Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2502.06089 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915144515190784 |
|---|---|
| author | Gourdeau, Pascale Lechner, Tosca Urner, Ruth |
| author_facet | Gourdeau, Pascale Lechner, Tosca Urner, Ruth |
| contents | We study the problem of computable multiclass learnability within the Probably Approximately Correct (PAC) learning framework of Valiant (1984). In the recently introduced computable PAC (CPAC) learning framework of Agarwal et al. (2020), both learners and the functions they output are required to be computable. We focus on the case of finite label space and start by proposing a computable version of the Natarajan dimension and showing that it characterizes CPAC learnability in this setting. We further generalize this result by establishing a meta-characterization of CPAC learnability for a certain family of dimensions: computable distinguishers. Distinguishers were defined by Ben-David et al. (1992) as a certain family of embeddings of the label space, with each embedding giving rise to a dimension. It was shown that the finiteness of each such dimension characterizes multiclass PAC learnability for finite label space in the non-computable setting. We show that the corresponding computable dimensions for distinguishers characterize CPAC learning. We conclude our analysis by proving that the DS dimension, which characterizes PAC learnability for infinite label space, cannot be expressed as a distinguisher (even in the case of finite label space). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_06089 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | On the Computability of Multiclass PAC Learning Gourdeau, Pascale Lechner, Tosca Urner, Ruth Machine Learning We study the problem of computable multiclass learnability within the Probably Approximately Correct (PAC) learning framework of Valiant (1984). In the recently introduced computable PAC (CPAC) learning framework of Agarwal et al. (2020), both learners and the functions they output are required to be computable. We focus on the case of finite label space and start by proposing a computable version of the Natarajan dimension and showing that it characterizes CPAC learnability in this setting. We further generalize this result by establishing a meta-characterization of CPAC learnability for a certain family of dimensions: computable distinguishers. Distinguishers were defined by Ben-David et al. (1992) as a certain family of embeddings of the label space, with each embedding giving rise to a dimension. It was shown that the finiteness of each such dimension characterizes multiclass PAC learnability for finite label space in the non-computable setting. We show that the corresponding computable dimensions for distinguishers characterize CPAC learning. We conclude our analysis by proving that the DS dimension, which characterizes PAC learnability for infinite label space, cannot be expressed as a distinguisher (even in the case of finite label space). |
| title | On the Computability of Multiclass PAC Learning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2502.06089 |