Deep neural networks for image classifi cation are well-known to be vulnerable to adversarial attacks. One such attack that has garnered recent attention is the adversarial backdoor attack, which has demonstrated the capability to perform targeted misclassifi cation of specifi c examples. In particular, backdoor attacks attempt to force a model to learn spurious relations between backdoor trigger patterns and false labels. In response to this threat, numerous defensive measures have been proposed; however, defenses against backdoor attacks focus on backdoor pattern detection, which may be unreliable against novel or unexpected types of backdoor pattern designs. We introduce a novel re-contextualization of the adversarial setting, where the presence of an adversary implicitly admits the existence of multiple database contributors. Then, under the mild assumption of contributor awareness, it becomes possible to exploit this knowledge to defend against backdoor attacks by destroying the false label associations. We propose a contributor-aware universal defensive framework for learning in the presence of multiple, potentially adversarial data sources that utilizes semi-supervised ensembles and learning from crowds to fi lter the false labels produced by adversarial triggers. Importantly, this defensive strategy is agnostic to backdoor pattern design, as it functions without needing—or even attempting—to perform either adversary identifi cation or backdoor pattern detection during either training or inference. Our empirical studies demonstrate the robustness of the proposed framework against adversarial backdoor attacks from multiple simultaneous adversaries.

Most studies on learning from noisy labels rely on unrealistic models of i.i.d. label noise, such as class-conditional transition matrices. More recent work on instance-dependent noise models are more realistic; however, these models assume a single generative process for label noise across the entire dataset, which does not accurately capture the real-world labeling dynamics observed in modern dataset collection. We propose a new model of label noise, which we call “labeler-dependent noise (LDN).” LDN extends and generalizes instance-dependent noise to multiple labelers, based on the observation that modern datasets are typically constructed via distributed data gathering and labeling methods. Furthermore, inspired by recent research in social learning theory, we introduce a modular learning framework under our proposed labeler-dependent noise model called “labeler-aware learning”, in which each labeler is estimated and compared against the others during the learning process. Our experimental results demonstrate how previous state-of-the-art approaches for learning from noisy labels are unable to handle the general model of label noise, while our labeler-aware learning framework remains robust even when large fractions of labelers are spammers.