{"title":"神经计算和随时间学习的统一框架","authors":"Stefano Melacci, Alessandro Betti, Michele Casoni, Tommaso Guidi, Matteo Tiezzi, Marco Gori","doi":"arxiv-2409.12038","DOIUrl":null,"url":null,"abstract":"This paper proposes Hamiltonian Learning, a novel unified framework for\nlearning with neural networks \"over time\", i.e., from a possibly infinite\nstream of data, in an online manner, without having access to future\ninformation. Existing works focus on the simplified setting in which the stream\nhas a known finite length or is segmented into smaller sequences, leveraging\nwell-established learning strategies from statistical machine learning. In this\npaper, the problem of learning over time is rethought from scratch, leveraging\ntools from optimal control theory, which yield a unifying view of the temporal\ndynamics of neural computations and learning. Hamiltonian Learning is based on\ndifferential equations that: (i) can be integrated without the need of external\nsoftware solvers; (ii) generalize the well-established notion of gradient-based\nlearning in feed-forward and recurrent networks; (iii) open to novel\nperspectives. The proposed framework is showcased by experimentally proving how\nit can recover gradient-based learning, comparing it to out-of-the box\noptimizers, and describing how it is flexible enough to switch from fully-local\nto partially/non-local computational schemes, possibly distributed over\nmultiple devices, and BackPropagation without storing activations. Hamiltonian\nLearning is easy to implement and can help researches approach in a principled\nand innovative manner the problem of learning over time.","PeriodicalId":501301,"journal":{"name":"arXiv - CS - Machine Learning","volume":"45 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A Unified Framework for Neural Computation and Learning Over Time\",\"authors\":\"Stefano Melacci, Alessandro Betti, Michele Casoni, Tommaso Guidi, Matteo Tiezzi, Marco Gori\",\"doi\":\"arxiv-2409.12038\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"This paper proposes Hamiltonian Learning, a novel unified framework for\\nlearning with neural networks \\\"over time\\\", i.e., from a possibly infinite\\nstream of data, in an online manner, without having access to future\\ninformation. Existing works focus on the simplified setting in which the stream\\nhas a known finite length or is segmented into smaller sequences, leveraging\\nwell-established learning strategies from statistical machine learning. In this\\npaper, the problem of learning over time is rethought from scratch, leveraging\\ntools from optimal control theory, which yield a unifying view of the temporal\\ndynamics of neural computations and learning. Hamiltonian Learning is based on\\ndifferential equations that: (i) can be integrated without the need of external\\nsoftware solvers; (ii) generalize the well-established notion of gradient-based\\nlearning in feed-forward and recurrent networks; (iii) open to novel\\nperspectives. The proposed framework is showcased by experimentally proving how\\nit can recover gradient-based learning, comparing it to out-of-the box\\noptimizers, and describing how it is flexible enough to switch from fully-local\\nto partially/non-local computational schemes, possibly distributed over\\nmultiple devices, and BackPropagation without storing activations. Hamiltonian\\nLearning is easy to implement and can help researches approach in a principled\\nand innovative manner the problem of learning over time.\",\"PeriodicalId\":501301,\"journal\":{\"name\":\"arXiv - CS - Machine Learning\",\"volume\":\"45 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-18\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Machine Learning\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.12038\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Machine Learning","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.12038","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
A Unified Framework for Neural Computation and Learning Over Time
This paper proposes Hamiltonian Learning, a novel unified framework for
learning with neural networks "over time", i.e., from a possibly infinite
stream of data, in an online manner, without having access to future
information. Existing works focus on the simplified setting in which the stream
has a known finite length or is segmented into smaller sequences, leveraging
well-established learning strategies from statistical machine learning. In this
paper, the problem of learning over time is rethought from scratch, leveraging
tools from optimal control theory, which yield a unifying view of the temporal
dynamics of neural computations and learning. Hamiltonian Learning is based on
differential equations that: (i) can be integrated without the need of external
software solvers; (ii) generalize the well-established notion of gradient-based
learning in feed-forward and recurrent networks; (iii) open to novel
perspectives. The proposed framework is showcased by experimentally proving how
it can recover gradient-based learning, comparing it to out-of-the box
optimizers, and describing how it is flexible enough to switch from fully-local
to partially/non-local computational schemes, possibly distributed over
multiple devices, and BackPropagation without storing activations. Hamiltonian
Learning is easy to implement and can help researches approach in a principled
and innovative manner the problem of learning over time.