Full text
| |
| Author(s): |
Pietro Zanin
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | São Paulo. |
| Institution: | Universidade de São Paulo (USP). Instituto de Física (IF/SBI) |
| Defense date: | 2022-08-08 |
| Examining board members: |
Nestor Felipe Caticha Alfonso;
Juan Pablo Neirotti;
Daniel Adrián Stariolo
|
| Advisor: | Nestor Felipe Caticha Alfonso |
| Abstract | |
In this work we built and analyzed a model with the goal of enlarging and generalizing ideas from the unlearning algorithm to try to understand interacting neural networks. Besides basing this work in previous neural networks articles related to that subject, we also used some social science articles to further develop the work. The model is built by introducing a change in the Hamiltonian of other models, in which we introduce an interaction between different networks. Changing the magnitude of this interaction leads to different results, which are interesting and non-trivial. In general, we try to understand the regions where this interaction is beneficial and in which ways it can be beneficial. Despite the model being too complex to be compared to real data, it shows qualitative behaviors that mimic some social dynamics in an curious way. Additionally, the model is also interesting to the area of neural networks, as it shows a way to train the networks in an efficient way. (AU) | |
| FAPESP's process: | 21/07951-7 - Topics in neural networks: I. interaction between attractor neural networks. II. learning dynamics in deep learning architectures |
| Grantee: | Pietro Zanin |
| Support Opportunities: | Scholarships in Brazil - Master |