As an increasing number of data-driven deep learning models are deployed in our daily lives, the issue of algorithmic fairness has become a major concern. These models are trained on data that inevitably contains various biases, leading them to learn unfair representations that differ across demographic subgroups, resulting in unfair predictions. Previous work on fairness has attempted to remove subgroup information from learned features, aiming to contribute to similar representations across subgroups and lead to fairer predictions. However, identifying and removing this information is extremely challenging due to the "black box" nature of neural networks. Moreover, removing desired features without affecting other features is difficult, as features are often correlated, potentially harming model prediction performance. This paper aims to learn fair representations without degrading model prediction performance. We adopt knowledge distillation, allowing unfair models to learn fair representations directly from a fair teacher. The proposed method provides a novel approach to obtaining fair representations while maintaining valid prediction performance. We evaluate the proposed method, FairDistill, on four datasets (CIFAR-10, UTKFace, CelebA, and Adult) under diverse settings. Extensive experiments demonstrate the effectiveness and robustness of the proposed method.

While existing fairness interventions show promise in mitigating biased predictions, most studies concentrate on single-attribute protections. Although a few methods consider multiple attributes, they either require additional constraints or prediction heads, incurring high computational overhead or jeopardizing the stability of the training process. More critically, they consider per-attribute protection approaches, raising concerns about fairness gerrymandering where certain attribute combinations remain unfair. This work aims to construct a neutral domain containing fused information across all subgroups and attributes. It delivers fair predictions as the fused input contains neutralized information for all considered attributes. Specifically, we adopt mixup operations to generate samples with fused information. However, our experiments reveal that directly adopting the operations leads to degraded prediction results. The excessive mixup operations result in unrecognizable training data. To this end, we design three distinct mixup schemes that balance information fusion across attributes while retaining distinct visual features critical for training valid models. Extensive experiments with multiple datasets and up to eight sensitive attributes demonstrate that the proposed MultiFair method can deliver fairness protections for multiple attributes while maintaining valid prediction results.

As newly emerged powerful tools for question-answering tasks, large language models (LLMs) have attracted significant attention in recent years. Despite their superior performance on various tasks, LLMs often lack domain knowledge or up-to-date knowledge. Retrievalaugmented generation (RAG) has been proposed as a solution to supplement LLMs and mitigate hallucination. However, while RAG systems offer considerable improvements, they also introduce new attack surfaces and privacy risks for LLMs. Current RAG systems are considered as robust against privacy attacks as all malicious queries can be easily detected and defended. However, we find that a hidden privacy attack on RAG may jeopardize the privacy of RAG. To show the risk of RAG system, we propose RAGLeak, a stealthy novel attack method to infer membership information in RAG-based LLMs. Our method crops a part of the query, using the former part as the input and the latter part as the true answer. We investigate the grey-box setting where the attacker can access the perplexity and the black-box setting where the attacker only has the input/output access. For the grey-box setting, we decide the membership states by thresholding the output perplexity; for black-box setting, we calculate the similarity between the output and the true answer and set the similarity threshold. We evaluate our method on two datasets and three LLMs. The results show that RAGLeak can bypass few-shot based defense method and achieve an accuracy of over 0.8 on all datasets and LLMs. This competitive performance on RAG-based LLMs demonstrates that our method poses significant privacy risks without complicated finetuning or retraining. This work presents the potential risk of RAG on LLM and highlights the need for robust privacy-preserving techniques in RAG-based LLMs.

While in-processing fairness approaches show promise in mitigating biased predictions, their potential impact on privacy leakage remains under-explored. We aim to address this gap by assessing the privacy risks of fairness-enhanced binary classifiers via membership inference attacks (MIAs) and attribute inference attacks (AIAs). Surprisingly, our results reveal that enhancing fairness does not necessarily lead to privacy compromises. For example, these fairness interventions exhibit increased resilience against MIAs and AIAs. This is because fairness interventions tend to remove sensitive information among extracted features and reduce confidence scores for the majority of training data for fairer predictions. However, during the evaluations, we uncover a potential threat mechanism that exploits prediction discrepancies between fair and biased models, leading to advanced attack results for both MIAs and AIAs. This mechanism reveals potent vulnerabilities of fair models and poses significant privacy risks of current fairness methods. Extensive experiments across multiple datasets, attack methods, and representative fairness approaches confirm our findings and demonstrate the efficacy of the uncovered mechanism. Our study exposes the under-explored privacy threats in fairness studies, advocating for thorough evaluations of potential security vulnerabilities before model deployments.

While in-processing fairness approaches show promise in mitigating bias predictions, their potential impact on privacy leakage remains under-explored. We aim to address this gap by assessing the privacy risks of fairness-enhanced binary classifiers with membership inference attacks (MIAs). Surprisingly, our results reveal that these fairness interventions exhibit increased resilience against existing attacks, indicating that enhancing fairness does not necessarily lead to privacy compromises. However, we find current attack methods are ineffective as they typically degrade into simple threshold models with limited attack effectiveness. Following this observation, we discover a novel threat dubbed Fairness Discrepancy Membership Inference Attacks (FD-MIA) that exploits prediction discrepancies between fair and biased models. This attack reveals more potent vulnerabilities and poses significant privacy risks to model privacy. Extensive experiments across multiple datasets, attack methods, and representative fairness approaches confirm our findings and demonstrate the efficacy of the proposed attack method. Our study exposes the overlooked privacy threats in fairness studies, advocating for thorough evaluations of potential security vulnerabilities before model deployments.

Color is a powerful tool for communication across language and cultural boundaries. Through understanding color emotion and color psychology, designers can establish a deeper emotional connection with the audience and achieve more effective visual communication. The purpose of this paper is to introduce healing colors and their specific applications in visual communication design. In order to explore the specific application value of color healing in visual communication design, this paper mines product color information extracts colors and screen clusters according to the K-Means clustering method, and finally constructs product color data sets. Mining consumer emotional information, combined with the product color data set, calculates the consumer emotional weight and the sense of product safety experience through experimental technical means so as to judge the application value of healing color. According to the final results of the calculation of consumer safety experience, orange (0.6084) and green (0.5718) have higher weights, which also lead to higher product sales. This shows that healing colors affect consumers’ psychological and physiological feelings while also influencing product sales.

As a booming research area in the past decade, deep learning technologies have been driven by big data collected and processed on an unprecedented scale. However, privacy concerns arise due to the potential leakage of sensitive information from the training data. Recent research has revealed that deep learning models are vulnerable to various privacy attacks, including membership inference attacks, attribute inference attacks, and gradient inversion attacks. Notably, the efficacy of these attacks varies from model to model. In this paper, we answer a fundamental question: Does model architecture affect model privacy? By investigating representative model architectures from convolutional neural networks (CNNs) to Transformers, we demonstrate that Transformers generally exhibit higher vulnerability to privacy attacks than CNNs. Additionally, we identify the micro design of activation layers, stem layers, and LN layers, as major factors contributing to the resilience of CNNs against privacy attacks, while the presence of attention modules is another main factor that exacerbates the privacy vulnerability of Transformers. Our discovery reveals valuable insights for deep learning models to defend against privacy attacks and inspires the research community to develop privacy-friendly model architectures.

Previous studies have developed fairness methods for biased models that exhibit discriminatory behaviors towards specific subgroups. While these models have shown promise in achieving fair predictions, recent research has identified their potential vulnerability to score-based membership inference attacks (MIAs). In these attacks, adversaries can infer whether a particular data sample was used during training by analyzing the model's prediction scores. However, our investigations reveal that these score-based MIAs are ineffective when targeting fairness-enhanced models in binary classifications. The attack models trained to launch the MIAs degrade into simplistic threshold models, resulting in lower attack performance. Meanwhile, we observe that fairness methods often lead to prediction performance degradation for the majority subgroups of the training data. This raises the barrier to successful attacks and widens the prediction gaps between member and non-member data. Building upon these insights, we propose an efficient MIA method against fairness-enhanced models based on fairness discrepancy results (FD-MIA). It leverages the difference in the predictions from both the original and fairness-enhanced models and exploits the observed prediction gaps as attack clues. We also explore potential strategies for mitigating privacy leakages. Extensive experiments validate our findings and demonstrate the efficacy of the proposed method.

Color is one of the three major elements of print advertising, and different color combinations can trigger different emotional experiences of human beings. At present, the application of color in advertising in China is relatively mature, but it is limited to the traditional application method and has not been combined with big data technology. From the perspective of business needs, this research analyzes the process of visual creativity from the perspective of business value-added, and analyzes the role of big data in it. Then it introduces the semantics of common colors and how to incorporate color semantics into advertising design. And a sequence mining-based advertising click-through rate prediction model is proposed. The Criteo dataset is used as the training set. The AUC value of the model is 0.702 and the loss value is 0.415. Compared with other models, AUC values increased by 10.16%, 4.70%, 2.69% and 2.30%, respectively. Losses decreased by 10.17%, 9.19%, 6.11% and 7.57%, respectively. Finally, the online shopping data of 20 consumers was used as the test set to predict their color preferences, and the prediction accuracy was about 70%. Among them, the prediction accuracy of the group with stable shopping habits was 72.76%, and that of the group who liked to try new things was 70.60%, both meeting the expectation. Through experiments, it is concluded that the model has good performance and stability, and can more accurately judge consumers' consumption preferences.

With learning models widely deployed in daily life, researchers are discovering that many of them generate discriminatory predictions towards sensitive attributes such as gender or race. A common way of tackling this problem is to learn fair features without sensitive information by removing features. This method requires extra heads for attribute predictions and eliminates the information via adversarial settings. Although the processes can impose fairness, they reduce the accuracy of the models relative to the originals. In this research, we generate continuous domains containing information from different subgroups with mixup operations. We then learn domain invariant features with similarity constraints. Different from previous methods, the proposed method, CIFair, can learn fair features without feature removal operations or task-irrelevant learning objectives. Finally, we evaluated our approach on the CelebA dataset with different sensitive attributes under multiple settings. All experimental results demonstrate that CIFair is able to impose enforce better fair prediction results than previous methods while maintaining model accuracy performance.

In recent years, it has been revealed that machine learning models can produce discriminatory predictions. Hence, fairness protection has come to play a pivotal role in machine learning. In the past, most studies on fairness protection have used traditional machine learning methods to enforce fairness. However, these studies focus on low dimensional inputs, such as numerical inputs, whereas more recent deep learning technologies have encouraged fairness protection with image inputs through deep model methods. These approaches involve various object functions and structural designs that break the spurious correlations between targets and sensitive features. With these connections broken, we are left with fairer predictions. To better understand the proposed methods and encourage further development in the field, this paper summarizes fairness protection methods in terms of three aspects: the problem settings, the models, and the challenges. Through this survey, we hope to reveal research trends in the field, discover the fundamentals of enforcing fairness, and summarize the main challenges to producing fairer models.

Facial images are widely adopted for computer vision tasks such as face recognition or attribute classifications. Consequently, the adoption of mass real facial images leads to significant identification privacy leakage concerns. Meanwhile, the model classification results suffer unfair predictions towards features such as genders due to biased training data distributions. Although methods have been proposed to resolve the privacy and fairness issues separately, simultaneous protection methods are merely studied. In this study, for facial attributes classifications, we propose one unified framework with GAN models to generate synthetic images for privacy protections and contrastive learning based loss designs to enforce fairness protections simultaneously. Meanwhile, unlike other privacy or fairness protection methods, the proposed methods can maintain high data and model utilities. We evaluate our approaches with the high image resolution dataset CelebA-HD, and the results show our methods meet both privacy and fairness requirements.

As a booming research area in the past decade, deep learning technologies have been driven by big data collected and processed on an unprecedented scale. However, privacy concerns arise due to the potential leakage of sensitive information from the training data. Recent research has revealed that deep learning models are vulnerable to various privacy attacks, including membership inference attacks, attribute inference attacks, and gradient inversion attacks. Notably, the efficacy of these attacks varies from model to model. In this paper, we answer a fundamental question: Does model architecture affect model privacy? By investigating representative model architectures from CNNs to Transformers, we demonstrate that Transformers generally exhibit higher vulnerability to privacy attacks compared to CNNs. Additionally, We identify the micro design of activation layers, stem layers, and LN layers, as major factors contributing to the resilience of CNNs against privacy attacks, while the presence of attention modules is another main factor that exacerbates the privacy vulnerability of Transformers. Our discovery reveals valuable insights for deep learning models to defend against privacy attacks and inspires the research community to develop privacy-friendly model architectures.

Reliable detection of obstacles around the vehicle is crucial for autonomous cars. We present a novel and robust ground shape invariant feature method for general obstacle detection with a car-mounted monocular fisheye camera. Both stationary and moving obstacles can be detected by our approach without recovering the camera motion. Firstly, In order to compute the ground shape invariant feature, the image is mapped into the top view image. And then feature points are extracted and matched between adjacent frames. Secondly, the points are grouped according to image patch partition. Finally, the ground shape invariant feature is computed for each group of points to detect obstacle points. Extensive experiments have been carried out with prerecorded video sequences including various obstacle types, various scenes and various illumination conditions. The experimental results show promising detection performance of the proposed method.