Unsupervised Learning and Deep Generative Networks
Event details
| Date | 23.03.2018 |
| Hour | 14:00 › 14:45 |
| Speaker | Stéphane Mallat |
| Location | |
| Category | Conferences - Seminars |
Summary
Generative convolutional networks have obtained spectacular results to synthesize complex signals such as images, speech, music, with barely any mathematical understanding. This lecture will move towards this world by beginning from well relatively understood maximum entropy modelization. We shall review deep Generative networks such as GAN and Variational Encoders, which can synthesize realizations of non-stationary processes or highly complex processes such as speech or music. We will show that they can be simplified by defining the estimation as an inverse problem. It builds a bridge with maximum entropy estimation. Applications will be shown on images, speech and music generation.
Biography
Stéphane Mallat has made fundamental contributions to the development of wavelet theory. He has also done work in applied mathematics, signal processing, music synthesis, and image segmentation. He has developed (with Yves Meyer) the multi-resolution analysis construction for compactly supported wavelets, which made the implementation of wavelets practical for engineering applications. He has also developed (with Sifen Zhong) the wavelet transform modulus maxima method for image characterization. He has introduced the scattering transform that constructs invariance for object recognition purposes. Stéphane Mallat is the author of A Wavelet Tour of Signal Processing, a common text in applied mathematics and engineering courses. He has held teaching positions at New York University, Massachusetts Institute of Technology, École polytechnique, and at the École normale supérieure. He is currently Professor of Data Science at Collège de France.
Generative convolutional networks have obtained spectacular results to synthesize complex signals such as images, speech, music, with barely any mathematical understanding. This lecture will move towards this world by beginning from well relatively understood maximum entropy modelization. We shall review deep Generative networks such as GAN and Variational Encoders, which can synthesize realizations of non-stationary processes or highly complex processes such as speech or music. We will show that they can be simplified by defining the estimation as an inverse problem. It builds a bridge with maximum entropy estimation. Applications will be shown on images, speech and music generation.
Biography
Stéphane Mallat has made fundamental contributions to the development of wavelet theory. He has also done work in applied mathematics, signal processing, music synthesis, and image segmentation. He has developed (with Yves Meyer) the multi-resolution analysis construction for compactly supported wavelets, which made the implementation of wavelets practical for engineering applications. He has also developed (with Sifen Zhong) the wavelet transform modulus maxima method for image characterization. He has introduced the scattering transform that constructs invariance for object recognition purposes. Stéphane Mallat is the author of A Wavelet Tour of Signal Processing, a common text in applied mathematics and engineering courses. He has held teaching positions at New York University, Massachusetts Institute of Technology, École polytechnique, and at the École normale supérieure. He is currently Professor of Data Science at Collège de France.
Practical information
- General public
- Free
Organizer
- Laboratoire d'imagerie biomédicale, Prof. Michael Unser