This study investigates the trade-offs between fairness, privacy, and utility in image classification using machine learning (ML). Recent research suggests that generalization techniques can improve the balance between privacy and utility. One focus of this work is sharpness-aware training (SAT) and its integration with differential privacy (DP-SAT) to further improve this balance. Additionally, we examine fairness in both private and non-private learning models trained on datasets with synthetic and real-world biases. We also measure the privacy risks involved in these scenarios by performing membership inference attacks (MIAs) and explore the consequences of eliminating high-privacy risk samples, termed outliers. Moreover, we introduce a new metric, named harmonic score, which combines accuracy, privacy, and fairness into a single measure. Through empirical analysis using generalization techniques, we achieve an accuracy of 81.11% under (8, 10^-5)-DP on CIFAR-10, surpassing the 79.5% reported by De et al. (2022). Moreover, our experiments show that memorization of training samples can begin before the overfitting point, and generalization techniques do not guarantee the prevention of this memorization. Our analysis of synthetic biases shows that generalization techniques can amplify model bias in both private and non-private models. Additionally, our results indicate that increased bias in training data leads to reduced accuracy, greater vulnerability to privacy attacks, and higher model bias. We validate these findings with the CelebA dataset, demonstrating that similar trends persist with real-world attribute imbalances. Finally, our experiments show that removing outlier data decreases accuracy and further amplifies model bias.

This paper explores the application of large language models (LLMs), like ChatGPT, for biometric tasks. We specifically examine the capabilities of ChatGPT in performing biometric-related tasks, with an emphasis on face recognition, gender detection, and age estimation. Since biometrics are considered as sensitive information, ChatGPT avoids answering direct prompts, and thus we crafted a prompting strategy to bypass its safeguard and evaluate the capabilities for biometrics tasks. Our study reveals that ChatGPT recognizes facial identities and differentiates between two facial images with considerable accuracy. Additionally, experimental results demonstrate remarkable performance in gender detection and reasonable accuracy for the age estimation tasks. Our findings shed light on the promising potentials in the application of LLMs and foundation models for biometrics.

Face inpainting, the technique of restoring missing or damaged regions in facial images, is pivotal for applications like face recognition in occluded scenarios and image analysis with poor-quality captures. This process not only needs to produce realistic visuals but also preserve individual identity characteristics. The aim of this paper is to inpaint a face given periocular region (eyes-to-face) through a proposed new Generative Adversarial Network (GAN)-based model called Eyes-to-Face Network (E2F-Net). The proposed approach extracts identity and non-identity features from the periocular region using two dedicated encoders have been used. The extracted features are then mapped to the latent space of a pre-trained StyleGAN generator to benefit from its state-of-the-art performance and its rich, diverse and expressive latent space without any additional training. We further improve the StyleGAN's output to find the optimal code in the latent space using a new optimization for GAN inversion technique. Our E2F-Net requires a minimum training process reducing the computational complexity as a secondary benefit. Through extensive experiments, we show that our method successfully reconstructs the whole face with high quality, surpassing current techniques, despite significantly less training and supervision efforts. We have generated seven eyes-to-face datasets based on well-known public face datasets for training and verifying our proposed methods. The code and datasets are publicly available1.

Online Social Networks are responsible for disclosing a large amount of sensitive information. Often, users unknowingly disclose vast amounts of sensitive and potentially (un)related data, oblivious to the associated privacy risks. Our research provides a comprehensive evaluation of the linkability between user profiles and shared content across various OSNs, a factor that has considerable implications for privacy leakage. We introduce a novel method for quantifying the linkability between profiles across multiple networks, based on key features and metrics that capture profile similarities. We applied this methodology to a dataset of user profiles across three online social networks named Flickr, Facebook, and Twitter. Our approach includes examining both structured and unstructured data related to user profiles, enabling us to offer a valuable understanding of linkability trends and identify potential privacy risks. Through our findings, we aim to inform the development of privacy-enhancing technologies and contribute to improving the current privacy landscape within OSNs. Our research underscores the critical need for robust privacy measures in the face of the growing interconnectedness of user data across different social networks.

This study introduces a new technique for generating synthetic faces using eyes-to-face inpainting methods. The proposed method can synthesize a face image using a combination of the eyes of two different individuals and use it as an input for inpainting, demonstrating its vast potential for various applications in biometrics. Despite minor biases in age and gender, our method proved effective in training reliable age- and gender-detection models using the generated datasets. We also addressed the challenge of training face recognition models using synthetic datasets, and the results demonstrated satisfactory accuracy across four benchmark face recognition datasets. This method could be particularly beneficial for underrepresented groups, for whom there is a scarcity of face samples in biometric datasets.

This paper presents the summary of the Efficient Face Recognition Competition (EFaR) held at the 2023 International Joint Conference on Biometrics (IJCB 2023). The competition received 17 submissions from 6 different teams. To drive further development of efficient face recognition models, the submitted solutions are ranked based on a weighted score of the achieved verification accuracies on a diverse set of benchmarks, as well as the deployability given by the number of floating-point operations and model size. The evaluation of submissions is extended to bias, cross-quality, and large-scale recognition benchmarks. Overall, the paper gives an overview of the achieved performance values of the submitted solutions as well as a diverse set of baselines. The submitted solutions use small, efficient network architectures to reduce the computational cost, some solutions apply model quantization. An outlook on possible techniques that are underrepresented in current solutions is given as well.

Online Social Networks are responsible for disclosing a large amount of sensitive information. Users unintentionally reveal their sensitive information and are unaware of the privacy risks involved. But the users should be well informed about their privacy quotient and should know where they stand on the privacy measuring scale. In this paper, we proposed an adaptive privacy measuring framework called PriMe that can measure the privacy leakage score for each action of a user in an OSN and subsequently adjust the privacy settings based on the preferred privacy scopes and boundaries. Various types of data, actions, and personal characteristics of each user have been considered to ensure the calculated privacy leakage score is accurate.

Abstract An important subfield of brain–computer interface is the classification of motor imagery (MI) signals where a presumed action, for example, imagining the hands' motions, is mentally simulated. The brain dynamics of MI is usually measured by electroencephalography (EEG) due to its noninvasiveness. The next generation of brain–computer interface systems can benefit from the generative deep learning (GDL) models by providing end‐to‐end (e2e) machine learning and increasing their accuracy. In this study, to exploit the e2e‐property of deep learning models, a novel GDL methodology is proposed where only minimal objective‐free preprocessing steps are needed. Furthermore, to deal with the complicated multi‐class MI–EEG signals, an innovative multilevel GDL‐based classifying scheme is proposed. The effectiveness of the proposed model and its robustness against noisy MI–EEG signals is evaluated using two different GDL models, that is, deep belief network and stacked sparse autoencoder in e2e manner. Experimental results demonstrate the effectiveness of the proposed methodology with improved accuracy compared with the widely used filter bank common spatial patterns algorithm.