2/18/2026
FedPSA: Modeling Behavioral Staleness in Asynchronous Federated Learning
Addresses staleness challenges in asynchronous federated learning with behavioral modeling.
2/18/2026
Fine-Tuning LLMs to Generate Economical and Reliable Actions for the Power Grid
Fine-tunes LLMs to generate corrective transmission switching actions for power grid operations.
2/18/2026
NeuroSymActive: Differentiable Neural-Symbolic Reasoning with Active Exploration for Knowledge Graph Question Answering
Integrates neural networks with symbolic reasoning for knowledge graph question answering using active exploration.
2/18/2026
CDRL: A Reinforcement Learning Framework Inspired by Cerebellar Circuits and Dendritic Computational Strategies
RL architecture inspired by cerebellar circuits to improve sample efficiency, noise robustness and generalization.
2/18/2026
GMAIL: Generative Modality Alignment for generated Image Learning
Framework addressing modality discrepancies between synthetic and real images for training machine learning models.
2/18/2026
Far Out: Evaluating Language Models on Slang in Australian and Indian English
Evaluates LLM performance on slang in Australian and Indian English across 7 models using new datasets.
2/18/2026
Orchestration-Free Customer Service Automation: A Privacy-Preserving and Flowchart-Guided Framework
Framework for orchestration-free customer service automation using Task-Oriented Flowcharts to guide end-to-end LLM-based workflows without manual intervention.
2/18/2026
ActionCodec: What Makes for Good Action Tokenizers
Study of action tokenization design for Vision-Language-Action models, examining what makes effective action tokenizers for VLA optimization.
2/18/2026
Logit Distance Bounds Representational Similarity
Mathematical analysis of representational similarity in discriminative models including autoregressive language models using logit distance bounds.
2/18/2026
Molecular Design beyond Training Data with Novel Extended Objective Functionals of Generative AI Models Driven by Quantum Annealing Computer
Framework combining deep generative models with quantum annealing for molecular design beyond training data distribution.
2/18/2026
SecCodeBench-V2 Technical Report
SecCodeBench-V2 benchmark for evaluating LLM code generation security across 98 scenarios and 22 CWE categories in 5 languages.
2/18/2026
Improving MLLMs in Embodied Exploration and Question Answering with Human-Inspired Memory Modeling
Memory framework for embodied AI agents using episodic-semantic separation to improve long-horizon question answering and exploration.
2/18/2026
Dynamic Training-Free Fusion of Subject and Style LoRAs
Training-free dynamic fusion framework for combining multiple LoRAs to generate subjects and styles in image generation.
2/18/2026
VLM-DEWM: Dynamic External World Model for Verifiable and Resilient Vision-Language Planning in Manufacturing
VLM-DEWM cognitive architecture for vision-language planning in manufacturing with persistent state tracking and interpretable reasoning.
2/18/2026
Beyond Static Pipelines: Learning Dynamic Workflows for Text-to-SQL
Dynamic workflow learning system for Text-to-SQL that adapts execution paths at inference time instead of static pipelines.
2/18/2026
STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens
STAPO method stabilizing RL fine-tuning for LLMs by filtering spurious tokens to prevent late-stage training collapse.
2/18/2026
Zombie Agents: Persistent Control of Self-Evolving LLM Agents via Self-Reinforcing Injections
Security analysis of self-evolving LLM agents where persistent memory can be exploited for covert payload injection attacks.
2/18/2026
Bayesian Optimization for Design Parameters of 3D Image Data Analysis
Bayesian optimization pipeline for selecting and tuning deep learning models for 3D biomedical image segmentation and classification.
2/18/2026
Relative Geometry of Neural Forecasters: Linking Accuracy and Alignment in Learned Latent Geometry
Framework studying how neural networks represent latent geometry in dynamical systems through representational alignment analysis.
2/18/2026
A Content-Based Framework for Cybersecurity Refusal Decisions in Large Language Models
Content-based framework for cybersecurity refusal decisions in LLMs and agents, addressing consistency and brittleness issues.
2/18/2026
Lifelong Scalable Multi-Agent Realistic Testbed and A Comprehensive Study on Design Choices in Lifelong AGV Fleet Management Systems
Open-source simulator (LSMART) for evaluating multi-agent path finding algorithms in fleet management with automated guided vehicles.
2/18/2026
Learning to Retrieve Navigable Candidates for Efficient Vision-and-Language Navigation
LLM-based navigation agent for vision-and-language tasks that learns to retrieve navigable candidates efficiently in unseen environments.
2/18/2026
ChartEditBench: Evaluating Grounded Multi-Turn Chart Editing in Multimodal Language Models
Benchmark for evaluating multi-turn chart editing in multimodal language models with iterative refinement.
2/18/2026
Understanding vs. Generation: Navigating Optimization Dilemma in Multimodal Models
Framework addressing trade-offs between generative and understanding capabilities in multimodal models via Reason-Reflect-Refine algorithm.
2/18/2026
The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safety
Analysis showing fine-tuning causes alignment collapse in language models and proposing geometric understanding of safety degradation.
2/18/2026
Decision Quality Evaluation Framework at Pinterest
Decision quality evaluation framework for assessing content moderation by both human agents and LLMs at scale.
2/18/2026
Avey-B
Encoder-only adaptation of Avey attention-free architecture for efficient NLP under compute constraints.
2/18/2026
CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing
Scalable non-destructive LLM editing method using second-order constraints to preserve capabilities while modifying target behaviors.
2/18/2026
From Prompts to Protection: Large Language Model-Enabled In-Context Learning for Smart Public Safety UAV
LLM-based in-context learning for optimizing public safety UAV navigation and control during emergency response.
2/18/2026
Prover Agent: An Agent-Based Framework for Formal Mathematical Proofs
AI agent framework integrating LLMs with Lean formal proof assistant for automated theorem proving with auxiliary lemma generation.
2/18/2026
OpenAgentSafety: A Comprehensive Framework for Evaluating Real-World AI Agent Safety
Comprehensive safety evaluation framework for real-world AI agents across diverse task domains and tool interactions.
2/18/2026
FRSICL: LLM-Enabled In-Context Learning Flight Resource Allocation for Fresh Data Collection in UAV-Assisted Wildfire Monitoring
LLM-based in-context learning approach for UAV resource allocation in wildfire monitoring systems.
2/18/2026
MultiSHAP: A Shapley-Based Framework for Explaining Cross-Modal Interactions in Multimodal AI Models
Shapley-based explanation framework for understanding cross-modal interactions in multimodal AI models.
2/18/2026
Learning When to Plan: Efficiently Allocating Test-Time Compute for LLM Agents
Framework for LLM agents to learn when to plan via reinforcement learning, improving efficiency and performance on long-horizon tasks.
2/18/2026
A Fano-Style Accuracy Upper Bound for LLM Single-Pass Reasoning in Multi-Hop QA
Theoretical analysis establishing accuracy upper bounds for LLM single-pass reasoning in multi-hop question answering, formalizing capacity overflow bottlenecks.
2/18/2026
Toward Safer Diffusion Language Models: Discovery and Mitigation of Priming Vulnerability
Identifies and mitigates priming vulnerability in diffusion language models exploited by jailbreak attacks during iterative denoising inference.
2/18/2026
Generalized Parallel Scaling with Interdependent Generations
Generalized parallel LLM inference scaling uses interdependent generations to share information across parallel responses, improving quality and efficiency.
2/18/2026
SR-Scientist: Scientific Equation Discovery With Agentic AI
SR-Scientist elevates LLMs from equation proposers to autonomous AI scientists for scientific discovery, writing and executing code for symbolic regression.
2/18/2026
Comparative Expressivity for Structured Argumentation Frameworks with Uncertain Rules and Premises
Studies expressivity of structured argumentation frameworks with uncertainty in rules and premises, advancing formal argumentation theory.
2/18/2026
Beyond Fact Retrieval: Episodic Memory for RAG with Generative Semantic Workspaces
Generative semantic workspaces enhance RAG with episodic memory for long-context reasoning, addressing context window limitations and sequence length degradation.
2/18/2026
Advanced Assistance for Traffic Crash Analysis: An AI-Driven Multi-Agent Approach to Pre-Crash Reconstruction
Multi-agent AI framework reconstructs pre-crash vehicle scenarios from fragmented collision data using collaborative reconstruction and reasoning stages.
2/18/2026
Chain of Summaries: Summarization Through Iterative Questioning
Chain of Summaries generates information-dense summaries of web content via iterative questioning, enabling LLMs to process diverse content formats within context limits.
2/18/2026
Aeon: High-Performance Neuro-Symbolic Memory Management for Long-Horizon LLM Agents
Aeon proposes neuro-symbolic memory management for long-horizon LLM agents, using hierarchical structures to overcome context window and attention limitations.
2/18/2026
ErrorMap and ErrorAtlas: Charting the Failure Landscape of Large Language Models
ErrorMap charts failure sources in LLM benchmarks, disentangling causes like formatting errors and calculation mistakes from reasoning deficiencies.
2/18/2026
PolySHAP: Extending KernelSHAP with Interaction-Informed Polynomial Regression
PolySHAP improves Shapley value computation for explainable AI using polynomial regression, reducing computational cost of KernelSHAP approximation.
2/18/2026
ScholarGym: Benchmarking Large Language Model Capabilities in the Information-Gathering Stage of Deep Research
ScholarGym benchmarks LLM capabilities in iterative research workflows, evaluating information-gathering stages of deep research systems beyond end-to-end metrics.
2/18/2026
FlowSteer: Interactive Agentic Workflow Orchestration via End-to-End Reinforcement Learning
FlowSteer: end-to-end reinforcement learning framework for agentic workflow orchestration. Learns policies for multi-step task automation with minimal manual specification.
2/18/2026
MARS: Modular Agent with Reflective Search for Automated AI Research
MARS: AI agent framework for automated machine learning research. Budget-aware planning, modular script generation, and reflective search for computationally expensive experiments.
2/18/2026
LQA: A Lightweight Quantized-Adaptive Framework for Vision-Language Models on the Edge
LQA: lightweight framework for deploying vision-language models on edge devices. Combines quantization with gradient-free test-time adaptation for distribution shift robustness.
2/18/2026
stable-worldmodel-v1: Reproducible World Modeling Research and Evaluation
stable-worldmodel-v1: reproducible open implementation of world models for agent reasoning and planning. Standardizes evaluation and reduces publication-specific code fragmentation.