REMIND Your Neural Network to Prevent Catastrophic Forgetting

By Tyler L. Hayes, Kushal Kafle, Robik Shrestha, Manoj Acharya, Christopher Kanan. 2019

349 citations recorded by the source corpus.

People learn throughout life. However, incrementally updating conventional neural networks leads to catastrophic forgetting. A common remedy is replay, which is inspired by how the brain consolidates memory. Replay involves fine-tuning a network on a mixture of new and old instances. While there is neuroscientific evidence that the brain replays compressed memories, existing methods for convolutional networks replay raw images. Here, we propose REMIND, a brain-inspired approach that enables efficient replay with compressed representations. REMIND is trained in an online manner, meaning it learns one example at a time, which is closer to how humans learn. Under the same constraints, REMIND outperforms other methods for incremental class learning on the ImageNet ILSVRC-2012 dataset. We probe REMIND's robustness to data ordering schemes known to induce catastrophic forgetting. We demonstrate REMIND's generality by pioneering online learning for Visual Question Answering (VQA).

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): “Such biological advantage is further incorporated by replaying compressed data [444] or feature representations [25], [158].”
Long-Tailed Class Incremental Learning (2022, methodology): “Replay-based methods are very effective and recall knowledge from previous tasks by maintaining a small memory of samples [2,4,7,13,32,41], representations [12], or synthetic data [35].”
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).”
Representation Compensation Networks for Continual Semantic Segmentation (2022, background): “The knowledge can be stored by multiple types, like examples [4, 6, 9, 11, 73, 83], prototypes [35, 106, 107], generative networks [60], etc.”
Memory Replay with Data Compression for Continual Learning (2022, critical): “REMIND (Hayes et al., 2020) saved the old features and reconstructed the synthesized features for replay, but it froze the majority of feature extractor after learning the initial phase, limiting the learning of representations for incremental tasks.”
ACAE-REMIND for Online Continual Learning with Compressed Feature Replay (2021, methodology): “To circumvent the difficulties of image replay, recent work has focused on feature replay [10, 21].”
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)).”
Replay in Deep Learning: Current Approaches and Missing Biological Elements (2021, positive): “While there have been several replay selection methods proposed, many existing works have found uniform random sampling of previous memories to work well, especially for large-scale problems (Chaudhry et al., 2018; Wu, Chen et al., 2019; Hayes et al., 2020).”
Rainbow Memory: Continual Learning with a Memory of Diverse Samples (2021, background): “propose to replay ‘compressed memory’ to increase the memory utilization [18].”
Self-Supervised Training Enhances Online Continual Learning (2021, methodology): “We study the common continual learning paradigm in which pre-training precedes continual learning [28, 29, 55, 10, 71, 32, 33, 34, 68, 4].”
Online Continual Learning in Image Classification: An Empirical Survey (2021, background): “Feature replay is a promising direction where latent features of the old samples at a given layer (feature extraction layer) are relayed instead of raw data [65, 116, 117, 118].”
PLOP: Learning without Forgetting for Continual Semantic Segmentation (2020, background): “Several solutions exist to address this problem: for instance, rehearsal learning consists in keeping a limited amount of training data from old classes either as raw images [55, 54, 6, 10], compressed features [24, 33], or generated training data [34, 60, 43].”
Meta-Consolidation for Continual Learning (2020, background): “While a handful of existing methods [29, 63] take this into consideration, many others [48, 38, 88] assume availability of task-identifying information during inference.”
Continual Prototype Evolution: Learning Online from Non-Stationary Data Streams (2020, methodology): “The replay data is either obtained directly from operational memoryM with stored samples [37, 31, 48, 18] or generated using generative models [42, 23, 39, 50].”
MgSvF: Multi-Grained Slow versus Fast Framework for Few-Shot Class-Incremental Learning (2020, background): “R ECENT years have witnessed a great development of class-incremental learning [1], [2], [3], [4], [5], [6], [7], [8], which aims at enabling a learner to acquire new knowledge from new data while preserving the learned knowledge from previous data.”
Class-Incremental Learning: Survey and Performance Evaluation on Image Classification (2020, methodology): “…select methods for our evaluation: property 2 excludes online methods [18], [19], property 3 excludes growing architectures [26], [27], property 4 excludes methods that require a pretrained network [28], [29], and property 5 excludes methods that are not yet scalable to larger datasets [30], [31].”
A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020, methodology): “te new knowledge [7,8,14,28,75]. The performance of these algorithms depends heavily on the existence of a bounded memory of the past. We also present results with ﬁxedrepresentation based algorithms [6,23,24,56], which do not update the model but are also less dependent on memory. LwF [36] uses dis”