TinyML Talks - Software-Hardware Co-design for Tiny AI Systems

Explore a comprehensive tinyML talk on software and hardware co-design for tiny AI systems. Delve into efficient AI models through hardware-friendly model compression and topology-aware Neural Architecture Search, optimizing quality-efficiency trade-offs. Learn about cross-optimization design and efficient distributed learning for swift and scalable AI systems with specialized hardware. Discover enhancements in quality-efficiency trade-offs for alternative applications like Electronic Design Automation (EDA) and Adversarial Machine Learning. Gain insights into the future of full-stack tiny AI solutions, covering topics such as intended machines, integration, computation, accuracy engineering, neural networks, distributed learning, privacy, and edge computing. Join Yiran Chen, Chair of ACM SIGDA, as he presents a vision for the future of tiny AI systems in this hour-long exploration of cutting-edge technologies and methodologies.

Intro
The Age of the Intended Machine
Basic Solutions
Integration
Computation
Common Practice
LevelBased Solution
Accuracy Engineering
Vertical Integration
Chips
Architecture
Natural Quality Stream
Tensor Partitions
Dynamic Programming
Advanced Neural Network
Graph Computing
Zero Computation
Zero Pruning
Israelites
Distributed Learning
Distributed Mobile Training
Clustering
Lottery ticket hypothesis
Deep learning
Delivery network
Privacy
Neural Network Research
Neural Active Search
Topology Awareness
Predictor
DAG
Neural Network Design
Summary
Nonconventional powers
Questions
Edge Impulse