On Tiny Episodic Memories in Continual Learning

By Arslan Chaudhry, Marcus Rohrbach, Mohamed Elhoseiny, Thalaiyasingam Ajanthan, Puneet Kumar Dokania, Philip H. S. Torr, Marc'Aurelio Ranzato. 2019

896 citations recorded by the source corpus.

Subfields: 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 Meta & Continual Learning

Citations in the corpus

A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “For example, Reser-voir Sampling [130], [131] randomly stores a ﬁxed amount of training samples from each input batch.”
The Ideal Continual Learner: An Agent That Never Forgets (2023, background): “, joint training of all tasks): Chaudhry et al. (2020) and Mirzadeh et al. (2021), among many others, advocated that performing multitask learning gives the best performance for continual learning, while Wu et al. (2022) claimed the opposite. (Q2) How does the rehearsal mechanism provably affect gen”
CODA-Prompt: COntinual Decomposed Attention-Based Prompting for Rehearsal-Free Continual Learning (2022, neutral): “We also compare to Experience Replay [11] to provide additional context to our results.”
Generative Negative Text Replay for Continual Vision-Language Pretraining (2022, methodology): “After the training of step t, we follow the practices [1,9] to update the memory by adopting reservoir sampling to select samples from the avaiable data to save.”
Task-Free Continual Learning via Online Discrepancy Distance Learning (2022, neutral): “(5), which provides a practical way to investigate the relationship between the memory and forgetting behaviour of existing approaches [6, 8] at each training step without requiring any task information (See more details in Appendix-D from SM).”
On the Effectiveness of Lipschitz-Driven Rehearsal in Continual Learning (2022, methodology): “Among other valid CL strategies, rehearsal approaches caught the attention of a large body of literature [62, 21, 15] thanks to their advantages.”
Repeated Augmented Rehearsal: A Simple but Strong Baseline for Online Continual Learning (2022, methodology): “However, whether rehearsal is appropriate for continual learning, considering the risk of overﬁtting the memory when using data from the memory to directly contribute to the gradient computation, has been debated vigorously [Lopez-Paz and Ranzato, 2017, Chaudhry et al., 2019, Verwimp et al., 2021].”
Exploring Example Influence in Continual Learning (2022, neutral): “Rehearsal-based methods [29, 7, 22, 6, 1, 2, 28, 3, 25] tackle the challenge of SP dilemma by retaining a subset of old tasks in a stored memory buffer with bounded resources.”
S-Prompts Learning with Pre-trained Transformers: An Occam's Razor for Domain Incremental Learning (2022, methodology): “They include non-prompting methods EWC [32], LwF [40] ER [8], GDumb [53], BiC [67], DER++ [4] and Co2L [6], as well as the prompting methods DyTox [14] and L2P [66] and self-supervised learning method CaSSLe [18].”
Incremental Task Learning with Incremental Rank Updates (2022, methodology): “For example, iCaRL [27] is the first replay method, which learns in a class-incremental way by selecting and storing exemplars closest to the feature mean of each class; MetaExperience Replay (MER) [28] combines experience replay with optimizationbased meta-learning to optimize the symmetric trade-o”
CLiMB: A Continual Learning Benchmark for Vision-and-Language Tasks (2022, methodology): “Experience Replay (ER) [Chaudhry et al., 2019] caches a small percentage of training examples in a memory buffer after training on each task, and periodically performs a “replay” training step using cached examples to refresh the model.”
Transfer without Forgetting (2022, background): “We focus our comparison on state-of-the-art rehearsal algorithms, as they prevail on most benchmarks in literature [14,8,63].”
DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning (2022, methodology): “ER [7, 13], GDumb [45], BiC [62], DER++ [3] and Co(2)L [4].”
Information-theoretic Online Memory Selection for Continual Learning (2022, neutral): “We include GSS and CBRS in Figure 8.”
Continual Normalization: Rethinking Batch Normalization for Online Continual Learning (2022, methodology): “We follow the standard setting in Chaudhry et al. (2019a) to split the original CIFAR100 (Krizhevsky & Hinton, 2009) or Mini IMN (Vinyals et al., 2016) datasets into a sequence of 20 tasks, three of which are used for hyper-parameter cross-validation, and the remaining 17 tasks are used for…”
ELLE: Efficient Lifelong Pre-training for Emerging Data (2022, methodology): “Following Chaudhry et al. (2019b), we jointly train PLMs on a mixture samples from both D i and D − 1 in each batch, and the sampling ratio of D i and D − 1 is set to 9 : 1 in every batch.”
Architecture Matters in Continual Learning (2022, background): “…eﬀective as they keep a small subset of previously seen data and either uses them directly when learning from new data (e.g., experience replay) (Chaudhry et al., 2019; Rolnick et al., 2019; Riemer et al., 2018) or incorporate them as part of the optimization process (e.g., gradient projection)…”
Continual Learning with Recursive Gradient Optimization (2022, methodology): “…to lie in a close vicinity of previous solutions by introducing quadratic penalty term to the loss function (Kirkpatrick et al., 2017; Zenke et al., 2017; Yin et al., 2020) or constraining the direction of parameter update (Farajtabar et al., 2019; Chaudhry et al., 2019a; Saha et al., 2021).”
Class-Incremental Continual Learning Into the eXtended DER-Verse (2022, methodology): “Despite its simplicity, this method has proven to be highly effective even with a minimal memory footprint [15] and serves as the basis for recent methods that propose variations on the strategies for the selection of samples to include in the memory buffer [12] or the sampling of examples from it [”