STRABLE: Benchmarking Tabular Machine Learning with Strings

Name: STRABLE: Benchmarking Tabular Machine Learning with Strings
Uploaded: 2026-10-22T06:57:17.919731+00:00
Duration: 439 s
Description: STRABLE is a new benchmark of 108 real-world tables containing both numbers and strings to study machine learning. The study evaluated 445 different learning pipelines, finding that simple embeddings work well for categorical data, while LLMs are useful for free text. It establishes STRABLE as a foundation for research into tabular learning with strings.

STRABLE is a new benchmark of 108 real-world tables containing both numbers and strings to study machine learning. The study evaluated 445 different learning pipelines, finding that simple embeddings work well for categorical data, while LLMs are useful for free text. It establishes STRABLE as a foundation for research into tabular learning with strings.

Open the full episode