A Modern Self-Referential Weight Matrix That Learns to Modify Itself
Type
Conference PaperAuthors
Irie, KazukiSchlag, Imanol
Csordás, Róbert
Schmidhuber, Juergen
KAUST Department
AI Initiative, King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia.Computer, Electrical and Mathematical Science and Engineering (CEMSE) Division
Computer Science Program
Date
2022-06-17Abstract
The weight matrix (WM) of a neural network (NN) is its program. The programs of many traditional NNs are learned through gradient descent in some error function, then remain fixed. The WM of a self-referential NN, however, can keep rapidly modifying all of itself during runtime. In principle, such NNs can meta-learn to learn, and meta-meta-learn to meta-learn to learn, and so on, in the sense of recursive self-improvement. While NN architectures potentially capable of implementing such behaviour have been proposed since the '90s, there have been few if any practical studies. Here we revisit such NNs, building upon recent successes of fast weight programmers and closely related linear Transformers. We propose a scalable self-referential WM (SRWM) that learns to use outer products and the delta update rule to modify itself. We evaluate our SRWM in supervised few-shot learning and in multi-task reinforcement learning with procedurally generated game environments. Our experiments demonstrate both practical applicability and competitive performance of the proposed SRWM. Our code is public.Acknowledgements
We would like to thank Karl Cobbe for answering some practical questions about ProcGen. Kazuki Irie wishes to thank Anand Gopalakrishnan for letting him know about ProcGen. This research was partially funded by ERC Advanced grant no: 742870, project AlgoRNN, and by Swiss National Science Foundation grant no: 200021 192356, project NEUSYM. We are thankful for hardware donations from NVIDIA & IBM. The resources used for the project were partially provided by Swiss National Supercomputing Centre (CSCS) project d115.Publisher
MLResearchPressConference/Event Name
ICML 2022 : 39th International Conference on Machine LearningarXiv
2202.05780Additional Links
https://arxiv.org/pdf/2202.05780.pdfhttps://www.semanticscholar.org/paper/A-Modern-Self-Referential-Weight-Matrix-That-Learns-Irie-Schlag/4fd61f6b860acc9c5da8766b7c9064f0ec896301
Relations
Is Supplemented By:- [Software]
Title: IDSIA/modern-srwm: Official repository for the paper "A Modern Self-Referential Weight Matrix That Learns to Modify Itself" (ICML 2022 & NeurIPS 2021 Deep RL Workshop). Publication Date: 2021-11-25. github: IDSIA/modern-srwm Handle: 10754/686757