Learning to Remember: A Synaptic Plasticity Driven Framework for Continual Learning

By Oleksiy Ostapenko, Mihai Marian Puscas, Tassilo Johannes Klein, Patrick Jähnichen, Moin Nabi. 2019

356 citations recorded by the source corpus.

Models trained in the context of continual learning (CL) should be able to learn from a stream of data over an undefined period of time. The main challenges herein are: 1) maintaining old knowledge while simultaneously benefiting from it when learning new tasks, and 2) guaranteeing model scalability with a growing amount of data to learn from. In order to tackle these challenges, we introduce Dynamic Generative Memory (DGM) - synaptic plasticity driven framework for continual learning. DGM relies on conditional generative adversarial networks with learnable connection plasticity realized with neural masking. Specifically, we evaluate two variants of neural masking: applied to (i) layer activations and (ii) to connection weights directly. Furthermore, we propose a dynamic network expansion mechanism that ensures sufficient model capacity to accommodate for continually incoming tasks. The amount of added capacity is determined dynamically from the learned binary mask. We evaluate DGM in the continual class-incremental setup on visual classification tasks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Meta & Continual Learning

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Meta & Continual Learning

Citations in the corpus

A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “A majority of state-of-the-art approaches have focused on GANs to enjoy its advantages in ﬁne-grained generation, but usually suffer from label inconsistency in continual learning [164], [168].”
Learnability and Algorithm for Continual Learning (2023, methodology): “Pseudoreplay methods generate pseudo replay data for old tasks to serve as the replay data (Kamra et al., 2017; Shin et al., 2017; Wu et al., 2018; Seff et al., 2017; Kemker & Kanan, 2018; Hu et al., 2019; Rostami et al., 2019; Ostapenko et al., 2019).”
A Theoretical Study on Solving Continual Learning (2022, background): “Some systems learn to generate pseudo training data of old tasks to be used in replay, called pseudo-replay [41, 42, 43, 44, 45, 46, 21, 47, 48].”
ALIFE: Adaptive Logit Regularizer and Feature Replay for Incremental Semantic Segmentation (2022, methodology): “To avoid these problems, recent works employ a pseudo replay scheme that synthesizes images [20, 31, 34] or latent features [38] of previous categories.”
Adaptive Orthogonal Projection for Batch and Online Continual Learning (2022, background): “Related work includes (Shin et al. 2017; Wu et al. 2018; Kamra, Gupta, and Liu 2017; Seff et al. 2017; Hu et al. 2019; Ostapenko et al. 2019; Hayes et al. 2019; von Oswald et al. 2020).”
Continual Learning through Retrieval and Imagination (2022, background): “DGM (Ostapenko et al. 2019) relies on conditional GANs with neural masking, which needs to expand network parameters for new tasks.”
Class-Incremental Learning by Knowledge Distillation with Adaptive Feature Consolidation (2022, background): “For example, [34] synthesizes the fake samples of the old tasks to balance the number of examples between old and new classes by employing class-conditional image synthesis networks [33].”
Uncertainty-aware Contrastive Distillation for Incremental Semantic Segmentation (2022, background): “Existing works in the literature can be grouped into three main categories: rehearsal-based methods [16], [30], [31], [32], [33], parameter isolation approaches [34], [35] and distillation-based techniques [14], [36], [37].”
Static-Dynamic Co-Teaching for Class-Incremental 3D Object Detection (2021, background): “…model and the current model; 2) rehearsal/replay-based methods store a subset of exemplars from previous classes (Rebuffi et al. 2017; Castro et al. 2018; Wu et al. 2019) or produce synthesized exemplars for previous classes using a generative model (Shin et al. 2017; Ostapenko et al. 2019).”
Self-Supervised Models are Continual Learners (2021, methodology): “Following [17], these works can be organized into three macro-categories: replay-based [4, 12, 35, 39, 43, 44], regularization-based [2, 9–11, 18, 21, 30, 31, 33, 47, 51, 55], and parameter isolation methods [45,46].”
Continual Learning via Local Module Composition (2021, methodology): “Continual learning methods typically address the problem of forgetting through parameter regularization [46, 70, 99], replay [89, 78, 3, 73, 54, 12, 97, 39, 81, 13] or dynamic architectures (and MoEs) [83, 87, 52, 85, 53, 41].”
Bridging Non Co-occurrence with Unlabeled In-the-wild Data for Incremental Object Detection (2021, neutral): “They usually keep a small number of exemplars [22, 29, 3], or generate synthetic images [26, 19] or features [12, 30] to achieve this purpose.”
Class-Incremental Learning for Action Recognition in Videos (2021, neutral): “On the other hand, pseudo-rehearsal techniques [28, 32] generate samples in the previously observed classes using generative adversarial networks (GANs) [11, 27].”
Recent Advances of Continual Learning in Computer Vision: An Overview (2021, neutral): “A few other methods [212, 213] explore the combination of generative replay and parameter isolation based methods for image classification.”
RECALL: Replay-based Continual Learning in Semantic Segmentation (2021, background): “Catastrophic forgetting is a long-standing problem [36, 14] which has been recently tackled in a variety of visual tasks such as image classification [22, 35, 26, 43, 33], object detection [40, 25] and semantic segmentation [29, 31, 5, 23].”
Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning (2021, methodology): “Experience replay with stored data [2, 3, 7, 8, 13, 17, 22, 27, 38, 44, 45, 46, 57] or samples from a generative model [25, 26, 43, 49, 54] is highly effective when storing training data or training/saving a generative model is possible, unlike the DFCIL setting.”
Continual Learning by Using Information of Each Class Holistically (2021, methodology): “…learn to generate pseudo old data and use them to jointly train with the new task data (Shin et al. 2017; Wu et al. 2018; Kamra, Gupta, and Liu 2017; Seff et al. 2017; Hu et al. 2019; Lesort et al. 2018; Ostapenko et al. 2019; Hayes et al. 2019) (also known as rehearsal (von Oswald et al. 2020)).”