Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2305.17154 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912460583206912 |
|---|---|
| author | Tětková, Lenka Brüsch, Thea Scheidt, Teresa Karen Mager, Fabian Martin Aagaard, Rasmus Ørtoft Foldager, Jonathan Alstrøm, Tommy Sonne Hansen, Lars Kai |
| author_facet | Tětková, Lenka Brüsch, Thea Scheidt, Teresa Karen Mager, Fabian Martin Aagaard, Rasmus Ørtoft Foldager, Jonathan Alstrøm, Tommy Sonne Hansen, Lars Kai |
| contents | Current work on human-machine alignment aims at understanding machine-learned latent spaces and their correspondence to human representations. G{ä}rdenfors' conceptual spaces is a prominent framework for understanding human representations. Convexity of object regions in conceptual spaces is argued to promote generalizability, few-shot learning, and interpersonal alignment. Based on these insights, we investigate the notion of convexity of concept regions in machine-learned latent spaces. We develop a set of tools for measuring convexity in sampled data and evaluate emergent convexity in layered representations of state-of-the-art deep networks. We show that convexity is robust to basic re-parametrization and, hence, meaningful as a quality of machine-learned latent spaces. We find that approximate convexity is pervasive in neural representations in multiple application domains, including models of images, audio, human activity, text, and medical images. Generally, we observe that fine-tuning increases the convexity of label regions. We find evidence that pretraining convexity of class label regions predicts subsequent fine-tuning performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2305_17154 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | On convex decision regions in deep network representations Tětková, Lenka Brüsch, Thea Scheidt, Teresa Karen Mager, Fabian Martin Aagaard, Rasmus Ørtoft Foldager, Jonathan Alstrøm, Tommy Sonne Hansen, Lars Kai Machine Learning Artificial Intelligence Current work on human-machine alignment aims at understanding machine-learned latent spaces and their correspondence to human representations. G{ä}rdenfors' conceptual spaces is a prominent framework for understanding human representations. Convexity of object regions in conceptual spaces is argued to promote generalizability, few-shot learning, and interpersonal alignment. Based on these insights, we investigate the notion of convexity of concept regions in machine-learned latent spaces. We develop a set of tools for measuring convexity in sampled data and evaluate emergent convexity in layered representations of state-of-the-art deep networks. We show that convexity is robust to basic re-parametrization and, hence, meaningful as a quality of machine-learned latent spaces. We find that approximate convexity is pervasive in neural representations in multiple application domains, including models of images, audio, human activity, text, and medical images. Generally, we observe that fine-tuning increases the convexity of label regions. We find evidence that pretraining convexity of class label regions predicts subsequent fine-tuning performance. |
| title | On convex decision regions in deep network representations |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2305.17154 |