Tag
The paper introduces ContractRL, a contract-constrained sequential repair protocol that models verifier-guided JSON repair as a bounded decision process using shielded group-relative policy optimization for tool calls. It achieves 0.9362 semantic success with far fewer generated tokens than Patch-SFT and full regeneration baselines.
This paper presents SkinAgent AI, a multimodal agentic framework for non-diagnostic skincare support that combines visual analysis with safe, auditable orchestration using large language models.
The study proposes a selective deferral framework for AI systems predicting crash severity in dementia-related cases, emphasizing auditability, calibrated uncertainty, and human review to improve public-sector decision support.
ai·rete·rag is a hosted product that uses a Rete rule engine for auditable decisions and RAG to provide explanations, designed for compliance-sensitive industries.
This study proposes modular deep learning augmentations for next-day wildfire spread prediction, enhancing audibility and trustworthiness through attention biases, retrieval augmentation, and dual-stream gating, evaluated on benchmarks with performance metrics.
This paper introduces a process-centric benchmark for evaluating AI-assisted peer review systems, aiming to improve transparency and reliability beyond final decision accuracy.
This paper presents ThyroidXAgent, a clinician-interactive agentic AI system that coordinates specialized diagnostic tools for thyroid ultrasound, storing outputs as auditable evidence records. Developed on a large multicentre dataset, it achieves strong results in nodule segmentation, benign-malignant classification, and report generation.
Proposes Causal-Audit, a framework for explicit and auditable causal reasoning in LLMs using target-aware causal graph construction and path-level evidence aggregation, outperforming existing methods on benchmarks.
FirstResearch introduces a structured framework for LLM scientific discovery agents that generates a Research Question Certificate containing primitive definitions, assumptions, mechanism, falsifiable hypothesis, and failure update rules, making the proposed research question inspectable before execution. Preliminary evaluations using LLM judges show that the certificate-centered approach outperforms baseline systems in audibility and score.