Academic Faculty

Name	Discipline/Position	Phone	Office
Fineberg Jay	Nonlinear	02 6585 207	Levin 107
Gat Omri	Nonlinear	02 6586 361	Levin 17
Gazit Doron	Nuclear Hadronic and Few-Body Physics	02 6586 858	Popick 125
Gazit Snir	Condensed Matter Physics	02 6586 946	Danciger B 209
Giveon Amit	High Energy Physics	02 6584 646	Kaplun 207
Glasner Ami	Astrophysics and Relativity	02 6584 727	Popick 110
Gnat Orly	Astrophysics and Relativity	02 6586 851	Popick 112
Hochberg Yonit	High Energy Physics		Ross -2 1B
Hoffman Yehuda	Astrophysics and Relativity	02 6584 417	Popick 122
Horesh Assaf	Astrophysics and Relativity	02 6584 928	Kaplun Bldg. 110

Upcoming Events

Yesterday

Nonlinear Physics seminar: "Statistical mechanics of wide but finite artificial neural networks"
- Start time: 12:00pm
- End date: Wednesday, December 6th
- End time: 01:30pm
- Lecturer: Mr. Gadi Naveh (HUJI)Abstract: Over the past decade, Deep neural networks (DNNs) have brought significant advancements to various fields of science and technology. Key to their success is feature learning, enabling them to extract useful representations from raw data. However, DNNs remain challenging to interpret and lack a comprehensive analytical understanding. Recent studies revealed that DNNs in the infinite width limit are equivalent to Gaussian Processes (GPs), which are well-understoodanalytically but lack feature learning capabilities due to pre-determined features.In this thesis, we aim to bridge the gap between GPs and finite DNNs, using approaches inspired by theoretical physics. Employing statistical field theory, we treat the DNN output as a random object from a statistical ensemble. We performed extensive computer simulations of DNNs trained in both stylized and real-world settings. As for analytical methods, we first employ a perturbative approach to obtain the leading correction in 1/width to the GP limit. Subsequently, we develop a self-consistent mean-field type theory, providing a GP description of finite DNNs with a data-dependent kernel that embodies feature learning properties.
- View this event in Google Calendar
Position at Kyoto
- Start time: 02:00pm
- End date: Wednesday, December 6th
- End time: 03:00pm
- View this event in Google Calendar

Search form

Academic Faculty

Pages