The topology of neural networks and their activation patterns
Event details
| Date | 20.11.2015 |
| Hour | 14:15 › 15:30 |
| Speaker | Paolo Masulli (UNIL) |
| Location |
MA 10
|
| Category | Conferences - Seminars |
The dynamical evolution of a network is strongly associated with its pattern of internal connections, but the lack of periodic patterns in the vast majority of biological networks, and in recurrent neural networks in particular, makes it difficult to understand this correlation from a theoretical and formal approach. We use algebraic topology to encode the connectivity structure of a network and build invariants that give us information on the dynamical evolution of a network, looking in particular at the example of recurrent boolean artificial neural networks, in order to relate topology and activation patterns. Our final goal is to shed light on the problem of how the more complex temporal activation patterns that are observed in biological networks are related with their topology.
Practical information
- Informed public
- Free
Organizer
- Kathryn Hess