CODA-Prompt: COntinual Decomposed Attention-Based Prompting for Rehearsal-Free Continual Learning

By James Smith, Leonid Karlinsky, Vyshnavi Gutta, Paola Cascante-Bonilla, Assaf Arbelle, Rameswar Panda, Rogério Schmidt Feris, Zsolt Kira. 2022

609 citations recorded by the source corpus.

Computer vision models suffer from a phenomenon known as catastrophic forgetting when learning novel concepts from continuously shifting training data. Typical solutions for this continual learning problem require extensive rehearsal of previously seen data, which increases memory costs and may violate data privacy. Recently, the emergence of large-scale pre-trained vision transformer models has enabled prompting approaches as an alternative to data-rehearsal. These approaches rely on a key-query mechanism to generate prompts and have been found to be highly resistant to catastrophic forgetting in the well-established rehearsal-free continual learning setting. However, the key mechanism of these methods is not trained end-to-end with the task sequence. Our experiments show that this leads to a reduction in their plasticity, hence sacrificing new task accuracy, and inability to benefit from expanded parameter capacity. We instead propose to learn a set of prompt components which are assembled with input-conditioned weights to produce input-conditioned prompts, resulting in a novel attention-based end-to-end key-query scheme. Our experiments show that we outperform the current SOTA method DualPrompt on established benchmarks by as much as 4.5% in average final accuracy. We also outperform the state of art by as much as 4.4% accuracy on a continual learning benchmark which contains both class-incremental and domain-incremental task shifts, corresponding to many practical settings. Our code is available at https://github.com/GT-RIPL/CODA-Prompt

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Large Language Models, 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

1. Han-Jia Ye
2. De-chuan Zhan
3. Da-Wei Zhou
4. Ziwei Liu
5. Jun Zhu
6. Hang Su
7. Qiwen Wang
8. Liyuan Wang

In Computer Vision and Pattern Recognition

1. Han-Jia Ye
2. De-chuan Zhan
3. Jun Zhu
4. Da-Wei Zhou
5. Ziwei Liu
6. Hang Su
7. Liyuan Wang
8. Xingxing Zhang

In Large Language Models

1. Liyuan Wang
2. Jun Zhu
3. Hang Su
4. Ziwei Liu
5. Keon-Hee Park
6. Kyungwoo Song
7. Gyeong-Moon Park
8. Xingxing Zhang

In Meta & Continual Learning

1. Han-Jia Ye
2. Xingxing Zhang
3. De-chuan Zhan
4. Liyuan Wang
5. Da-Wei Zhou
6. Jun Zhu
7. Hang Su
8. Qiwen Wang

Citations in the corpus

• A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, methodology): “Representative strategies include selecting the most relevant prompts from a prompt pool (L2P [458]), performing a weighted summation of the prompt pool with attention factors (CODAPrompt [402]), using explicitly task-sharing and task-specific prompts (DualPrompt [457]) or only task-specific prompts”
• Recent Advances of Continual Learning in Computer Vision: An Overview (2021, neutral): “It was pointed out by Zhai et al. [168] that this method is not memory efficient.”
• InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (2024, methodology): “For all the methods in all the experiments except for the comparison with SLCA, the batch size is set to 128 to follow many existing continual learning methods based on PEFT [4, 5].”
• One Size Fits All for Semantic Shifts: Adaptive Prompt Tuning for Continual Learning (2023, background): “Please see Appendix B for the details and complexity analysis.”
• A Closer Look at Rehearsal-Free Continual Learning * (2022, methodology): “Compared to the prompting methods L2P, DualPrompt, and the recent CODA-Prompt, L2 has clear and significant improvements, whereas EWC has poor performance.”
• Deep Class-Incremental Learning: A Survey (2023, background): “CIL with Pre-trained Models: Recently, pre-trained models have shown to work competitively with their strong transferability, especially for ViT-based methods [72], [73], [74], [75].”
• Consistent Prompting for Rehearsal-Free Continual Learning (2024, neutral): “Competitors We focus on comparing our model with the state-of-the-art prompt-based methods, i.e., L2P [46], Du-alPrompt [45], ESN [44] and CODAprompt [36].”
• Expandable Subspace Ensemble for Pre-Trained Model-Based Class-Incremental Learning (2024, methodology): “With a generalizable PTM as initialization, algorithms tend to freeze the pre-trained weight and append minimal additional parameters ( e.g ., prompts [31]) to accommodate incremental tasks [49, 60–62].”
• Class-Incremental Learning: A Survey (2023, background): “CIL with Pre-trained Models : Recently, pre-trained models have shown to work competitively with their strong transferability, especially for ViT-based methods [69], [70], [71], [72].”
• Continual Segmentation with Disentangled Objectness Learning and Class Recognition (2024, neutral): “Recently, more works [13, 37, 41, 41, 49, 50] started to leverage the dynamic nature of Transformers.”
• Pre-trained Vision and Language Transformers are Few-Shot Incremental Learners (2024, background): “Some recent approaches [9, 31, 32] have suggested meth-ods for generating prompts to adapt the domain space for effective continual learning.”
• A Unified Continual Learning Framework with General Parameter-Efficient Tuning (2023, neutral): “In the supplementary, we compare our LAE with the contemporary CODA-Prompt [44] approach on ImageNet-R and DomainNet [35] datasets.”