What Learning Systems do Intelligent Agents Need? Complementary Learning Systems Theory Updated.

By Dharshan Kumaran, Demis Hassabis, James L. McClelland. 2016

760 citations recorded by the source corpus.

Subfields: Cellular and Molecular Neuroscience, Cognitive Neuroscience, Neuroscience, Cognitive Science & Psychology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cellular and Molecular Neuroscience

In Cognitive Neuroscience

In Neuroscience

In Cognitive Science & Psychology

Citations in the corpus

Overcoming catastrophic forgetting in neural networks (2016, neutral): “This approach [often called system-level consolidation (4, 5)] is impractical for learning large numbers of tasks, as in our setting it would require the amount of memories being stored and replayed to be proportional to the number of tasks.”
Continual Lifelong Learning with Neural Networks: A Review (2018, background): “The CLS theory (Kumaran et al., 2016; McClelland et al., 1995) provides the basis for a computational frameworkmodellingmemory consolidation and retrieval in which the complementary tasks of memorization and generalization are mediated by the interplay of the mammalian hippocampus and neocortex (see”
A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “As for regional collaboration, the complementary learning system (CLS) theory [231], [291] has been widely used to inspire continual learning, which attributes the advantages of biological learning and memory to the complementary”
On the role of neurogenesis in overcoming catastrophic forgetting (2018, background): “mples is expensive, learning algorithms should be able to replay learned schema to consolidate internal representations, allowing for continual learning in the absence of external sensory stimulation [3]. The mammalian brain remains the best model of unsupervised incremental learning, which makes bi”
Lifelong Learning of Spatiotemporal Representations With Dual-Memory Recurrent Self-Organization (2018, background): “The complementary learning systems (CLS) theory (McClelland et al., 1995; Kumaran et al., 2016) holds that these two interdependent operations are mediated by the interplay of the mammalian hippocampus and neocortex, providing the means for episodic memory (specific experience) and semantic memory (”
Deep Generative Dual Memory Network for Continual Learning (2017, background): “The widely acknowledged complementary learning systems theory (McClelland et al., 1995; Kumaran et al., 2016) suggests that this separation has been achieved in the human brain via evolution of two separate areas: (a) the neocortex, which is a long term memory specializing in consolidating new infor”
Born to Learn: the Inspiration, Progress, and Future of Evolved Plastic Artificial Neural Networks (2017, background): “Inspiration for new approaches could come from complementary learning systems (McClelland et al., 1995; Kumaran et al., 2016) that humans seem to possess, which include fast and slow learning components.”
Synaptic weight decay with selective consolidation enables fast learning without catastrophic forgetting (2019, background): “Catastrophic forgetting is a problem in neuronal networks, but not so much in human continual learning (French, 1999; Kumaran et al., 2016).”
Continual Learning through Retrieval and Imagination (2022, background): “The Complementary Learning Systems (CLS) theory (Gelbard-Sagiv et al. 2008; Kumaran et al. 2016) holds that the hippocampus in human brain stores episodic-like memory that can be reactivated during sleep or in unconscious and conscious recall, thus consolidating knowledge in the neocortex via the re”
DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning (2022, background): “Differently, DualPrompt takes inspiration from CLS and presents a different approach to attach complementary prompts to the pre-trained backbone to learn task-invariant and task-specific instructions.”
Learning Fast, Learning Slow: A General Continual Learning Method based on Complementary Learning System (2022, background): “In particular, the CLS theory (Kumaran et al., 2016) posits that efﬁcient learning requires two complementary learning systems: the hippocampus exhibits short-term adaptation and rapid learning of episodic information which is then gradually consolidated to the neocortex for slow learning of…”
Learning to Prompt for Continual Learning (2021, methodology): “First, motivated by the episodic memory in the hippocampus according to the Complementary Learning Systems (CLS) theory [23, 36], many state-of-the-art methods [3, 4, 8] rely on a rehearsal buffer to re-train a portion of past examples.”
Replay in Deep Learning: Current Approaches and Missing Biological Elements (2021, neutral): “The complementary learning systems (CLS) theory describes long-term memory consolidation based on the interplay between the hippocampus and the neocortex (McClelland et al., 1995; Kumaran et al., 2016).”
Continual learning with hypernetworks (2019, neutral): “We close with a speculative systems interpretation (Kumaran et al., 2016; Hassabis et al., 2017) of our work as a model for modulatory top-down signals in cortex.”
Three scenarios for continual learning (2019, background): “Catastrophic forgetting is a major obstacle to the development of artiﬁcial intelligence applications capable of true lifelong learning [27, 28], and enabling neural networks to sequentially learn multiple tasks has become a topic of intense research.”
Generative replay with feedback connections as a general strategy for continual learning (2018, background): “One of the main challenges currently facing deep learning is that artificial neural networks are extremely bad at retaining old information when trained on something new, and enabling these networks to sequentially learn multiple tasks without catastrophic forgetting has become a topic of intense re”