Enregistré dans:
| Auteurs principaux: | , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2023
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2306.06545 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866914779677851648 |
|---|---|
| author | Valkov, Lazar Srivastava, Akash Chaudhuri, Swarat Sutton, Charles |
| author_facet | Valkov, Lazar Srivastava, Akash Chaudhuri, Swarat Sutton, Charles |
| contents | Modular approaches that use a different composition of modules for each problem are a promising direction in continual learning (CL). However, searching through the large, discrete space of module compositions is challenging, especially because evaluating a composition's performance requires a round of neural network training. We address this challenge through a modular CL framework, PICLE, that uses a probabilistic model to cheaply compute the fitness of each composition, allowing PICLE to achieve both perceptual, few-shot and latent transfer. The model combines prior knowledge about good module compositions with dataset-specific information. We evaluate PICLE using two benchmark suites designed to assess different desiderata of CL techniques. Comparing to a wide range of approaches, we show that PICLE is the first modular CL algorithm to achieve perceptual, few-shot and latent transfer while scaling well to large search spaces, outperforming previous state-of-the-art modular CL approaches on long problem sequences. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2306_06545 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | A Probabilistic Framework for Modular Continual Learning Valkov, Lazar Srivastava, Akash Chaudhuri, Swarat Sutton, Charles Machine Learning Modular approaches that use a different composition of modules for each problem are a promising direction in continual learning (CL). However, searching through the large, discrete space of module compositions is challenging, especially because evaluating a composition's performance requires a round of neural network training. We address this challenge through a modular CL framework, PICLE, that uses a probabilistic model to cheaply compute the fitness of each composition, allowing PICLE to achieve both perceptual, few-shot and latent transfer. The model combines prior knowledge about good module compositions with dataset-specific information. We evaluate PICLE using two benchmark suites designed to assess different desiderata of CL techniques. Comparing to a wide range of approaches, we show that PICLE is the first modular CL algorithm to achieve perceptual, few-shot and latent transfer while scaling well to large search spaces, outperforming previous state-of-the-art modular CL approaches on long problem sequences. |
| title | A Probabilistic Framework for Modular Continual Learning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2306.06545 |