Search
2 results for "labeling"
- ai · arxiv/cs.LG · 4 min
Weak Labels Fail Across Time Even When Domain Transfer Works
A study of CRISPR experiments reveals supervision drift—where the labeling mechanism itself shifts—causes model collapse in temporal transfer despite strong in-domain performance.
Apr 21, 2026 Read → - ai · arxiv/cs.LG · 3 min
Framework uses AI outputs as features, not proxies, for labeled data
Generative Augmented Inference treats LLM predictions as informative signals rather than direct substitutes, reducing human labeling needs by 75–90% across operations tasks.
Apr 17, 2026 Read →