2/25/2026
Multimodal Multi-Agent Empowered Legal Judgment Prediction
JurisMMA framework uses multimodal multi-agent systems to predict legal case outcomes by decomposing trial processes and handling multiple allegations.
2/25/2026
Talking to Yourself: Defying Forgetting in Large Language Models
SA-SFT method mitigates catastrophic forgetting in LLM fine-tuning by generating self-dialogues and mixing with task data before training.
2/25/2026
ConceptRM: The Quest to Mitigate Alert Fatigue through Consensus-Based Purity-Driven Data Cleaning for Reflection Modelling
ConceptRM addresses alert fatigue in intelligent agents by using consensus-based data cleaning and reflection models to filter false alerts.
2/25/2026
CAGE: A Framework for Culturally Adaptive Red-Teaming Benchmark Generation
CAGE: Framework systematically adapting red-teaming prompts to cultural contexts for culturally-aware LLM safety evaluation and vulnerability detection.
2/25/2026
Closing the Expertise Gap in Residential Building Energy Retrofits: A Domain-Specific LLM for Informed Decision-Making
Domain-specific LLM fine-tuned on physics-based energy simulations providing retrofit recommendations for residential buildings to bridge expertise gap.
2/25/2026
MoBiQuant: Mixture-of-Bits Quantization for Token-Adaptive Elastic LLMs
MoBiQuant: Mixture-of-bits quantization enabling token-adaptive elastic LLM deployment across varying computational resources and precisions.
2/25/2026
When Backdoors Go Beyond Triggers: Semantic Drift in Diffusion Models Under Encoder Attacks
Demonstrates encoder-side poisoning induces trigger-free semantic corruption in diffusion models through Jacobian-based geometric analysis.
2/25/2026
OpenPort Protocol: A Security Governance Specification for AI Agent Tool Access
OpenPort Protocol: Governance-first specification for secure, least-privilege tool access in AI agents with auditability and abuse resistance.
2/25/2026
Controllable Exploration in Hybrid-Policy RLVR for Multi-Modal Reasoning
Controllable exploration strategy for reinforcement learning with verifiable rewards in multi-modal LLMs to prevent entropy collapse and policy degradation.
2/25/2026
Global Prior Meets Local Consistency: Dual-Memory Augmented Vision-Language-Action Model for Efficient Robotic Manipulation
Dual-memory augmented vision-language-action model improving inference efficiency and action generation for robotic manipulation tasks.
2/25/2026
Evaluating the Reliability of Digital Forensic Evidence Discovered by Large Language Model: A Case Study
Framework automating forensic artifact extraction and validation using LLM analysis and Digital Forensic Knowledge Graph on forensic images.
2/25/2026
Analyzing Latency Hiding and Parallelism in an MLIR-based AI Kernel Compiler
Benchmark methodology for MLIR-based AI kernel compiler analyzing vectorization, multi-threading, and double buffering for edge device optimization.
2/25/2026
Mitigating "Epistemic Debt" in Generative AI-Scaffolded Novice Programming using Metacognitive Scripts
Studies pedagogical guardrails needed when using LLMs for novice programming to avoid cognitive outsourcing and ensure skill acquisition.
2/25/2026
Golden Layers and Where to Find Them: Improved Knowledge Editing for Large Language Models Via Layer Gradient Analysis
Identifies optimal layers for knowledge editing in LLMs via layer gradient analysis to improve edit success rates without degrading performance.
2/25/2026
Model Merging in the Essential Subspace
ESM: Framework using principal component analysis on layer gradients to merge multiple task-specific models while reducing task interference.
2/25/2026
CodeHacker: Automated Test Case Generation for Detecting Vulnerabilities in Competitive Programming Solutions
CodeHacker: Automated agent framework generating adversarial test cases to expose vulnerabilities in LLM-generated code solutions.
2/25/2026
Right to History: A Sovereignty Kernel for Verifiable AI Agent Execution
Sovereignty kernel framework enabling tamper-evident, verifiable logging of AI agent execution for regulatory compliance and auditability.
2/25/2026
KnapSpec: Self-Speculative Decoding via Adaptive Layer Selection as a Knapsack Problem
KnapSpec: Training-free framework reformulating draft model selection as knapsack problem to optimize LLM inference throughput via adaptive layer skipping.
2/25/2026
An Approach to Combining Video and Speech with Large Language Models in Human-Robot Interaction
Multimodal framework combining vision-language models and speech processing with fuzzy logic for robotic arm control in human-robot interaction.
2/25/2026
What Matters for Simulation to Online Reinforcement Learning on Real Robots
Systematic ablation study across 100 real-world robot runs identifying which design choices enable successful online reinforcement learning on physical robots.
2/25/2026
MultiModalPFN: Extending Prior-Data Fitted Networks for Multimodal Tabular Learning
Extends TabPFN foundation model to handle multimodal data (images, text, tabular) for healthcare and marketing applications.
2/25/2026
Exploring Anti-Aging Literature via ConvexTopics and Large Language Models
Applies ConvexTopics clustering and LLMs to organize biomedical anti-aging literature and detect research trends.
2/25/2026
Quantifying the Expectation-Realisation Gap for Agentic AI Systems
Empirical study quantifying gaps between expected and actual performance of AI agents across software engineering, clinical documentation, and decision support domains.
2/25/2026
InterviewSim: A Scalable Framework for Interview-Grounded Personality Simulation
Framework for evaluating LLM personality simulation using interview data from 671k+ interactions, grounding generation in authentic personal data rather than proxies.
2/25/2026
What Makes a Good Query? Measuring the Impact of Human-Confusing Linguistic Features on LLM Performance
Analysis of linguistic feature dimensions affecting LLM performance using 22-feature vector covering clause complexity, anaphora, and answerability.
2/25/2026
Learning Physical Principles from Interaction: Self-Evolving Planning via Test-Time Memory
PhysMem: memory framework enabling VLM robot planners to learn physical principles through interaction and test-time experience.
2/25/2026
Circuit Tracing in Vision-Language Models: Understanding the Internal Mechanisms of Multimodal Thinking
Circuit tracing framework for analyzing internal mechanisms of vision-language models using transcoders and attribution graphs.
2/25/2026
No One Size Fits All: QueryBandits for Hallucination Mitigation
QueryBandits: model-agnostic contextual bandit framework for hallucination mitigation in closed-source LLMs.
2/25/2026
Case-Aware LLM-as-a-Judge Evaluation for Enterprise-Scale RAG Systems
LLM-as-a-judge evaluation framework for enterprise RAG systems handling multi-turn case-based workflows with enterprise-specific failure modes.
2/25/2026
Three Concrete Challenges and Two Hopes for the Safety of Unsupervised Elicitation
Analysis of challenges in unsupervised elicitation and easy-to-hard generalization for steering language models toward truthful outputs.
2/25/2026
Examining and Addressing Barriers to Diversity in LLM-Generated Ideas
Investigation of mechanisms reducing LLM idea diversity compared to humans, with analysis of fixation effects and potential mitigation strategies.
2/25/2026
Protein Language Models Diverge from Natural Language: Comparative Analysis and Improved Inference
Analysis of how protein language models differ from natural language transformers and methods to improve inference on biological sequences.
2/25/2026
Hybrid LLM-Embedded Dialogue Agents for Learner Reflection: Designing Responsive and Theory-Driven Interactions
Hybrid dialogue agent design combining rule-based pedagogy with LLM-based responses for learner reflection in educational settings.
2/25/2026
Wireless Federated Multi-Task LLM Fine-Tuning via Sparse-and-Orthogonal LoRA
Federated learning framework for multi-task LLM fine-tuning via sparse-orthogonal LoRA over wireless networks with heterogeneous datasets.
2/25/2026
LESA: Learnable Stage-Aware Predictors for Diffusion Model Acceleration
LESA: learnable predictors for accelerating diffusion transformer inference through feature caching adapted to diffusion process dynamics.
2/25/2026
A Generalized Apprenticeship Learning Framework for Capturing Evolving Student Pedagogical Strategies
Apprenticeship learning framework for intelligent tutoring systems using reinforcement learning to overcome sample inefficiency and reward design challenges.
2/25/2026
Actor-Curator: Co-adaptive Curriculum Learning via Policy-Improvement Bandits for RL Post-Training
Actor-Curator: Automated curriculum learning framework using policy-improvement bandits for efficient RL post-training of large language models.
2/25/2026
What Drives Students' Use of AI Chatbots? Technology Acceptance in Conversational AI
Study applying Technology Acceptance Model to understand what drives student adoption of conversational AI chatbots for learning.
2/25/2026
Personal Information Parroting in Language Models
R&R detector suite measuring memorization of personal information (emails, phone numbers, IPs) in language models trained on web data.
2/25/2026
OptiLeak: Efficient Prompt Reconstruction via Reinforcement Learning in Multi-tenant LLM Services
OptiLeak: RL-enhanced attack method exposing prompt leakage vulnerabilities in multi-tenant LLM services via shared Key-Value cache side-channels.
2/25/2026
Enhancing Hate Speech Detection on Social Media: A Comparative Analysis of Machine Learning Models and Text Transformation Approaches
Comparative study of ML models (CNN, LSTM, BERT) for detecting hate speech and offensive language on social media using text transformation techniques.
2/25/2026
TrajGPT-R: Generating Urban Mobility Trajectory with Reinforcement Learning-Enhanced Generative Pre-trained Transformer
TrajGPT-R: transformer model generating urban mobility trajectories using reinforcement learning and pre-training for privacy-preserving data synthesis.
2/25/2026
CAMEL: Confidence-Gated Reflection for Reward Modeling
CAMEL: confidence-gated reward model combining interpretability with efficiency, using log-probability margins for prediction confidence estimation.
2/25/2026
PRECTR-V2:Unified Relevance-CTR Framework with Cross-User Preference Mining, Exposure Bias Correction, and LLM-Distilled Encoder Optimization
PRECTR-V2: unified framework combining search relevance matching and CTR prediction with cross-user preference mining and LLM-distilled encoders.
2/25/2026
UrbanFM: Scaling Urban Spatio-Temporal Foundation Models
UrbanFM: foundation model for urban spatio-temporal data analysis addressing generalization across regions and tasks in urban computing.
2/25/2026
Agile V: A Compliance-Ready Framework for AI-Augmented Engineering -- From Concept to Audit-Ready Delivery
Agile V: compliance framework embedding AI agents for requirements, design, build, test, and deployment with verification and audit trails.
2/25/2026
AdapTools: Adaptive Tool-based Indirect Prompt Injection Attacks on Agentic LLMs
AdapTools: attack framework demonstrating indirect prompt injection vulnerabilities in agentic LLMs using external data services and tool integration.
2/25/2026
Communication-Inspired Tokenization for Structured Image Representations
Communication-inspired discrete image tokenizer for vision systems, optimizing for semantic structure and object-level representations rather than texture compression.
2/25/2026
RMIT-ADM+S at the MMU-RAG NeurIPS 2025 Competition
RMIT-ADM+S: RAG system with dynamic retrieval strategy routing based on query complexity, enabling efficient operation on consumer hardware.
2/25/2026
SibylSense: Adaptive Rubric Learning via Memory Tuning and Adversarial Probing
SibylSense: inference-time learning approach adapting reward rubrics for open-ended generation tasks, addressing reward hacking and scaling rubric construction.