PyMC Labs to open-source MFM: prior-data fitted networks do Bayesian media mix modeling in one pass
twiecki · x · 2026-10-23
PyMC Labs principal data scientist Carlos Trujillo will present MFM (Foundation Models for Bayesian Media Mix Modeling) at ODSC AI West on Oct 28, 4:05 PM PT.
The approach builds on prior-data fitted networks (PFN): train once on simulated data, then return a posterior in a single forward pass. MFM uses this to split a brand's sales into baseline, channel, and control contributions — each with an uncertainty band — with no per-dataset fitting and no MCMC chains to diagnose.
The team says it will open-source the project; replying "MFM" to the tweet gets you notified at release.
More from Venture
- How Hyperscalers Ensure Middlemen Die First When the AI Bubble Bursts — mattwbaker · 2026-10-23
- Nvidia plunges 7.21% to $212.31 in a day, still holding a $5.2T market cap — michalmalewicz · 2026-10-23
- $SPCX selloff blamed mostly on token unlock, macro AI selloff only ~30% of drop — JOBhakdi · 2026-10-23
- New paper: 88% of VC-backed founders get zero exit value, SF homeowners prove it — chrisalbon · 2026-10-23
- Growth intern who helped scale Valley from $30k to $800k+ MRR shares his GTM playbook — itsOmSarraf_ · 2026-10-23
- Anthropic's earliest backers from Silicon Valley to Singapore eye one of VC's biggest paydays — nmasc_ · 2026-10-23