SELFIE: Refurbishing Unclean Samples for Robust Deep Learning

By Hwanjun Song, Minseok Kim, Jae-Gil Lee. 2019

467 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Jiaheng Wei
2. Lu Jiang
3. Chris Dongjoo Kim
4. Tongliang Liu
5. Gang Niu
6. Fang Wen
7. Zhaowei Zhu
8. Bo Zhang

Citations in the corpus

• Continual Learning on Noisy Data Streams via Self-Purified Replay (2021, background): “Techniques that repair labels [39, 80, 50, 74, 27, 56] or directly learn them [76, 88] are also viable options for learning from noisy labeled data.”
• From Synthetic to Real: Unsupervised Domain Adaptation for Animal Pose Estimation (2021, background): “This small-loss trick , which is based on the memorization effect [3, 42] of deep networks, has also adopted by other works [17, 30] to select more reliable labels.”
• Early-Learning Regularization Prevents Memorization of Noisy Labels (2020, methodology): “SELFIE [35] uses label replacement to correct a subset of the labels selected by considering past model outputs.”
• Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (2021, background): “Self-label correction [51, 61, 73, 74] is a more appealing approach.”
• Data collection and quality challenges in deep learning: a data-centric AI perspective (2021, methodology): “SELFIE introduces re-furbishable samples where labels can be corrected with high precision.”
• Robust Learning by Self-Transition for Handling Noisy Labels (2020, methodology): “SELFIE [33] trains a DNN on selectively refurbished samples together with small-loss samples.”
• Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (2020, unknown): “For example, one usually needs to tune a warm-up epoch (Tanaka et al. 2018; Song, Kim, and Lee 2019; Arazo et al. 2019) before which no label correction is applied.”
• Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations (2021, background): “• Interventions: human interventions in data generation [Jiang et al., 2020] and non-representative data collection process [Song et al., 2019] might disturb the original noisy-label pattern.”
• Beyond Synthetic Noise: Deep Learning on Controlled Noisy Labels (2019, background): “, 2017), label cleaning/correction (Reed et al., 2014; Goldberger & BenReuven, 2017; Li et al., 2017b; Veit et al., 2017; Song et al., 2019), example weighting (Jiang et al.”