Energy-Based Adversarial Training and Video Prediction, NIPS 2016 | Yann LeCun, Facebook AI Research
NIPS 2016 Workshop on Adversarial Training
https://arxiv.org/abs/1609.03126
We introduce the "Energy-based Generative Adversarial Network" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce contrastive samples with minimal energies, while the discriminator is trained to assign high energies to these generated samples. Viewing the discriminator as an energy function allows to use a wide variety of architectures and loss functionals in addition to the usual binary classifier with logistic output. Among them, we show one instantiation of EBGAN framework as using an auto-encoder architecture, with the energy being the reconstruction error, in place of the discriminator. We show that this form of EBGAN exhibits more stable behavior than regular GANs during training. We also show that a single-scale architecture can be trained to generate high-resolution images.
Видео Energy-Based Adversarial Training and Video Prediction, NIPS 2016 | Yann LeCun, Facebook AI Research канала Preserve Knowledge
https://arxiv.org/abs/1609.03126
We introduce the "Energy-based Generative Adversarial Network" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce contrastive samples with minimal energies, while the discriminator is trained to assign high energies to these generated samples. Viewing the discriminator as an energy function allows to use a wide variety of architectures and loss functionals in addition to the usual binary classifier with logistic output. Among them, we show one instantiation of EBGAN framework as using an auto-encoder architecture, with the energy being the reconstruction error, in place of the discriminator. We show that this form of EBGAN exhibits more stable behavior than regular GANs during training. We also show that a single-scale architecture can be trained to generate high-resolution images.
Видео Energy-Based Adversarial Training and Video Prediction, NIPS 2016 | Yann LeCun, Facebook AI Research канала Preserve Knowledge
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
![Introduction to GANs, NIPS 2016 | Ian Goodfellow, OpenAI](https://i.ytimg.com/vi/9JpdAg6uMXs/default.jpg)
![David Duvenaud | Reflecting on Neural ODEs | NeurIPS 2019](https://i.ytimg.com/vi/YZ-_E7A3V2w/default.jpg)
![Your Brain on Energy-Based Models: Applying and Scaling EBMs to Problems...- Will Grathwohl](https://i.ytimg.com/vi/taob_eW0OFs/default.jpg)
![How AI Powers Self-Driving Tesla with Elon Musk and Andrej Karpathy](https://i.ytimg.com/vi/FnFksQo-yEY/default.jpg)
![Variational Inference: Foundations and Modern Methods (NIPS 2016 tutorial)](https://i.ytimg.com/vi/ogdv_6dbvVQ/default.jpg)
![Machine learning + neuroscience = biologically feasible computing | Benjamin Migliori | TEDxSanDiego](https://i.ytimg.com/vi/iPdKMs9cEAs/default.jpg)
![Ian Goodfellow: Generative Adversarial Networks (NIPS 2016 tutorial)](https://i.ytimg.com/vi/HGYYEUSm-0Q/default.jpg)
![](https://i.ytimg.com/vi/pJArCv_MiaM/default.jpg)
![Meet Geoffrey Hinton, U of T's Godfather of Deep Learning](https://i.ytimg.com/vi/zPqFbkdcvWQ/default.jpg)
![Yann LeCun - How does the brain learn so much so quickly? (CCN 2017)](https://i.ytimg.com/vi/cWzi38-vDbE/default.jpg)
![Semantic Segmentation using Adversarial Networks, NIPS 2016 | Pauline Luc, Facebook AI Research](https://i.ytimg.com/vi/aRZVBG4o0Hk/default.jpg)
![Yann LeCun: “AI Breakthroughs & Obstacles to Progress, Mathematical and Otherwise”](https://i.ytimg.com/vi/1_KhJv0Em5Y/default.jpg)
![A Connection Between GANs, Inverse Reinforcement Learning, and Energy Based Models, NIPS 2016](https://i.ytimg.com/vi/fu7uBNWTzU8/default.jpg)
![Connecting Generative Adversarial Networks and Actor Critic Methods, NIPS 2016 | David Pfau](https://i.ytimg.com/vi/BkDqCV8hieE/default.jpg)
![Yann LeCun - Graph Embedding, Content Understanding, and Self-Supervised Learning](https://i.ytimg.com/vi/UGPT64wo7lU/default.jpg)
![Video Classification with Deep Learning](https://i.ytimg.com/vi/LAV56E-mWoo/default.jpg)
![The Great AI Debate - NIPS2017 - Yann LeCun](https://i.ytimg.com/vi/93Xv8vJ2acI/default.jpg)
![Convolutional LSTMs for video prediction : self-driving cars & medical image processing](https://i.ytimg.com/vi/MjFpgyWH-pk/default.jpg)
![Why Deep Learning Now? | AI Revolution Documentary](https://i.ytimg.com/vi/b3IyDNB_ciI/default.jpg)
![Serving a Billion Personalized News Feeds](https://i.ytimg.com/vi/Xpx5RYNTQvg/default.jpg)