{"id":"arxiv:2312.00276","title":"Metalearning Continual Learning Algorithms","date":"2023-12-01","section":"Reinforcement learning","quality":0.12499,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/","source_url":"https://arxiv.org/abs/2312.00276","text_available":true,"authors":["Kazuki Irie","Róbert Csordás","Jürgen Schmidhuber"],"outgoing":[{"source":"arxiv:2312.00276","target":"arxiv:2006.16236","evidence":"title","id":"arxiv:2006.16236","title":"Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d119844a2dce8a3b156558743b90add84774d00894634b6478d03abb4f5216f9/"},{"source":"arxiv:2312.00276","target":"doi:10.1073/pnas.1611835114","evidence":"title","id":"doi:10.1073/pnas.1611835114","title":"Overcoming catastrophic forgetting in neural networks","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/849082368a44d06b93ef3ba95bcfddc6672b7a5e8064ad2b2d19601639266e8b/"}],"incoming":[{"source":"arxiv:2505.22954","target":"arxiv:2312.00276","evidence":"title","id":"arxiv:2505.22954","title":"Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c29f716871f80314868b5647692de387f26e12166b93bd362915fd272b6e081c/"},{"source":"arxiv:2512.24695","target":"arxiv:2312.00276","evidence":"title","id":"arxiv:2512.24695","title":"Nested Learning: The Illusion of Deep Learning Architectures","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d19bf2367894704a701d0741e7db6d8b148719af1d634b55f5797aab5f617d06/"},{"source":"arxiv:2607.13104","target":"arxiv:2312.00276","evidence":"id","id":"arxiv:2607.13104","title":"Self-Improvements in Modern Agentic Systems: A Survey","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/"},{"source":"url:people.idsia.ch/~juergen/most-cited-neural-nets.html","target":"arxiv:2312.00276","evidence":"title","id":"url:people.idsia.ch/~juergen/most-cited-neural-nets.html","title":"The most cited neural nets of the 20th century (Schmidhuber)","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a07ae08185ac4ab008e5384d2dab1a6ce15cf327b9414fb3ccc3afa8e5ac914a/"}],"text_parts":[{"start":0,"end":15592,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html"},{"start":15592,"end":30777,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-2.html"},{"start":30777,"end":46565,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-3.html"},{"start":46565,"end":62141,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html"},{"start":62141,"end":71238,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html"}],"contents":[{"title":"Metalearning Continual Learning Algorithms","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"Abstract","offset":46,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"1 Introduction","offset":1898,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"2 Background","offset":7464,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"2.1 Continual Learning","offset":7862,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"2.2 Metalearning via Sequence Learning a.k.a. In-Context Learning","offset":9252,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"2.3 Self-Referential Weight Matrices and Recursive Self-Transformers","offset":13803,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"General description.","offset":13877,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-1.html#text"},{"title":"3 Method","offset":20039,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-2.html#text"},{"title":"4 Experiments","offset":29668,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-2.html#text"},{"title":"4.1 Two-Task Setting: Comprehensible Study","offset":29734,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-2.html#text"},{"title":"4.2 Analysis: Emergence of In-Context Catastrophic Forgetting","offset":33858,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-3.html#text"},{"title":"4.3 General Evaluation","offset":36337,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-3.html#text"},{"title":"5 Discussion","offset":41696,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-3.html#text"},{"title":"Other Limitations.","offset":41713,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-3.html#text"},{"title":"6 Conclusion","offset":47575,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"Acknowledgements","offset":48657,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"Appendix A Experimental Details","offset":49038,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.1 Continual and Metalearning Terminologies","offset":49074,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"Continual learning.","offset":49225,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"Metalearning.","offset":50395,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.2 Datasets","offset":51970,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.3 Training Details & Hyper-Parameters","offset":53523,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.4 Evaluation Procedure","offset":54606,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.5 ACL Objectives with More Tasks","offset":55314,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.6 Auxiliary 1-shot Learning Objective","offset":57254,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.7 Details of the Split-MNIST experiment","offset":57694,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-4.html#text"},{"title":"A.8 Details of the Split-CIFAR100 and 5-datasets Experiment using ViT","offset":63057,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"Appendix B Extra Experimental Results","offset":63938,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"B.1 Ablation Studies on the Choice of Meta-Validation Dataset","offset":63980,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"B.2 Performance on Split-Omniglot","offset":66485,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"B.3 Varying the Number of In-Context Examples/Demonstrations","offset":67083,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"B.4 Varying the Number of Tasks in the ACL Meta-Training Objective","offset":67794,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"B.5 Further Discussion on Limitations","offset":67971,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"Domain generalization.","offset":68126,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"Length generalization.","offset":69315,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"B.6 More Visualizations","offset":70116,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"},{"title":"Instructions for reporting errors","offset":70274,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/part-5.html#text"}],"text_code_points":71238,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/source.md"}