Balancing Privacy, Robustness, and Efficiency in Machine Learning
It addresses the fundamental challenge of balancing key system properties for real-world ML deployment, but is incremental as it critiques existing approaches without presenting new solutions.
This paper argues that achieving robustness, privacy, and efficiency simultaneously in machine learning is infeasible under current threat models, advocating for a research agenda to formalize this trilemma and explore trade-offs through relaxed assumptions.
This position paper argues that achieving robustness, privacy, and efficiency simultaneously in machine learning systems is infeasible under prevailing threat models. The tension between these goals arises not from algorithmic shortcomings but from structural limitations imposed by worst-case adversarial assumptions. We advocate for a systematic research agenda aimed at formalizing the robustness-privacy-efficiency trilemma, exploring how principled relaxations of threat models can unlock better trade-offs, and designing benchmarks that expose rather than obscure the compromises made. By shifting focus from aspirational universal guarantees to context-aware system design, the machine learning community can build models that are truly appropriate for real-world deployment.