Inference-based machine learning and statistical mechanics share deep isomorphisms, and utilize many of the same computational techniques (such as efficient techniques for Markov chain Monte Carlo sampling). Despite this, there has been historically poor communication between these two fields. This meeting—the first of its kind—brings together key people from both communities to find common ground and exciting new opportunities at the interface.

The meeting will feature a series of short talks on various topics of relevance to the interface of these two fields, with the goal of exposing fruitful areas of research where both communities can make significant contributions and reap significant rewards.


smml:2017 is a satellite meeting that takes place immediately before the Berkeley Stat Mech Meeting, and will take place Thu 12 Jan 2017 @ 9.00A - Fri 13 Jan 2017 @ 2.00P.


Due to last-minute technical difficulties with our original venue, the workshop has been moved to 50 Birge Hall, a short distance away (~500 ft).

50 Birge Hall, University of California, Berkeley (what3words: rewarding.custom.spider)

50 Birge Hall also has a capacity of 159, allowing us to admit more people from the waitlist.



  • Isomorphisms between statistical mechanics and statistical inference
  • What can stat mech do for machine learning?
  • What can machine learning do for stat mech?
  • Efficient integrators for sampling from intractable distributions
  • Validating machine learning implementations using stat mech principles
  • Efficient model comparison and its relation to free energy calculation (nonequilibrium methods for model comparison, relative/absolute model evidences, Bayes factors, and free energies)


Talks are 30 min (including roughly 10 min of introductory material and 20 min of problem, application, and opportunities) followed by 15 min of discussion. All speakers have been requested to bring a slide of provocative questions and ideas to stimulate discussion.

Thu 12 Jan

9.00A coffee and pastries 50 Birge Hall
9.25A John D. Chodera (MSKCC) welcome remarks
9.30A Gavin E. Crooks (LBNL) discussion moderator
9.30A Jascha Sohl-Dickstein (Google) Deep unsupervised learning using nonequilibrium thermodynamics
10.15A Todd Gingrich (MIT, Physics of Living Systems) Sampling low-dissipation protocols
11.00A coffee break (30 min) 50 Birge Hall
11.30A Jonathan Weare (U Chicago) Some ensemble methods for faster MCMC
12.15P lunch on your own local restaurants
2.00P Vijay S. Pande (Stanford / a16z) discussion moderator
2.00P Danilo Rezende (Google DeepMind) Are deep generative models effective field theory machines?
2.45P Theofanis Karaletsos (Geometric Intelligence) Bayesian inference on graphs as hypothesis testing
3.30P coffee break (30 min) 50 Birge Hall
4.00P James Zou (Stanford) Entropy and bias in data exploration
4.45P Vijay S. Pande (Stanford / a16z) moderated discussion
5.30P dinner on your own local restaurants

Fri 13 Jan

9.00A coffee and pastries 50 Birge Hall
9.30A Aaron R. Dinner (U Chicago) discussion moderator
9.30A Rajesh Ranganath (Princeton) Operator variational inference
10.15A Dimitris Achlioptas (UCSC) Partition function estimation via error-correcting codes
11.00A coffee break (30 min) 50 Birge Hall
11.30A Surya Ganguli (Stanford) Statistical mechanics of learning and inference in high dimensions
12.15P Rhiju Das (Stanford) Eterna: Crowdsourcing RNA engineering with stat mech, experiments, and machine learning in the loop
1.00P Aaron R. Dinner (U Chicago) / John D. Chodera (MSKCC) wrap-up discussion
1.30P lunch on your own local restaurants


Please register via Eventbrite using this link.


The Berkeley Stat Mech meeting provides a list of recommended nearby hotels.


Participants are free to sample the many restaurants in and around Berkeley and San Francisco.
Please view recommendations on area restaurants here.


Please contact John Chodera <> for questions.