Finite-time Lyapunov exponents of deep neural networks
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2023
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911184228188160 |
|---|---|
| author | Storm, L. Linander, H. Bec, J. Gustavsson, K. Mehlig, B. |
| author_facet | Storm, L. Linander, H. Bec, J. Gustavsson, K. Mehlig, B. |
| contents | We compute how small input perturbations affect the output of deep neural networks, exploring an analogy between deep networks and dynamical systems, where the growth or decay of local perturbations is characterised by finite-time Lyapunov exponents. We show that the maximal exponent forms geometrical structures in input space, akin to coherent structures in dynamical systems. Ridges of large positive exponents divide input space into different regions that the network associates with different classes. These ridges visualise the geometry that deep networks construct in input space, shedding light on the fundamental mechanisms underlying their learning capabilities. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2306_12548 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Finite-time Lyapunov exponents of deep neural networks Storm, L. Linander, H. Bec, J. Gustavsson, K. Mehlig, B. Disordered Systems and Neural Networks Machine Learning We compute how small input perturbations affect the output of deep neural networks, exploring an analogy between deep networks and dynamical systems, where the growth or decay of local perturbations is characterised by finite-time Lyapunov exponents. We show that the maximal exponent forms geometrical structures in input space, akin to coherent structures in dynamical systems. Ridges of large positive exponents divide input space into different regions that the network associates with different classes. These ridges visualise the geometry that deep networks construct in input space, shedding light on the fundamental mechanisms underlying their learning capabilities. |
| title | Finite-time Lyapunov exponents of deep neural networks |
| topic | Disordered Systems and Neural Networks Machine Learning |
| url | https://arxiv.org/abs/2306.12548 |