Research
Mitra-v2: Scaling synthetic pretraining for tabular foundation models
The Brief
- Amazon Science published a blog post introducing Mitra-v2, which scales synthetic pretraining for tabular foundation models.
- Tabular data, with rows as customers, transactions or interactions and columns as attributes, underpins much business data.
- The models are aimed at predicting missing attributes, such as customer retention likelihood or expected demand.
Why it matters
It describes an approach to pretraining foundation models on synthetic tabular data, which could reduce reliance on large real datasets for prediction tasks on business tables.
Who’s affected
- machine learning researchers
- data scientists working with tabular business data
- Mitra-v2
- #tabular data
- #foundation models
- #synthetic pretraining
- #machine learning
- #amazon
Much of the data that businesses run on lives in tables: the rows might represent customers, transactions, or interactions, and their “attributes” are listed in the columns. Predicting missing attributes — such as likelihood of customer retention or how much demand to expect next…
Timeline
Amazon Science blog post on Mitra-v2 published
Amazon Science homepage
Mitra-v2: Scaling synthetic pretraining for tabular foundation models
