MSR-IISc AI Seminar Series: GFlowNets and System 2 Deep Learning - Yoshua Bengio

GFlowNets are instances of a larger family of approaches at the intersection of generative modeling and RL that can be used to train probabilistic inference functions in a way that is related to variational inference and opens a lot of new doors, especially for brain-inspired AI. Instead of maximizing some objective (like expected return), these approaches seek to sample latent random variables from a distribution defined by an energy function, for example a posterior distribution (given past data, current observations, etc). Recent work showed how GFlowNets can be used to sample a diversity of solutions in an active learning context. We will also discuss ongoing work to explore how to train such inference machinery for learning energy-based models, to approximately marginalize over infinitely many variables, perform efficient posterior Bayesian inference and incorporate inductive biases associated with conscious processing and reasoning in humans. These inductive biases include modular knowledge representation favoring systematic generalization, the causal nature of human thoughts, concepts, explanations and plans and the sparsity of dependencies captured by reusable relational or causal knowledge. Many open questions remain to develop these ideas, which will require many collaborating minds!
Slides and video details: www.microsoft.com/en-us/resea...
MSR-IISc AI Seminar Series: www.microsoft.com/en-us/resea...

[@valentinussofa4135]

Great talk. Thank you very much. 🙏

[@EagleMaster64]

wow sammy boy would love this

[@abdulshabazz8597]

My understanding is such: whereas, artificial neural networks model software primitives; gflownets model, among other things, discrete ALUs along with a builtin software primitive. Such a model would obviously perform well on a quantum computer.

[@maloxi1472]

"Obviously" 😄

[@abdulshabazz8597]

Imagine a mathematical construction which captures a discrete ALU model, a Program Status Word (PSW) flagset register, and a state machine! That would be some powerful computational power, even on a classical computing framework.

[@abtasamullahkhan3352]

With rare data, we can predict future.

[@codonology]

This is exactly what Codonology can solve. Integration between GFlowNets and Codonology may be able to create new AI tools for human-like learning and reasoning. (Founder of Codonology)

[@maloxi1472]

Could you summarize, without fluff, what "Codonology" is about ?
Your website content is frankly extremely confusing and that might explain why almost nobody seems to have given a thorough look at your ideas for the last decade.

[@codonology]

@@maloxi1472 Actually the summarization is: Codon, or -LCP-

[@maloxi1472]

@@codonology That is not a summary

[@codonology]

@@maloxi1472 Codonology is to build machine learning mechanism based on a hypothetical knowledge-learning and reasoning notion called “Codon”, a minimalism view about AI.