Departmental Seminar – “Inherent Weight Normalization in Stochastic Neural Networks” by Dr Georgios Is. Detorakis
| Dates: | 11 March 2026 |
|---|
| Times: | 14:00 - 15:00 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | Department of Computer Science |
|---|
| How much: | Free |
|---|
| Who is it for: | University staff |
|---|
| Speaker: | Dr Georgios Is. Detorakis |
|---|
|
Neural Sampling Machines (NSM) is a class of neural networks with binary threshold neurons that rely almost exclusively on multiplicative noise as a resource for inference and learning. The probability of activation of the NSM exhibits a self-normalizing property that mirrors Weight Normalization, a previously studied mechanism that fulfills many of the features of batch normalization in an online fashion. The always-on stochasticity of the NSM can leverage the stochasticity inherent to a physical substrate, such as analog non-volatile memories for in-memory computing, and is well-suited for Monte Carlo sampling, while requiring almost exclusively addition and comparison operations.
Speaker
Dr Georgios Is. Detorakis
Role: Lecturer in Neuromorphic Systems, Department of Computer Science
Organisation: University of Manchester
Biography: Georgios Is. Detorakis holds a B.Sc. in Applied Mathematics, an M.Sc. in Brain and Mind Sciences, and a Ph.D. in Computer Science and Computational Neuroscience. From 2013 to 2015, he was a postdoctoral researcher at L2S - CentraleSupélec. There, he worked on Parkinson’s disease, combining computational neuroscience and control theory. From 2015 to 2019, he was a postdoctoral researcher in the Cognitive Sciences Department at the University of California, Irvine. His research focused on neuromorphic computing and deep learning. Since then, he has worked as a data scientist and machine learning scientist in various companies. He focused on developing deep learning methods for time series analysis and forecasting, as well as natural language processing. At the same time, he continued researching the applications of deep learning to physical problems, especially those involving differential equations.
Travel and Contact Information
Find event
Kilburn_TH 1.3
Kilburn Building
Manchester
