Saliency Guided Experience Packing for Replay in Continual Learning

被引:4
|
作者
Saha, Gobinda [1 ]
Roy, Kaushik [1 ]
机构
[1] Purdue Univ, Elmore Family Sch Elect & Comp Engn, W Lafayette, IN 47907 USA
来源
2023 IEEE/CVF WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV) | 2023年
基金
美国国家科学基金会;
关键词
D O I
10.1109/WACV56688.2023.00524
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Artificial learning systems aspire to mimic human intelligence by continually learning from a stream of tasks without forgetting past knowledge. One way to enable such learning is to store past experiences in the form of input examples in episodic memory and replay them when learning new tasks. However, performance of such method suffers as the size of the memory becomes smaller. In this paper, we propose a new approach for experience replay, where we select the past experiences by looking at the saliency maps which provide visual explanations for the model's decision. Guided by these saliency maps, we pack the memory with only the parts or patches of the input images important for the model's prediction. While learning a new task, we replay these memory patches with appropriate zero-padding to remind the model about its past decisions. We evaluate our algorithm on CIFAR-100, miniImageNet and CUB datasets and report better performance than the state-of-the-art approaches. With qualitative and quantitative analyses we show that our method captures richer summaries of past experiences without any memory increase, and hence performs well with small episodic memory.
引用
收藏
页码:5262 / 5272
页数:11
相关论文
共 50 条
  • [31] Contrastive Correlation Preserving Replay for Online Continual Learning
    Yu, Da
    Zhang, Mingyi
    Li, Mantian
    Zha, Fusheng
    Zhang, Junge
    Sun, Lining
    Huang, Kaiqi
    IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, 2024, 34 (01) : 124 - 139
  • [32] An Investigation of Replay-based Approaches for Continual Learning
    Bagus, Benedikt
    Gepperth, Alexander
    2021 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), 2021,
  • [33] Latent Replay for Real-Time Continual Learning
    Pellegrini, Lorenzo
    Graffieti, Gabriele
    Lomonaco, Vincenzo
    Maltoni, Davide
    2020 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS), 2020, : 10203 - 10209
  • [34] Effective Data Selection and Replay for Unsupervised Continual Learning
    Liu, Hanmo
    Di, Shimin
    Li, Haoyang
    Li, Shuangyin
    Chen, Lei
    Zhou, Xiaofang
    Proceedings - International Conference on Data Engineering, 2024, : 1449 - 1463
  • [35] Towards Causal Replay for Knowledge Rehearsal in Continual Learning
    Churamani, Nikhil
    Cheong, Jiaee
    Kalkan, Sinan
    Gunes, Hatice
    AAAI BRIDGE PROGRAM ON CONTINUAL CAUSALITY, VOL 208, 2023, 208 : 63 - 70
  • [36] Replay-Driven Continual Learning for the Industrial Internet of Things
    Sen, Sagar
    Nielsen, Simon Myklebust
    Husom, Erik Johannes
    Goknil, Arda
    Tverdal, Simeon
    Pinilla, Leonardo Sastoque
    2023 IEEE/ACM 2ND INTERNATIONAL CONFERENCE ON AI ENGINEERING - SOFTWARE ENGINEERING FOR AI, CAIN, 2023, : 43 - 55
  • [37] CONTINUAL LEARNING OF NEW SOUND CLASSES USING GENERATIVE REPLAY
    Wang, Zhepei
    Subakan, Cem
    Tzinis, Efthymios
    Smaragdis, Paris
    Charlin, Laurent
    2019 IEEE WORKSHOP ON APPLICATIONS OF SIGNAL PROCESSING TO AUDIO AND ACOUSTICS (WASPAA), 2019, : 308 - 312
  • [38] Continual Learning of Medical Image Classification Based on Feature Replay
    Li, Xiaojie
    Li, Haifeng
    Ma, Lin
    2022 16TH IEEE INTERNATIONAL CONFERENCE ON SIGNAL PROCESSING (ICSP2022), VOL 1, 2022, : 426 - 430
  • [39] Generative feature-driven image replay for continual learning
    Thandiackal, Kevin
    Portenier, Tiziano
    Giovannini, Andrea
    Gabrani, Maria
    Goksel, Orcun
    IMAGE AND VISION COMPUTING, 2024, 150
  • [40] Selective Replay Enhances Learning in Online Continual Analogical Reasoning
    Hayes, Tyler L.
    Kanan, Christopher
    2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS, CVPRW 2021, 2021, : 3497 - 3507