Attacking All Tasks at Once Using Adversarial Examples in Multi-Task Learning
This work addresses the understudied vulnerability of multi-task models to adversarial attacks, which is crucial for improving the security of visual content understanding systems.
The paper tackles the problem of adversarial robustness in multi-task learning models by proposing a novel attack framework, Dynamic Gradient Balancing Attack (DGBA), which effectively attacks all tasks simultaneously, revealing a trade-off between task accuracy and robustness due to parameter sharing.
Visual content understanding frequently relies on multi-task models to extract robust representations of a single visual input for multiple downstream tasks. However, in comparison to extensively studied single-task models, the adversarial robustness of multi-task models has received significantly less attention and many questions remain unclear: 1) How robust are multi-task models to single task adversarial attacks, 2) Can adversarial attacks be designed to simultaneously attack all tasks in a multi-task model, and 3) How does parameter sharing across tasks affect multi-task model robustness to adversarial attacks? This paper aims to answer these questions through careful analysis and rigorous experimentation. First, we analyze the inherent drawbacks of two commonly-used adaptations of single-task white-box attacks in attacking multi-task models. We then propose a novel attack framework, Dynamic Gradient Balancing Attack (DGBA). Our framework poses the problem of attacking all tasks in a multi-task model as an optimization problem that can be efficiently solved through integer linear programming. Extensive evaluation on two popular MTL benchmarks, NYUv2 and Tiny-Taxonomy, demonstrates the effectiveness of DGBA compared to baselines in attacking both clean and adversarially trained multi-task models. Our results also reveal a fundamental trade-off between improving task accuracy via parameter sharing across tasks and undermining model robustness due to increased attack transferability from parameter sharing.