Inter-subject Contrastive Learning for Subject Adaptive EEG-based Visual Recognition

Lee, Pilhyeon; Hwang, Sunhee; Lee, Jewook; Shin, Minjung; Jeon, Seogkyu; Byun, Hyeran

doi:10.1109/bci53720.2022.9734886

Cited by 8 publications

(13 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Several approaches [5], [31], [32] hinge on adversarial training-based strategies to encourage the subject invariance as well as the class discriminability of the feature representation during the training process. Meanwhile, there appear the studies on subject adaptation [12], [23], aiming at correctly recognizing the target class only with a few EEG signals from the new users using a sufficient number of pre-collected data from source subjects. To this end, they exploit domain adaptation methods [33], [34] and contrastive learning [35], [36] to narrow the representation gap between the source subjects and the target one.…”

Section: B Subject-adaptive Eeg-based Classificationmentioning

confidence: 99%

“…For this purpose, a variety of subject-independent training methods can be utilized. Here we adopt two recent approaches, namely MMDbased [12] and contrastive learning-based subject adaptation methods [23]. The overview of the target subject adaptation process of the target model M trg is shown in Fig.…”

Section: Target Subject Adaptationmentioning

confidence: 99%

“…However, they are generally dependent on subjects; that is, they perform well only for the existing users in the training database but show poor performance for new users. To handle this issue, the subject adaptation paradigm has recently been proposed to alleviate the cost of calibration process [12], [23], where only a handful of EEG samples are acquired from the target subject while abundant data from other source subjects are available. With sophisticated subject-independent feature learning, recent methods achieve promising performance [12], [23], demonstrating the potential for efficient adaptation of EEG-based visual recognition models to new users.…”

Section: Introductionmentioning

confidence: 99%

“…Then the generated samples serve as pseudo-source data to be utilized for subjectindependent representation learning along with the handful of target subject data. Our framework is general and therefore can adopt any subject-independent training method; we use two representative methods [12], [23] for the experiments. In the experimental results on EEG-ImageNet40 [14], we verify the effectiveness of the proposed framework under the source-free subject adaptation scenario.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Subject Adaptive EEG-Based Visual Recognition

Lee

Hwang²,

Jeon

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

This paper focuses on subject adaptation for EEGbased visual recognition. It aims at building a visual stimuli recognition system customized for the target subject whose EEG samples are limited, by transferring knowledge from abundant data of source subjects. Existing approaches consider the scenario that samples of source subjects are accessible during training. However, it is often infeasible and problematic to access personal biological data like EEG signals due to privacy issues. In this paper, we introduce a novel and practical problem setup, namely source-free subject adaptation, where the source subject data are unavailable and only the pre-trained model parameters are provided for subject adaptation. To tackle this challenging problem, we propose classifier-based data generation to simulate EEG samples from source subjects using classifier responses. Using the generated samples and target subject data, we perform subjectindependent feature learning to exploit the common knowledge shared across different subjects. Notably, our framework is generalizable and can adopt any subject-independent learning method. In the experiments on the EEG-ImageNet40 benchmark, our model brings consistent improvements regardless of the choice of subject-independent learning. Also, our method shows promising performance, recording top-1 test accuracy of 74.6% under the 5-shot setting even without relying on source data. Our code can be found at https://github.com/DeepBCI/Deep-BCI.

show abstract

Section: B Subject-adaptive Eeg-based Classificationmentioning

confidence: 99%

Section: Target Subject Adaptationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Subject Adaptive EEG-Based Visual Recognition

Lee

Hwang²,

Jeon

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…They effectively learn visual representation by drawing positive pairs and pushing away negative ones in the high-dimensional space. Despite being originally introduced for unsupervised learning, the contrastive learning strategy proves to be effective in various vision fields [21,29,30]. Particularly, SupCon [25] achieves better top-1 accuracy than the state-of-the-art methods based on the cross-entropy loss on ImageNet [41] by simply grafting the contrastive loss to the supervised representation learning.…”

Section: Introductionmentioning

confidence: 99%

Fair Contrastive Learning for Facial Attribute Classification

Park¹,

Lee²,

Lee³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Learning visual representation of high quality is essential for image classification. Recently, a series of contrastive representation learning methods have achieved preeminent success. Particularly, SupCon [25] outperformed the dominant methods based on cross-entropy loss in representation learning. However, we notice that there could be potential ethical risks in supervised contrastive learning. In this paper, we for the first time analyze unfairness caused by supervised contrastive learning and propose a new Fair Supervised Contrastive Loss (FSCL) for fair visual representation learning. Inheriting the philosophy of supervised contrastive learning, it encourages representation of the same class to be closer to each other than that of different classes, while ensuring fairness by penalizing the inclusion of sensitive attribute information in representation. In addition, we introduce a group-wise normalization to diminish the disparities of intragroup compactness and inter-class separability between demographic groups that arouse unfair classification. Through extensive experiments on CelebA and UTK Face, we validate that the proposed method significantly outperforms SupCon and existing state-of-the-art methods in terms of the trade-off between top-1 accuracy and fairness. Moreover, our method is robust to the intensity of data bias and effectively works in incomplete supervised settings. Our code is available at https://github.com/sungho-CoolG/FSCL.

show abstract