DigestAI news desk

Cut through the AI noise.

Listen
Research

Researchers release TRACE framework to improve oncology LLM accuracy

Researchers introduced TRACE, a deployable framework designed to enhance the accuracy of large language models in oncology applications. The system organizes medical concepts and relations into an updatable tree-relational structure, refined using evidence derived from LM-loss. This structure is then retrieved during inference to provide task-adaptive evidence without requiring supervised labels…

1 source primary source

Key points

  • TRACE organizes oncology concepts into an updatable tree-relational structure for evidence retrieval during inference
  • Outperforms vanilla RAG and GraphRAG across ten oncology classification tasks and MedQuAD CancerGov benchmark
  • Supports task-adaptive evidence selection without supervised labels in zero-shot settings

TRACE outperforms baseline methods like vanilla RAG and GraphRAG across ten oncology classification tasks and the MedQuAD CancerGov QA benchmark. It also demonstrates robustness under leakage-controlled METABRIC inputs and produces interpretable evidence paths aligned with clinical reasoning. The authors claim this approach improves both label-free evaluation and supervised fine-tuning, suggesting a practical path toward more accurate and auditable oncology LLM deployments.

Read the original at arXiv cs.CL · by Jizheng Lai, Yingyun Li, Ying Qin, Haiyang Qian primary sourceOpen source ↗
Topics · follow one to build your own front page
TRACE

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Research

All →

Related stories

hot Google DeepMind announces Gemini 4 Argon with 1M output tokens for coding and cyber defense new Researchers release ArgGYM benchmark for testing defeasible reasoning in AI models hot OpenAI releases GPT-6.1 Sol, says it nearly matches Astra on agentic coding at one-fifth the price new Researchers release SimTrace for generating synthetic user behavior data hot Anthropic files for IPO with $518 billion infrastructure plan and $2 trillion valuation target new MetaPersona framework uses 11,000+ studies to build synthetic populations for AI tasks hot OpenAI's GPT-6 Astra: Changing Work and Research new Researchers propose DLFP controller to cut AI inference latency by up to 30%