients to prevent training collapse.
2/19/2026
A Content-Based Framework for Cybersecurity Refusal Decisions in Large Language Models
Content-based framework for consistent refusal decisions in LLM-based cybersecurity agents avoiding over-restriction and brittleness.
2/19/2026
B-DENSE: Branching For Dense Ensemble Network Learning
B-DENSE proposes branching approach to improve diffusion model inference speed while preserving structural information from intermediate trajectory steps.
2/19/2026
Verifier-Constrained Flow Expansion for Discovery Beyond the Data
Flow model expansion method using verifier constraints for scientific discovery beyond training data distribution in molecular design space.
2/19/2026
Anatomy of Capability Emergence: Scale-Invariant Representation Collapse and Top-Down Reorganization in Neural Networks
Mechanistic study of capability emergence tracking geometric measures across model scales, revealing scale-invariant representation collapse during training.
2/19/2026
AI-CARE: Carbon-Aware Reporting Evaluation Metric for AI Models
Carbon-aware evaluation metric for AI models that incorporates energy consumption and carbon emissions alongside traditional performance benchmarks.
2/19/2026
MoE-Spec: Expert Budgeting for Efficient Speculative Decoding
Expert budgeting optimization for efficient speculative decoding in Mixture-of-Experts LLMs to reduce memory pressure and maintain speedup.
2/19/2026
Multi-Objective Alignment of Language Models for Personalized Psychotherapy
Multi-objective alignment method for LLMs in psychotherapy applications balancing patient preferences with clinical safety using preference rankings.
2/19/2026
Can Generative Artificial Intelligence Survive Data Contamination? Theoretical Guarantees under Contaminated Recursive Training
Theoretical analysis of generative AI robustness under data contamination from AI-generated content in recursive training, with guarantees on model survival.
2/19/2026
On the Power of Source Screening for Learning Shared Feature Extractors
Investigation of source screening for learning shared feature extractors across heterogeneous data sources to filter low-quality or irrelevant data.
2/19/2026
Investigating GNN Convergence on Large Randomly Generated Graphs with Realistic Node Feature Correlations
Analysis of graph neural network convergence on large random graphs with correlated node features to assess GNN expressiveness in realistic settings.
2/19/2026
HiPER: Hierarchical Reinforcement Learning with Explicit Credit Assignment for Large Language Model Agents
Hierarchical reinforcement learning framework for training LLM agents on long-horizon tasks with sparse rewards using explicit credit assignment across action hierarchies.
2/19/2026
Muon with Spectral Guidance: Efficient Optimization for Scientific Machine Learning
Optimization method using orthogonalized updates for physics-informed neural networks and neural operators to handle ill-conditioned gradients and multi-scale behavior.
2/19/2026
Discrete Stochastic Localization for Non-autoregressive Generation
Non-autoregressive generation using masked diffusion language models with remasking samplers to reduce decoding latency while addressing error accumulation in iterative refinement.
2/19/2026
Deep TPC: Temporal-Prior Conditioning for Time Series Forecasting
Temporal-Prior Conditioning method for time series forecasting with LLMs, treating time as first-class modality across model depths.
2/19/2026
Graphon Mean-Field Subsampling for Cooperative Heterogeneous Multi-Agent Reinforcement Learning
Graphon-based mean-field method for multi-agent reinforcement learning with heterogeneous agents and computational efficiency improvements.
2/19/2026
Training-Free Adaptation of Diffusion Models via Doob's $h$-Transform
Training-free adaptation method for diffusion models using Doob's h-transform without additional training or differentiability assumptions.
2/19/2026
Linked Data Classification using Neurochaos Learning
Neurochaos Learning applied to linked data classification, demonstrating small-sample learning and low compute requirements on graph-structured data.