Incremental Learning in Online Scenario

By Jiangpeng He, Runyu Mao, Zeman Shao, Feng Zhu. 2020

186 citations recorded by the source corpus.

Modern deep learning approaches have achieved great success in many vision applications by training a model using all available task-specific data. However, there are two major obstacles making it challenging to implement for real life applications: (1) Learning new classes makes the trained model quickly forget old classes knowledge, which is referred to as catastrophic forgetting. (2) As new observations of old classes come sequentially over time, the distribution may change in unforeseen way, making the performance degrade dramatically on future data, which is referred to as concept drift. Current state-of-the-art incremental learning methods require a long time to train the model whenever new classes are added and none of them takes into consideration the new observations of old classes. In this paper, we propose an incremental learning framework that can work in the challenging online learning scenario and handle both new classes data and new observations of old classes. We address problem (1) in online mode by introducing a modified cross-distillation loss together with a two-step learning technique. Our method outperforms the results obtained from current state-of-the-art offline incremental learning methods on the CIFAR-100 and ImageNet-1000 (ILSVRC 2012) datasets under the same experiment protocol but in online scenario. We also provide a simple yet effective method to mitigate problem (2) by updating exemplar set using the feature of each new observation of old classes and demonstrate a real life application of online food image classification based on our complete framework using the Food-101 dataset.

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. Dacheng Tao
2. Jun Zhu
3. Mohit Bansal
4. Hang Su
5. Ali Farhadi
6. Bernt Schiele
7. Jianfeng Gao
8. Xingxing Zhang

In Meta & Continual Learning

1. Xingxing Zhang
2. Liyuan Wang
3. Hang Su
4. Jun Zhu
5. Timothée Lesort
6. Dacheng Tao
7. Irina Rish
8. Jiangpeng He

Citations in the corpus

• A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, methodology): “…memory buffer, representative strategies include calibrating features with task-specific parameters [277], performing knowledge distillation [31], [278], improving representations with contrastive learning [276], [279], using asymmetric cross-entropy [280] or constrained gradient directions [45],…”
• Online Continual Learning with Contrastive Vision Transformer (2022, background): “This paper focuses on a more realistic and challenging setting: online continual learning (online CL) [34,50,67], where the model learns a sequence of classification tasks with a single pass over the data and without task boundaries.”
• Online Continual Learning on Class Incremental Blurry Task Configuration with Anytime Inference (2021, background): “Despite being more realistic (Losing et al., 2018; He et al., 2020), online CL have not been popular (Prabhu et al., 2020) due to the difficulty and subtle differences in the setups in the literature.”
• Online Continual Learning Via Candidates Voting (2021, methodology): “(1) For methods using original data as exemplars [34, 2, 3, 29, 8, 9, 7, 35, 6, 42, 45, 13], the storage requirement for storing data in exemplar set is O(3 × S(2) × Q × C).”
• Recent Advances of Continual Learning in Computer Vision: An Overview (2021, neutral): “[59] pointed out that concept drift is much less explored in previous continual learning studies compared to catastrophic forgetting.”
• Replay in Deep Learning: Current Approaches and Missing Biological Elements (2021, neutral): “…& Popescu, 2019, 2020; Chaudhry et al., 2019; Riemer et al., 2019; Aljundi, Belilovsky et al., 2019; 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;…”
• Online Continual Learning in Image Classification: An Empirical Survey (2021, neutral): “WA[68], BiC[42], LUCIR[69] 3 3 IL2M[70], ILO[71]”
• MgSvF: Multi-Grained Slow versus Fast Framework for Few-Shot Class-Incremental Learning (2020, neutral): “Recently, there has been a large body of research in incremental learning [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43].”
• Continual Representation Learning for Biometric Identification (2020, methodology): “As shown in the most recent works [16, 32, 78], LWF is still a competitive method under the same memory setting.”
• Continual Learning With Structured Inheritance for Semantic Segmentation in Aerial Imagery (2021, methodology): “Moreover, some works store samples [26]–[30] or generate high-quality pseudo-samples [46], [47] of old tasks to further alleviate the catastrophic forgetting problem.”
• Continual Learning with Foundation Models: An Empirical Study of Latent Replay (2022, methodology): “5, pre-training on the CLIP dataset (400M examples) tends to produce better results on both large architectural families (ResNet101 and ViT).”
• In the Wild: From ML Models to Pragmatic ML Systems (2020, background): “Recent variants [15, 2, 37, 16] learn continuously in a task-free sequential setting but focus on slowly accumulating knowledge throughout a stream and do not include open-world & few-shot challenges.”
• Online Continual Learning For Visual Food Classification (2021, methodology): “proposed ILIO [9], which applies an accommodation ratio to generate a stronger constraint for knowledge distillation loss to achieve improved performance.”
• Self-Supervised Features Improve Open-World Learning (2021, background): “[26] tried to address the incremental learning in a more challenging online setting where update time and available data are limited.”
• Online Prototype Learning for Online Continual Learning (2023, background): “The anchor of OPE as well as its positive and negative samples are online prototypes, which means no instance is involved, while PCL takes instance-level representation as the anchor and cluster centers as the positive and negative samples.”
• Wandering Within a World: Online Contextualized Few-Shot Learning (2020, background): “Incremental learning (Castro et al., 2018; He et al., 2020; Hou et al., 2019; Rebuffi et al., 2017; Wu et al., 2019) is a form of continual learning, where each task is an incremental batch of several new classes.”
• Orchestrate Latent Expertise: Advancing Online Continual Learning with Multi-Level Supervision and Reverse Self-Distillation (2024, background): “In OCL, a model needs to learn a one-pass data stream with shifting distribution [24, 41, 52].”
• Unsupervised Continual Learning Via Pseudo Labels (2021, methodology): “…Hoiem, 2017 ] , ICARL [ Rebuffi et al. , 2017 ] , EEIL [ Castro et al. , 2018 ] , LUCIR [ Hou et al. , 2019 ] , WA [ Zhao et al. , 2020 ] and ILIO [ He et al. , 2020 ] , which are representative methods from all Regularization based , Bias-correction based and Rehearsal based categories as…”
• Are We Overfitting to Experimental Setups in Recognition (2020, background): “Recent variants have proposed a task-free setting where the data distribution changes without the model’s knowledge [15, 38, 16, 52].”