The AI/HPC Journal Club will hold a seminar on Friday, Sept. 20, 2019, at 10 a.m. in Building 240, Room 1416. Join us for a presentation by Angel Yanguas-Gil (AMD).
“The Insect Brain as a Model System for Edge Processing and On-chip AI Applications”
Abstract
Insects are adaptable systems that carry out multisensory integrations to perform goal-oriented tasks, and are capable of adapting to changes in their environment and process information and learn in a context and task dependent manner. Moreover they do so with limited resources and extremely low power consumption. This makes them a fantastic model system for brain-inspired computing approaches that are tailored to operate at the edge. In particular, the combination of low power, adaptability, on-line learning, and robustness makes them particularly appealing for applications in remote or harsh environments. In this talk, I will focus on how we can leverage existing knowledge in insect neuroscience to develop algorithms and hardware that are capable of context-dependent learning and processing. Starting from algorithms, we have explored three different implementations: a spiking model running in Intel’s neuromorphic chip Loihi, custom FPGA implementation, and finally architectures based on emerging materials that build on traditional memristors and cross-bar arrays.