Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer
An AI research agent can improve the score it sees without finding a modelling change that works on new materials. We ask a stricter question. After repeated experiments, does the selected change survive on data that never entered the loop, and can its code be reused? We separate the search into changes to features, models, representations, and training data. Seven searches produce 701 evaluated changes across ten Matbench endpoints. Agents receive only the mean over five inner folds, reducing reliance on any single development split. We then freeze the selected code and evaluate it once on an
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- FuzzySimilar title/name (fuzzy) · 59%wanshuiyin/Auto-claude-code-research-in-sleep →
“Fuzzy title match (0.73): “Auto Research for Materials: Auditable AI-Scientist Workflow” ≈ “wanshuiyin/Auto-claude-code-research-in-sleep””
- FuzzySimilar title/name (fuzzy) · 59%google-research/google-research →
“Fuzzy title match (0.73): “Auto Research for Materials: Auditable AI-Scientist Workflow” ≈ “google-research/google-research””
- FuzzySimilar title/name (fuzzy) · 59%argoproj/argo-workflows →
“Fuzzy title match (0.73): “Auto Research for Materials: Auditable AI-Scientist Workflow” ≈ “argoproj/argo-workflows””
- LinkedLinked via arxiv author · 85%Jingjie Ning →
“Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer”
- LinkedLinked via arxiv author · 85%Xiaochuan Li →
“Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer”
- LinkedLinked via arxiv author · 85%Shanshan Zhong →
“Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer”
- LinkedLinked via arxiv author · 85%Ji Zeng →
“Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer”
- LinkedLinked via arxiv author · 85%Guolin Ke →
“Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer”