Training Neural Networks for Modularity aids Interpretability
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909705665773568 |
|---|---|
| author | Golechha, Satvik Cope, Dylan Schoots, Nandi |
| author_facet | Golechha, Satvik Cope, Dylan Schoots, Nandi |
| contents | An approach to improve network interpretability is via clusterability, i.e., splitting a model into disjoint clusters that can be studied independently. We find pretrained models to be highly unclusterable and thus train models to be more modular using an ``enmeshment loss'' function that encourages the formation of non-interacting clusters. Using automated interpretability measures, we show that our method finds clusters that learn different, disjoint, and smaller circuits for CIFAR-10 labels. Our approach provides a promising direction for making neural networks easier to interpret. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2409_15747 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Training Neural Networks for Modularity aids Interpretability Golechha, Satvik Cope, Dylan Schoots, Nandi Machine Learning Artificial Intelligence An approach to improve network interpretability is via clusterability, i.e., splitting a model into disjoint clusters that can be studied independently. We find pretrained models to be highly unclusterable and thus train models to be more modular using an ``enmeshment loss'' function that encourages the formation of non-interacting clusters. Using automated interpretability measures, we show that our method finds clusters that learn different, disjoint, and smaller circuits for CIFAR-10 labels. Our approach provides a promising direction for making neural networks easier to interpret. |
| title | Training Neural Networks for Modularity aids Interpretability |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2409.15747 |