Additive Adversarial Learning for Unbiased Authentication

Abstract

Authentication is a task aiming to confirm the truth between data instances and personal identities. Typical authentication applications include face recognition, person re-identification, authentication based on mobile devices and so on. The recently-emerging data-driven authentication process may encounter undesired biases, i.e., the models are often trained in one domain (e.g., for people wearing spring outfits) while required to apply in other domains (e.g., they change the clothes to summer outfits). To address this issue, we propose a novel two-stage method that disentangles the class/identity from domaindifferences, and we consider multiple types of domaindifference. In the first stage, we learn disentangled representations by a one-versus-rest disentangle learning (OVRDL) mechanism. In the second stage, we improve the disentanglement by an additive adversarial learning (AAL) mechanism. Moreover, we discuss the necessity to avoid a learning dilemma due to disentangling causally related types of domain-difference. Comprehensive evaluation results demonstrate the effectiveness and superiority of the proposed method.

Selection Bias Explorations and Debias Methods for Natural Language Sentence Matching Datasets

“…Usually the propensity scores are estimated through online Result Randomization (Joachims et al, 2017). Liang et al (2019) studied the biasedness for authentication, and proposed an additive adversarial learning for unbiased learning.…”

A Comprehensive Study on Face Recognition Biases Beyond Demographics

“…It is designed to reduce overall intrapersonal variations while enlarging interpersonal differences simultaneously. Recent works aimed at mitigating demographic-bias in face recognition through adversarial learning [27], [42], [48], [47], margin-based approaches [79], [35], data augmentation [80], [40], [81], metric-learning [73], or score normalization [72].…”

“…Driven by (a) these legal efforts to guarantee fairness and (b) the findings that the performance of current face recognition solutions depends on the user's demographics, several approaches were proposed to mitigate demographics-bias in face recognition technologies. This was achieved through adversarial learning [27], [42], [48], [47], margin-based approaches [79], [35], data augmentation [80], [40], [81], metric-learning [73], or score normalization [72]. However, the research focus on demographic-bias does only tackle a minor proportion of all possible discriminatory effects.…”

Adversarial Learning with Bidirectional Attention for Visual Question Answering

“…The method of adversarial learning is inspired by human perception mechanisms [18]. In the process of answering questions, the more suitable and reasonable answers are given out by people comparing the differences between the information to find contradictions and differences [19][20][21][22]. Adversarial learning has made a series of good achievements in the field of artificial intelligence.…”