Multi-Task Inference with Context Aware Network Architectures

Abstract: As computer vision and robotics permeate industrial applications, the need for a reliable and thorough visual understanding of a scene has been solved with numerous CNNs. Several of these networks have complementary or overlapping objectives that cause them to perform redundant computation when executed independently. In an autonomous vehicle or drone, power and compute resources are limited so these inefficiencies present an opportunity for improvement. In this talk I will discuss our work in context aware network architectures that can be dynamically reconfigured to fit changing needs or priorities based on environmental context.

Bio: Skyler Anderson is a computer science and engineering masters student at The Pennsylvania State University advised by Professor Vijay Narayanan. He completed his Bachelor of Science in electrical engineering at The Pennsylvania State University. His research is in computer vision and deep learning with a focus on power efficient network architectures.