Sauraj Gambhir, Co-founder & CEO, Prior Labs | Slush 2025
summary
A pre-trained transformer called Tab PFN handles any row‑and‑column data (numbers, categories, text) instead of building separate models per use case, and it claims training on hundreds of millions of datasets with Nature publication and open-source release. Benchmarks show about 5,000× faster on classification and 3,000× faster on regression versus prior methods; the 2.5 release is top on Tab Arena and has ~2.5M downloads with thousands of deployments. Reported wins include a 40% improvement on Hitachi railway anomaly detection; upcoming releases add context-aware text input, million‑row scaling and causal what‑if modeling.
AI-written summary · about
related talks
25mHumans helping machines help humans run machines
30mGenerative AI in action: from POC to business value
33mMachine Learning 2.0 with Hugging Face
12mRevolutionizing Cloud-Native Applications with NLP
49mBen Hall & Barbara Fusinska
18mSmall AI Teams with Huge Impact
11mUsing AI to extend the range of data scientist
16mForm factors for your new AI coworkers