The Larger The Fairer? Small Neural Networks Can Achieve Fairness for Edge Devices
TimeTuesday, July 12th2:37pm - 3pm PDT
Location3005, Level 3
Event Type
Research Manuscript
AI/ML Design: System and Platform
DescriptionAs the neural networks have been increasingly deployed in edge/mobile devices, the fairness concerns gradually emerge for many edge applications, such as medical AI, economic AI, and language processing. One fundamental question is what is the fairest model? By scanning the existing neural networks, we observe that the larger model commonly has better fairness. One additional question then arises, whether we can find small neural networks that can fit edge devices while achieving high fairness? With these questions in mind, we apply the neural architecture search to explore the neural architectures to make better tradeoffs among fairness, latency, and accuracy.