Effect of scale on catastrophic forgetting in neural networks

By Vinay Venkatesh Ramasesh, Aitor Lewkowycz, Ethan Dyer. 2022

226 citations recorded by the source corpus.

How researchers see this paper

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

In Scaling & Learning Theory

Citations in the corpus

Llama 2: Open Foundation and Fine-Tuned Chat Models (2023, background): “For example, RLHF V3 struggled more than previous versions to compose rhyming lines in poems, as discerned through qualitative analysis, suggesting that further investigation into the causes of and mitigations for forgetting (Kirkpatrick et al., 2017; Nguyen et al., 2019; Ramasesh et al., 2021) coul”
Pretraining Language Models with Human Preferences (2023, background): “, 2022), but this strategy is also limited by the fact that large LMs are quite resistant to forgetting their training data (an effect that increases with model size; Carlini et al., 2022; Vu et al., 2022; Ramasesh et al., 2022).”
A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “In particular,sinceself-supervisedandpre-trainedrepresentations arenaturallymorerobusttocatastrophicforgetting[28],[87], [89],[93],creating,improvingandexploitingsuchrepresenta-tionaladvantageshasbecomeapromisingdirection.”
SLCA: Slow Learner with Classifier Alignment for Continual Learning on a Pre-trained Model (2023, background): “For example, the representations obtained from supervised pre-training have been shown to facilitate not only knowledge transfer but also robustness to catastrophic forgetting for downstream continual learning [30, 28, 56].”
Architecture Matters in Continual Learning (2022, background): “Recently, Ramasesh et al. (2022) shows that in the pre-training setups, the scale of model and dataset can improve CL performance.”
Steering Prototypes with Prompt-tuning for Rehearsal-free Continual Learning (2023, result): “As shown in Table 2, a pre-trained ViT backbone can significantly boost the performance of existing prototype-based methods, which aligns with the observation in [39].”
Memorization Without Overfitting: Analyzing the Training Dynamics of Large Language Models (2022, background): “There is also a growing body of work demonstrating that model scale and dataset scale can make models more resistant to forgetting [34, 35], as well as work characterizing how forgetting naturally occurs in image classifiers [36].”
Continual Learning with Foundation Models: An Empirical Study of Latent Replay (2022, background): “With the increasing popularity of such models as a foundation for downstream tasks, several recent works have explored their continual ﬁne-tuning (Mehta et al., 2022; Wu et al., 2022; Ramasesh et al., 2022).”
Learning from One Continuous Video Stream (2023, background): “Recent continual learning papers reported that some of the challenges of the problem are partially mitigated by using features pretrained in IID settings [27, 36, 45].”
Editing Models with Task Arithmetic (2022, unknown): “[81], we find that results improve with model scale.”
Consistent Prompting for Rehearsal-Free Continual Learning (2024, unknown): “Continual learning [2, 6, 26, 37, 39] aims to equip deep models with the capacity to continuously acquire new knowledge, e.g., learn to recognize new object categories, while handling catastrophic forgetting [18, 27, 28, 32] of the old knowledge.”