2/17/2026
DTBench: A Synthetic Benchmark for Document-to-Table Extraction
Benchmark dataset and evaluation of LLM performance on document-to-table extraction with complex reasoning requirements.
2/17/2026
Evaluating LLM-Generated ACSL Annotations for Formal Verification
Empirical study of LLM capability to generate and verify formal ACSL specifications for C programs automatically.
2/17/2026
GSRM: Generative Speech Reward Model for Speech RLHF
Generative Speech Reward Model for evaluating and improving naturalness in speech language model outputs via RLHF.
2/17/2026
A Comparative Analysis of Social Network Topology in Reddit and Moltbook
Empirical comparison of AI agent-driven social network topology (Moltbook) versus human-driven (Reddit).
2/17/2026
GREPO: A Benchmark for Graph Neural Networks on Repository-Level Bug Localization
GREPO benchmark for GNNs on repository-level bug localization, addressing context limitations of standard LLMs.
2/17/2026
You Can Learn Tokenization End-to-End with Reinforcement Learning
End-to-end learned tokenization for LLMs using RL to optimize token boundaries instead of fixed hardcoded compression.
2/17/2026
Experiential Reinforcement Learning
Experiential RL training paradigm embedding explanatory chains for LMs to learn from sparse delayed environmental feedback.
2/17/2026
Eureka-Audio: Triggering Audio Intelligence in Compact Language Models
Eureka-Audio: 1.7B parameter compact audio language model matching performance of 7B-30B models on ASR and audio understanding.
2/17/2026
Chemical Language Models for Natural Products: A State-Space Model Approach
NP-specific chemical language models using Mamba state-space models for natural product property prediction and generation.
2/17/2026
WoVR: World Models as Reliable Simulators for Post-Training VLA Policies with RL
WoVR: framework using learned world models as simulators for RL-based post-training of Vision-Language-Action policies.
2/17/2026
The Sufficiency-Conciseness Trade-off in LLM Self-Explanation from an Information Bottleneck Perspective
Information-theoretic analysis of trade-off between explanation sufficiency and conciseness in LLM chain-of-thought reasoning.
2/17/2026
From SFT to RL: Demystifying the Post-Training Pipeline for LLM-based Vulnerability Detection
Comprehensive study on LLM post-training pipeline (SFT, preference optimization) applied to vulnerability detection in code.
2/17/2026
EIDOS: Latent-Space Predictive Learning for Time Series Foundation Models
EIDOS foundation model family for time series using latent-space predictive learning instead of direct future prediction.
2/17/2026
Beyond Static Snapshots: Dynamic Modeling and Forecasting of Group-Level Value Evolution with Large Language Models
LLM-based social simulation framework modeling dynamic group-level value evolution over time rather than static snapshots.
2/17/2026
Annotation-Efficient Vision-Language Model Adaptation to the Polish Language Using the LLaVA Framework
Adapts LLaVA vision-language model framework to Polish language using automated pipeline for annotation-efficient multilingual VLM development.
2/17/2026
Policy Gradient with Adaptive Entropy Annealing for Continual Fine-Tuning
Reinforcement learning approach using policy gradients with entropy annealing for parameter-efficient fine-tuning of vision models preventing catastrophic forgetting.
2/17/2026
Empty Shelves or Lost Keys? Recall Is the Bottleneck for Parametric Factuality
Framework analyzing LLM factuality by distinguishing knowledge encoding failures from retrieval failures at the fact level.
2/17/2026
TabTracer: Monte Carlo Tree Search for Complex Table Reasoning with Large Language Models
TabTracer applies Monte Carlo Tree Search with LLMs for complex table reasoning, enabling step-level verification and backtracking to reduce errors.
2/17/2026
Anticipating Adversary Behavior in DevSecOps Scenarios through Large Language Models
Uses LLMs to predict adversary behavior and anticipate cyberattacks in DevSecOps cloud environments.
2/17/2026
Toward Autonomous O-RAN: A Multi-Scale Agentic AI Framework for Real-Time Network Control and Management
Multi-scale agentic AI framework for autonomous O-RAN network control and management, handling complex multi-layer control loops in 6G systems.
2/17/2026
DenseMLLM: Standard Multimodal LLMs are Intrinsic Dense Predictors
DenseMLLM enables multimodal LLMs to perform dense prediction tasks like semantic segmentation and depth estimation without task-specific decoders.
2/17/2026
A Multi-Agent Framework for Medical AI: Leveraging Fine-Tuned GPT, LLaMA, and DeepSeek R1 for Evidence-Based and Bias-Aware Clinical Query Processing
Multi-agent framework combining fine-tuned GPT, LLaMA, and DeepSeek R1 with evidence retrieval and bias detection for medical question answering.
2/17/2026
Deep Dense Exploration for LLM Reinforcement Learning via Pivot-Driven Resampling
Introduces Pivot-Driven Resampling for exploration in LLM reinforcement learning, improving trajectory discovery within limited sampling budgets.
2/17/2026
UniWeTok: An Unified Binary Tokenizer with Codebook Size $\mathit{2^{128}}$ for Unified Multimodal Large Language Model
Proposes UniWeTok, a unified discrete tokenizer with 2^128 codebook for multimodal LLMs enabling high-fidelity reconstruction, semantic extraction, and generation.
2/17/2026
GPT-5 vs Other LLMs in Long Short-Context Performance
Evaluates context window utilization across LLMs including GPT-5, comparing theoretical capacity vs practical performance on long-context tasks requiring detailed understanding.
2/17/2026
Knowing When Not to Answer: Abstention-Aware Scientific Reasoning
Framework for LLMs to abstain from answering uncertain scientific claims, decomposing claims into verifiable conditions.
2/17/2026
GeoEyes: On-Demand Visual Focusing for Evidence-Grounded Understanding of Ultra-High-Resolution Remote Sensing Imagery
Multimodal framework enabling adaptive visual focusing in vision-language models for ultra-high-resolution remote sensing analysis.
2/17/2026
SkillJect: Automating Stealthy Skill-Based Prompt Injection for Coding Agents with Trace-Driven Closed-Loop Refinement
Attack methodology demonstrating skill-based prompt injection vulnerability in coding agents via trace-driven refinement.
2/17/2026
Reasoning Language Models for complex assessments tasks: Evaluating parental cooperation from child protection case reports
Multi-stage workflow using reasoning language models to assess parental cooperation in child protection case reports.
2/17/2026
Evaluating LLMs in Finance Requires Explicit Bias Consideration
Analysis identifying five recurring biases in financial LLM evaluations: look-ahead, survivorship, narrative, objective, and cost.
2/17/2026
Dual-Signal Adaptive KV-Cache Optimization for Long-Form Video Understanding in Vision-Language Models
Optimization technique for reducing KV-cache memory overhead in vision-language models processing long-form video.
2/17/2026
A Hybrid TGN-SEAL Model for Dynamic Graph Link Prediction
Hybrid GNN model combining temporal graph networks with graph structure learning for dynamic link prediction.
2/17/2026
Multi-Agent Debate: A Unified Agentic Framework for Tabular Anomaly Detection
Multi-agent debate framework treating model disagreements as signals for improved tabular anomaly detection.
2/17/2026
AD-Bench: A Real-World, Trajectory-Aware Advertising Analytics Benchmark for LLM Agents
Benchmark for evaluating LLM agents on real-world advertising analytics tasks with multi-round trajectory-aware workflows.
2/17/2026
A Rational Analysis of the Effects of Sycophantic AI
Epistemological analysis of sycophancy in LLMs and its effects on user beliefs and worldview formation.
2/17/2026
Integrating Unstructured Text into Causal Inference: Empirical Evidence from Real Data
Framework using transformer language models to perform causal inference from unstructured text data.
2/17/2026
Reverse N-Wise Output-Oriented Testing for AI/ML and Quantum Computing Systems
Testing methodology framework addressing challenges in validating AI/ML and quantum computing systems.
2/17/2026
Whom to Query for What: Adaptive Group Elicitation via Multi-Turn LLM Interactions
Adaptive multi-turn LLM-based framework for optimizing respondent selection in information elicitation surveys.
2/17/2026
FMMD: A multimodal open peer review dataset based on F1000Research
Multimodal dataset for automated scholarly peer review research based on F1000Research platform data.
2/17/2026
KernelBlaster: Continual Cross-Task CUDA Optimization via Memory-Augmented In-Context Reinforcement Learning
Agentic RL framework using memory augmentation for continual CUDA code optimization across GPU architectures.
2/17/2026
Machine Learning as a Tool (MLAT): A Framework for Integrating Statistical ML Models as Callable Tools within LLM Agent Workflows
Design pattern for integrating pre-trained ML models as callable tools within LLM agent workflows for hybrid reasoning.
2/17/2026
Does Socialization Emerge in AI Agent Society? A Case Study of Moltbook
Large-scale empirical study examining whether AI agent societies develop social convergence dynamics similar to human systems.
2/17/2026
DeepFusion: Accelerating MoE Training via Federated Knowledge Distillation from Heterogeneous Edge Devices
Federated learning approach for training Mixture-of-Experts LLMs using heterogeneous edge devices via knowledge distillation.
2/17/2026
Train Less, Learn More: Adaptive Efficient Rollout Optimization for Group-Based Reinforcement Learning
Adaptive efficient rollout optimization for GRPO-based LLM post-training reducing computational waste when rollout groups share outcomes.
2/17/2026
Zero-Shot Instruction Following in RL via Structured LTL Representations
Approach for zero-shot instruction following in multi-task RL using linear temporal logic representations for temporally extended tasks.
2/17/2026
AXE: An Agentic eXploit Engine for Confirming Zero-Day Vulnerability Reports
AXE agentic system automatically exploits and validates zero-day vulnerability reports using vulnerability metadata and source code.
2/17/2026
Key Considerations for Domain Expert Involvement in LLM Design and Evaluation: An Ethnographic Study
Ethnographic study examining challenges in designing and evaluating LLMs with domain experts through 12-week pedagogical chatbot development.
2/17/2026
A Trajectory-Based Safety Audit of Clawdbot (OpenClaw)
Safety audit of Clawdbot, a self-hosted tool-using AI agent, evaluating risks across six dimensions using trajectory-based evaluation.
2/17/2026
InnoEval: On Research Idea Evaluation as a Knowledge-Grounded, Multi-Perspective Reasoning Problem
InnoEval framework for evaluating research ideas using LLMs with knowledge grounding and multi-perspective reasoning across evaluation dimensions.
2/17/2026
Differentially Private Retrieval-Augmented Generation
Methods for differentially private retrieval-augmented generation protecting sensitive data while reducing hallucinations in LLM outputs.