AI
Interval POMDP Shielding for Imperfect-Perception Agents
Researchers have developed an algorithm to improve the safety of autonomous systems with imperfect perception. These systems can make unsafe decisions...
AI
AI
When Audio-Language Models Fail to Leverage Multimodal Context for Dysarthric Speech Recognition
Researchers have found that current audio-language models are not effectively using additional clinical context to improve speech recognition for peop...
AI
Event-Causal RAG: A Retrieval-Augmented Generation Framework for Long Video Reasoning in Complex Scenarios
Researchers have developed a new AI framework called Event-Causal RAG that improves the ability of large vision-language models to understand long vid...
AI
MBABench: Evaluating LLM Agents on End-to-End Spreadsheet Tasks in Finance
Researchers have developed a benchmark to evaluate the performance of large language model (LLM) agents on end-to-end spreadsheet tasks in finance. Th...
AI
RULER: Representation-Level Verification of Machine Unlearning
Researchers have introduced RULER, a set of metrics to verify machine unlearning at the representation level. Current methods only check if a model's ...
AI
A Framework for Measuring Appropriate Reliance on Set-Valued AI Advice
A new framework for measuring how well humans rely on set-valued AI advice has been proposed by researchers. Set-valued advice, which provides a range...
AI
Teaching agentic AI to learn expert reasoning for rare disease diagnosis
Researchers have developed a method to teach artificial intelligence systems to reason like experts in diagnosing rare diseases. The approach, called ...
AI
ChainWorld: Composing Long-Horizon Desktop Workloads from Atomic OSWorld Tasks
Researchers have developed ChainWorld, a system that composes atomic tasks into long-horizon desktop workloads. This is done through directional compa...
AI
ContextSniper: AntTrail's Token-Efficient Code Memory for Repository-Level Program Repair
Researchers have developed ContextSniper, a module for large language model agents that helps them repair code issues in repositories more efficiently...
AI
ReasFlow: Assisting Reasoning-Centric Scientific Discovery in Applied Mathematics via a Knowledge-Based Multi-Agent System
Researchers have developed ReasFlow, a system that uses artificial intelligence to assist in scientific discovery in applied mathematics. The system i...
AI
Train the Model, Not the Reader: Decodability Supervision for Verifiable Activation Explanations
Researchers have proposed a new approach called Decodability Supervision for Verifiable Activation Explanations, which aims to improve the interpretab...
AI
Rethinking Self-Evolution: A Constrained Exploration-Exploitation Process for Mitigating Skill Overfitting
Researchers propose a framework called SkillBoost to help large language model agents learn from past interactions and adapt to new situations. The ap...