Imbalanced Continual Learning with Partitioning Reservoir Sampling

By Chris Dongjoo Kim, Jinseo Jeong, Gunhee Kim. 2020

136 citations recorded by the source corpus.

Continual learning from a sequential stream of data is a crucial challenge for machine learning research. Most studies have been conducted on this topic under the single-label classification setting along with an assumption of balanced label distribution. This work expands this research horizon towards multi-label classification. In doing so, we identify unanticipated adversity innately existent in many multi-label datasets, the long-tailed distribution. We jointly address the two independently solved problems, Catastropic Forgetting and the long-tailed label distribution by first empirically showing a new challenge of destructive forgetting of the minority concepts on the tail. Then, we curate two benchmark datasets, COCOseq and NUS-WIDEseq, that allow the study of both intra- and inter-task imbalances. Lastly, we propose a new sampling strategy for replay-based approach named Partitioning Reservoir Sampling (PRS), which allows the model to maintain a balanced knowledge of both head and tail classes. We publicly release the dataset and the code in our project page.

Subfields: Artificial Intelligence, 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. Tinne Tuytelaars
7. Hang Su
8. Qiwen Wang

In Meta & Continual Learning

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

Citations in the corpus

• A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “Besides, other realistic consider-ationshavebeenincorporated,suchasmultiplelabels [29],noisy labels [30], blurred boundary [31], [32], hierarchical granularity [33] and sub-populations [34], mixture of task similarity [35], long-tailed distribution [36], novel class discovery [37], [38],…”
• Long-Tailed Class Incremental Learning (2022, background): “[17] proposed a multi-label classification problem with long-tailed distribution.”
• Information-theoretic Online Memory Selection for Continual Learning (2022, neutral): “RS is generalized to class-balanced RS (CBRS) that selects uniform samples over each class (Chrysakis & Moens, 2020; Wiewel & Yang, 2021; Kim et al., 2020), but CBRS cannot resolve other forms of data imbalance.”
• Continual Normalization: Rethinking Batch Normalization for Online Continual Learning (2022, methodology): “We consider the split Mini IMN with ER and the COCOseq benchmark with PRS and report the result in Table 8.”
• Online Continual Learning on Class Incremental Blurry Task Configuration with Anytime Inference (2021, background): “We focus extensively on the task-free setup as it is challenging and being actively investigated recently (Kim et al., 2020; Lee et al., 2019; Aljundi et al., 2019c).”
• Continual Learning on Noisy Data Streams via Self-Purified Replay (2021, methodology): “We include all of GDumb [68] combinations and the single best performing combination of PRS [37] and CRS [86] for each dataset.”
• Recent Advances of Continual Learning in Computer Vision: An Overview (2021, neutral): “They thus proposed Incremental Learning with Dual Memory which has a second memory to store statistics of the previous tasks.”
• Total Recall: a Customized Continual Learning Method for Neural Semantic Parsers (2021, background): “…to reduce the catastrophic forgetting; iii) memory-based methods (Lopez-Paz and Ranzato, 2017; Wang et al., 2019; Han et al., 2020; Aljundi et al., 2019; Chrysakis and Moens, 2020; Kim et al., 2020) which stores historical instances and continually learn them along with instances in new tasks.”
• Replay in Deep Learning: Current Approaches and Missing Biological Elements (2021, neutral): “…Aljundi, Lin, Goujaud, & Bengio, 2019; He, Mao, Shao, & Zhu, 2020; Zhao, Xiao, Gan, Zhang, & Xia, 2020; Kurle, Cseke, Klushyn, van der Smagt, & Günnemann, 2020; Chrysakis & Moens, 2020; Kim, Jeong, & Kim, 2020; Tao, Chang, Hong, Wei, & Gong, 2020; Douil-lard, Cord, Ollion, Robert, & Valle, 2020).”
• Supervised Contrastive Replay: Revisiting the Nearest Class Mean Classifier in Online Class-Incremental Continual Learning (2021, methodology): “Although random seems reasonable for the balanced CIFAR100 dataset, when facing imbalanced datasets, combing SCR with other memory management methods may yield better performance [27, 13].”
• Online Continual Learning in Image Classification: An Empirical Survey (2021, background): “What differentiate various memory-based methods are the memory retrieval strategy (line 3) [17, 85], model update (line 4) [31, 13] and memory update strategy (line 5) [61, 63, 18].”
• Continual Prototype Evolution: Learning Online from Non-Stationary Data Streams (2020, methodology): “As this is computationally expensive, recent works explore other balancing schemes [24, 8], where we propose a simple but effective class-based Reservoir scheme with uniform retrieval.”
• The CLEAR Benchmark: Continual LEArning on Real-World Imagery (2022, background): “This uniform sampling procedure has shown to work well in prior works [8, 48] as well as some variants that tackle class-imbalanced data streams [2, 26].”
• Mitigating Forgetting in Online Continual Learning with Neuron Calibration (2021, methodology): “The existing replay-based approaches have taken into account of different perspectives of the model training process to remedy the stability-plasticity dilemma, such as regularizing the parameter change during training [4, 12, 23], selective memory storage or replay [1, 11], Bayesian and variational”
• Deep Class-Incremental Learning: A Survey (2023, background): “[212], [213], [214] propose to tackle the long-tailed class-incremental learning problem, where the head classes are easy to collect with adequate instances while tail classes are scarce.”