Learning Universal Predictors

Jordi Grau-Moya, Tim Genewein, Marcus Hutter, Laurent Orseau, Gregoire Deletang, Elliot Catt, Anian Ruoss, Li Kevin Wenliang, Christopher Mattern, Matthew Aitchison, Joel Veness
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:16178-16205, 2024.

Abstract

Meta-learning has emerged as a powerful approach to train neural networks to learn new tasks quickly from limited data by pre-training them on a broad set of tasks. But, what are the limits of meta-learning? In this work, we explore the potential of amortizing the most powerful universal predictor, namely Solomonoff Induction (SI), into neural networks via leveraging (memory-based) meta-learning to its limits. We use Universal Turing Machines (UTMs) to generate training data used to expose networks to a broad range of patterns. We provide theoretical analysis of the UTM data generation processes and meta-training protocols. We conduct comprehensive experiments with neural architectures (e.g. LSTMs, Transformers) and algorithmic data generators of varying complexity and universality. Our results suggest that UTM data is a valuable resource for meta-learning, and that it can be used to train neural networks capable of learning universal prediction strategies.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-grau-moya24a, title = {Learning Universal Predictors}, author = {Grau-Moya, Jordi and Genewein, Tim and Hutter, Marcus and Orseau, Laurent and Deletang, Gregoire and Catt, Elliot and Ruoss, Anian and Wenliang, Li Kevin and Mattern, Christopher and Aitchison, Matthew and Veness, Joel}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {16178--16205}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/grau-moya24a/grau-moya24a.pdf}, url = {https://proceedings.mlr.press/v235/grau-moya24a.html}, abstract = {Meta-learning has emerged as a powerful approach to train neural networks to learn new tasks quickly from limited data by pre-training them on a broad set of tasks. But, what are the limits of meta-learning? In this work, we explore the potential of amortizing the most powerful universal predictor, namely Solomonoff Induction (SI), into neural networks via leveraging (memory-based) meta-learning to its limits. We use Universal Turing Machines (UTMs) to generate training data used to expose networks to a broad range of patterns. We provide theoretical analysis of the UTM data generation processes and meta-training protocols. We conduct comprehensive experiments with neural architectures (e.g. LSTMs, Transformers) and algorithmic data generators of varying complexity and universality. Our results suggest that UTM data is a valuable resource for meta-learning, and that it can be used to train neural networks capable of learning universal prediction strategies.} }
Endnote
%0 Conference Paper %T Learning Universal Predictors %A Jordi Grau-Moya %A Tim Genewein %A Marcus Hutter %A Laurent Orseau %A Gregoire Deletang %A Elliot Catt %A Anian Ruoss %A Li Kevin Wenliang %A Christopher Mattern %A Matthew Aitchison %A Joel Veness %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-grau-moya24a %I PMLR %P 16178--16205 %U https://proceedings.mlr.press/v235/grau-moya24a.html %V 235 %X Meta-learning has emerged as a powerful approach to train neural networks to learn new tasks quickly from limited data by pre-training them on a broad set of tasks. But, what are the limits of meta-learning? In this work, we explore the potential of amortizing the most powerful universal predictor, namely Solomonoff Induction (SI), into neural networks via leveraging (memory-based) meta-learning to its limits. We use Universal Turing Machines (UTMs) to generate training data used to expose networks to a broad range of patterns. We provide theoretical analysis of the UTM data generation processes and meta-training protocols. We conduct comprehensive experiments with neural architectures (e.g. LSTMs, Transformers) and algorithmic data generators of varying complexity and universality. Our results suggest that UTM data is a valuable resource for meta-learning, and that it can be used to train neural networks capable of learning universal prediction strategies.
APA
Grau-Moya, J., Genewein, T., Hutter, M., Orseau, L., Deletang, G., Catt, E., Ruoss, A., Wenliang, L.K., Mattern, C., Aitchison, M. & Veness, J.. (2024). Learning Universal Predictors. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:16178-16205 Available from https://proceedings.mlr.press/v235/grau-moya24a.html.

Related Material