Frank Hutter on why deep learning failed on tabular data for a decade
Machine Learning Street Talk · youtube · 2026-09-29
Machine Learning Street Talk shares a clip from its interview with Prior Labs co-founder Frank Hutter on why deep learning struggled with tabular data for a decade.
- Tables are messy and heterogeneous, with wildly varying feature types and semantics across datasets;
- Hyped models like TabNet failed to generalise to new datasets;
- There was no "ImageNet of tables" to anchor large-scale pretraining.
The breakthrough came from learning to transfer at the level of patterns across many different tables — TabPFN-3.5 now tops the TabArena benchmark. Full interview on MLST.
More from Models
- OpenAI Agent Gains Unauthorized Access to Australian Medicare Service — TechNadu · 2026-09-29
- Opus 5.5's music videos wow users, but general intelligence hasn't jumped — aiamblichus · 2026-09-29
- MIT researcher: Opus 5.5 built an interactive JS slide deck for a 15-minute talk — davidbau · 2026-09-29
- Dev Slams Sonnet 5.5 as 'Very Stupid, Not a Good Model' — D3VAUX · 2026-09-29
- Sebastian Raschka explains the tech behind viral Jev, from bag-of-words to LLM classifiers — Ahead of AI (Sebastian Raschka) · 2026-09-29
- Bindu Reddy: OpenAI is a lost cause, Gemini is the only hope for a Fable-7 class model — bindureddy · 2026-09-29