2020/10/02 by Vito Walter Anelli, Tommaso Di Noia, Anelli, Vito Walter +5
Computer Science · #Adversarial Robustness in Machine Learning #Anomaly Detection Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Information Retrieval (cs.IR) #Machine Learning (cs.LG) #Multimedia (cs.MM)
paper · pdf · doi:10.48550/arxiv.2010.00984
openalex publication_date 2020/10/02 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Visual-based recommender systems (VRSs) enhance recommendation performance by\nintegrating users' feedback with the visual features of product images\nextracted from a deep neural network (DNN). Recently, human-imperceptible\nimages perturbations, defined \adversarial attacks, have been\ndemonstrated to alter the VRSs recommendation performance, e.g., pushing/nuking\ncategory of products. However, since adversarial training techniques have\nproven to successfully robustify DNNs in preserving classification accuracy, to\nthe best of our knowledge, two important questions have not been investigated\nyet: 1) How well can these defensive mechanisms protect the VRSs performance?\n2) What are the reasons behind ineffective/effective defenses? To answer these\nquestions, we define a set of defense and attack settings, as well as\nrecommender models, to empirically investigate the efficacy of defensive\nmechanisms. The results indicate alarming risks in protecting a VRS through the\nDNN robustification. Our experiments shed light on the importance of visual\nfeatures in very effective attack scenarios. Given the financial impact of VRSs\non many companies, we believe this work might rise the need to investigate how\nto successfully protect visual-based recommenders. Source code and data are\navailable at\nhttps://anonymous.4open.science/r/868f87ca-c8a4-41ba-9af9-20c41de33029/.\n