AI
The Cost of a Physics Prior Is Bounded by the Ablation Gap
Researchers have found that the cost of imposing a physics prior in machine learning models is largely dependent on the ablation gap, which measures h...
AI
AI
CellPath-Bench: A Multidimensional Benchmark for Whole-Slide Cellular Representations in Pathology Foundation Models
Researchers have developed a benchmark called CellPath-Bench to evaluate the capabilities of pathology foundation models in representing whole-slide c...
AI
Can Legal AI Know When It Is Wrong? And Do Students Know When It Is?
Researchers tested three large language models (LLMs) on a set of Indian contract law cases. The models consistently provided incorrect verdicts with ...
AI
When Trust Meets Truth: Trust-Truth Separability in LLM-as-Judge
Researchers from various institutions have published a paper titled 'When Trust Meets Truth: Trust-Truth Separability in LLM-as-Judge' on arXiv. The s...
AI
ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models
Researchers have proposed a new framework called ReFrame to improve the safety of multimodal large language models. These models can process text and ...
AI
Large Language Models at the Intersection of Software Engineering and Software Security:An Evidence-Centered Structured Survey and Research Agenda
Large Language Models (LLMs) are expanding their capabilities beyond code completion to participate in security-sensitive workflows. However, the evid...
AI
Root cause analysis via difference graph discovery from linear time-series data
Researchers have developed a method to identify the root causes of anomalies in complex systems by analyzing time-series data. They use 'difference gr...
AI
From Attention Masks to Inert Zero-Vector Tokens: OAttention and O-Closure for Token Dynamics
Researchers have proposed two new concepts in natural language processing (NLP): OAttention and O-Closure. These concepts aim to improve how attention...
AI
SENTRY: Deterministic, Intelligent Risk Assessment for IT Change Management
Researchers have developed a risk assessment platform called SENTRY to help large financial institutions manage technology changes. Current methods re...
AI
Personalized Privacy Control in LLMs via Attention Head Intervention
Researchers propose a new approach to controlling personalized privacy in large language models (LLMs). They introduce 'personalized privacy' that con...
AI
Enhancing LLMs in Predictive Political QA with Semi-Structured Data
Researchers have proposed a new framework called PSL to improve large language models (LLMs) in predictive political question answering. They argue th...
AI
Ontology-supported AI Model and Dataset Management
Researchers have developed an ontology-supported AI model and dataset management platform. The platform aims to improve the exchange and analysis of A...