Generative Models from the perspective of Continual Learning

Lesort, Timothée; Caselles-Dupré, Hugo; Garcia-Ortiz, Michael; Stoian, Andrei; Filliat, David

doi:10.48550/arxiv.1812.09111

Cited by 7 publications

(15 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We do not compare model-based methods, because databased methods are known to outperform them in classincremental learning [22,41], and they are orthogonal to data-based methods, such that they can potentially be combined with our approaches for better performance [15]. Datasets.…”

Section: Methodsmentioning

confidence: 99%

“…Therefore, the amount of knowledge kept by knowledge distillation depends on the degree of similarity between the data distribution used to learn the previous tasks in the previous stages and the one used to distill the knowledge in the later stages. To guarantee to have a certain amount of similar data, some prior works [3,30,33] reserved a small amount of memory to keep a coreset, and others [22,32,38,41,42] trained a generative model and replay the generated data when training a new model. Note that the model-based and data-based approaches are orthogonal in most cases, thus they can be combined for better performance [15].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Overcoming Catastrophic Forgetting With Unlabeled Data in the Wild

Lee

Shin

et al. 2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

136

View full text Add to dashboard Cite

Lifelong learning with deep neural networks is wellknown to suffer from catastrophic forgetting: the performance on previous tasks drastically degrades when learning a new task. To alleviate this effect, we propose to leverage a large stream of unlabeled data easily obtainable in the wild. In particular, we design a novel classincremental learning scheme with (a) a new distillation loss, termed global distillation, (b) a learning strategy to avoid overfitting to the most recent task, and (c) a confidence-based sampling method to effectively leverage unlabeled external data. Our experimental results on various datasets, including CIFAR and ImageNet, demonstrate the superiority of the proposed methods over prior methods, particularly when a stream of unlabeled data is accessible: our method shows up to 15.8% higher accuracy and 46.5% less forgetting compared to the state-of-the-art method. The code is available at https://github. comB. We design a 3-step learning scheme to improve the effectiveness of global distillation: (i) training a teacher specialized for the current task, (ii) training a model by distilling the knowledge of the previous model, the teacher learned in (i), and their ensemble, and (iii) finetuning to avoid overfitting to the current task.C. We propose a confidence-based sampling method to effectively leverage a large stream of unlabeled data.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Overcoming Catastrophic Forgetting With Unlabeled Data in the Wild

Lee

Shin

et al. 2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

136

View full text Add to dashboard Cite

show abstract

“…Other approaches using distillation loss include the encoder-based method (Rannen et al 2017) and incremental moment matching (Lee et al 2017). Some of the recent papers also consider generating data from previous tasks using a deep generative model (Lesort et al 2018;Shin et al 2017;van de Ven and Tolias 2018). Then, the model for the main task can be updated in a multi-task learning fashion using both the generated data and the data of the new task.…”

Section: Related Workmentioning

confidence: 99%

Lifelong Object Detection

Zhou,

Chang,

Sosa

et al. 2020

Preprint

View full text Add to dashboard Cite

Recent advances in object detection have benefited significantly from rapid developments in deep neural networks. However, neural networks suffer from the well-known issue of catastrophic forgetting, which makes continual or lifelong learning problematic. In this paper, we leverage the fact that new training classes arrive in a sequential manner and incrementally refine the model so that it additionally detects new object classes in the absence of previous training data. Specifically, we consider the representative object detector, Faster R-CNN, for both accurate and efficient prediction. To prevent abrupt performance degradation due to catastrophic forgetting, we propose to apply knowledge distillation on both the region proposal network and the region classification network, to retain the detection of previously trained classes. A pseudo-positive-aware sampling strategy is also introduced for distillation sample selection. We evaluate the proposed method on PASCAL VOC 2007 and MS COCO benchmarks and show competitive mAP and 6x inference speed improvement, which makes the approach more suitable for real-time applications. Our implementation will be publicly available.

show abstract

“…While most work in CL is focused on discriminative models, there is a recent interest in CL approaches for generative models (Achille et al, 2018;Nguyen et al, 2017). Proposed methods often rely on using generated samples to avoid forgetting (Wu et al, 2018;Lesort et al, 2018a). This technique, which we use in this paper, is termed Generative Replay.…”

Section: Continual State Representation Learning For Rlmentioning

confidence: 99%

“…(Graves et al, 2018) and others) that are widely used for SRL. Previous work have shown that VAEs can learn continually using generated samples from previous tasks, a method called Generative Replay (Lesort et al, 2018a). S-TRIGGER uses Generative Replay to remember information relative to previously encountered environments.…”

Section: Introductionmentioning

confidence: 99%

S-TRIGGER: Continual State Representation Learning via Self-Triggered Generative Replay

Caselles-Dupré¹,

Garcia-Ortiz²,

Filliat³

2019

Preprint

Self Cite

View full text Add to dashboard Cite

We consider the problem of building a state representation model for control, in a continual learning setting. As the environment changes, the aim is to efficiently compress the sensory state's information without losing past knowledge, and then use Reinforcement Learning on the resulting features for efficient policy learning. To this end, we propose S-TRIGGER, a general method for Continual State Representation Learning applicable to Variational Auto-Encoders and its many variants. The method is based on Generative Replay, i.e. the use of generated samples to maintain past knowledge. It comes along with a statistically sound method for environment change detection, which self-triggers the Generative Replay. Our experiments on VAEs show that S-TRIGGER learns state representations that allows fast and high-performing Reinforcement Learning, while avoiding catastrophic forgetting. The resulting system is capable of autonomously learning new information without using past data and with a bounded system size. Code for our experiments is attached in Appendix.

show abstract

Generative Models from the perspective of Continual Learning

Cited by 7 publications

References 0 publications

Overcoming Catastrophic Forgetting With Unlabeled Data in the Wild

Overcoming Catastrophic Forgetting With Unlabeled Data in the Wild

Lifelong Object Detection

S-TRIGGER: Continual State Representation Learning via Self-Triggered Generative Replay

Contact Info

Product

Resources

About