SWIS: Shared Weight Bit Sparsity for Efficient Neural Network Acceleration

Explore the SWIS – Shared Weight bIt Sparsity framework for efficient neural network acceleration in this 20-minute conference talk from the tinyML Research Symposium 2021. Delve into the quantization technique that improves performance and storage compression through offline weight decomposition and scheduling algorithms. Learn how SWIS achieves significant accuracy improvements when quantizing MobileNet-v2, and discover its potential for up to 6X speedup and 1.8X energy improvement over state-of-the-art bit-serial architectures. Follow the presentation as it covers the introduction, base sparsity, quantization error, base serial multiplier, SWIS architecture, computation animation, scheduling, retraining, and concludes with a Q&A session.