Yoshua Bengio: Generative Flow Networks | IACS Distinguished Lecturer
Yoshua Bengio
Full Professor, Université de Montréal
ABSTRACT: Generative Flow Networks (or GFlowNets) have been introduced as a method to sample a diverse set of candidates in an active learning context, with a training objective that makes them approximately sample in proportion to a given reward function. We show a number of additional theoretical properties of GFlowNets. They can be used to estimate joint probability distributions and corresponding marginal distributions (when some variables are unspecified) and are particularly interesting to represent distributions over composite objects like sets and graphs. They amortize in a single but trained generative pass the work typically done by computationally expensive MCMC methods. They can be used to estimate partition functions and free energies, conditional probabilities of supersets or of larger graphs (supergraphs) given a subset of an included subgraph, as well as marginal distributions over all supersets of a set or supergraphs of a graph. The talk will highlight the relations and differences to standard approaches in generative modeling and reinforcement learning and summarize early experimental results obtained in the context of exploring the space of molecules to discover ones with properties of interest.
Видео Yoshua Bengio: Generative Flow Networks | IACS Distinguished Lecturer канала Harvard Institute for Applied Computational Science
Full Professor, Université de Montréal
ABSTRACT: Generative Flow Networks (or GFlowNets) have been introduced as a method to sample a diverse set of candidates in an active learning context, with a training objective that makes them approximately sample in proportion to a given reward function. We show a number of additional theoretical properties of GFlowNets. They can be used to estimate joint probability distributions and corresponding marginal distributions (when some variables are unspecified) and are particularly interesting to represent distributions over composite objects like sets and graphs. They amortize in a single but trained generative pass the work typically done by computationally expensive MCMC methods. They can be used to estimate partition functions and free energies, conditional probabilities of supersets or of larger graphs (supergraphs) given a subset of an included subgraph, as well as marginal distributions over all supersets of a set or supergraphs of a graph. The talk will highlight the relations and differences to standard approaches in generative modeling and reinforcement learning and summarize early experimental results obtained in the context of exploring the space of molecules to discover ones with properties of interest.
Видео Yoshua Bengio: Generative Flow Networks | IACS Distinguished Lecturer канала Harvard Institute for Applied Computational Science
Показать
Комментарии отсутствуют
Информация о видео
12 июля 2022 г. 18:37:50
01:00:19
Другие видео канала
![CS 109b: Data Science II: Advanced Topics in Data Science](https://i.ytimg.com/vi/NaLMvDt8OAg/default.jpg)
![Rayid Ghani: Doing Practical Data Science for Social Impact | IACS Seminar](https://i.ytimg.com/vi/aK3jiNIQ5t0/default.jpg)
![Alan Heavens: Cosmology with Weak Lensing | IACS Seminar](https://i.ytimg.com/vi/h7s1DGXRsqc/default.jpg)
![Cengiz Pehlevan: A Function Approximation of Perspective on Sensory Representations | IACS Seminar](https://i.ytimg.com/vi/qsYx-6rZmoU/default.jpg)
!["Machine Learning for Partial Differential Equations" by Michael Brenner](https://i.ytimg.com/vi/9Rycnz2O-aY/default.jpg)
![Isaac Lagaris: Neural Modeling and Differential Equations | IACS Seminar](https://i.ytimg.com/vi/aKFimFiWULE/default.jpg)
!["Computational Biology in Translational Cancer Research"](https://i.ytimg.com/vi/XGYruD6wsPc/default.jpg)
![Susan Murphy: We used RL; but did it work? | IACS Distinguished Lecturer](https://i.ytimg.com/vi/F2t8D3BjkO4/default.jpg)
!["Learning Quantum Emergence with AI" by Eun-Ah Kim](https://i.ytimg.com/vi/gBTNlQD6qKE/default.jpg)
![Alan Aspuru-Guzik: Billions and Billions of Molecules: Exploring Chemical Space | IACS Seminar](https://i.ytimg.com/vi/vN1ZWOgs06E/default.jpg)
![Grushka-Cockayne: Forecasting Airport Transfer Passenger Flow Using Real-Time Data & Machine Seminar](https://i.ytimg.com/vi/rCD-0MktwNA/default.jpg)
![Thore Graepel: Automatic Curricula in Deep Multi-Agent Reinforcement Learning | IACS Seminar](https://i.ytimg.com/vi/GB5XWY_IvvE/default.jpg)
![Verena Kaynig: Connectomics - Mapping the Brain | IACS Seminar](https://i.ytimg.com/vi/4a0urY43GFU/default.jpg)
![CSE Symposium: XY: Basketball Meets Big Data 1/24/2014](https://i.ytimg.com/vi/JvNkZdZJBt4/default.jpg)
![Thomson Nguyen: Machine Learning for Small Business Lending | IACS Seminar](https://i.ytimg.com/vi/pG3Y7mXkXs8/default.jpg)
![Scott Kuindersma: Controlling Multi-Contact Robot Behaviors with Optimization | IACS Seminar](https://i.ytimg.com/vi/4GWeik_5tXo/default.jpg)
![1/19 Hanna Halaburda- "Hopes and worries of blockchain commercial use”](https://i.ytimg.com/vi/vDKVD1QhnpQ/default.jpg)
![Emmanuel Candès: Recent Progress in Predictive Inference | IACS Distinguished Lecturer](https://i.ytimg.com/vi/14jzthqF45I/default.jpg)
![Keynote Fernando Pérez | Jupyter Meets Earth: An Open, Collaborative Approach for Earth Data Science](https://i.ytimg.com/vi/GUyt_VXU8Aw/default.jpg)
![Jessica Stauth: It’s Not Just Good Leadership Advice, It’s Good Data Science Practice | IACS Seminar](https://i.ytimg.com/vi/psyJSES96Ks/default.jpg)
![WiDS 2022 Women of Color in Data Science](https://i.ytimg.com/vi/3Ve9yHj-Beo/default.jpg)