New benchmark reveals plasma diagnostic ML models struggle with sensor failures

A new research paper evaluates the robustness of machine learning models used in plasma diagnostic systems for tokamak fusion devices. The study, conducted on the TokaMark dataset, tested various architectures including XGBoost, LSTM, and Transformer models against six simulated sensor failure scenarios. A key finding indicates that sequence models like LSTM are particularly vulnerable to sensor failures occurring close to plasma disruptions, significantly degrading their performance, while statistical models show more stability. AI

IMPACT Highlights critical vulnerabilities in ML models used for fusion energy research, potentially impacting safety and operational reliability.

RANK_REASON Research paper detailing a new benchmark for ML model robustness in a specific scientific domain. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• CNN
• long short-term memory
• Neerav Gupta
• Robustness Score (RS)
• TokaMark
• Transformer++
• XGBoost

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →