Explore the Akida event-based neural processor in this 38-minute tinyML Talks webcast featuring Kristofor Carlson from BrainChip Inc. Dive into the key distinguishing factors of Akida's computing architecture, including aggressive 1 to 4-bit weight and activation quantization, event-based implementation of machine-learning operations, and distributed computation across multiple neural processing units. Learn how these architectural innovations lead to significant reductions in MACs, parameter memory usage, and peak bandwidth requirements compared to traditional 8-bit machine learning accelerators. Discover Akida's on-chip learning capabilities using a proprietary bio-inspired algorithm, and examine its performance in few-shot learning for both visual and auditory applications. Gain insights into the chip's design and potential applications in edge computing through demonstrations and a comprehensive overview.
The Akida Neural Processor - Low Power CNN Inference and Learning
Overview
Syllabus
Introduction
Sponsors
Upcoming
Speaker Introduction
Speaker Background
Akida Neural Processor
Eventbased computation
Low bit precision
Edge learning
Edge learning demo
Akida chip overview
Summary
Questions
Conclusion
Taught by
tinyML
