[11/08] Hiddenite: 4K-PE Hidden Network Inference 4D-Tensor Engine Exploiting On-Chip Model Construction Achieving 34.8-to-16.0TOPS/W for CIFAR-100 and ImageNet

[jack-kjlee]

• When : 11/08
• Who : Kyeong-Jun Lee (이경준) (POSTECH, [leekj9444 (at) postech (dot) ac (dot) kr])
• What
• Title: Hiddenite: 4K-PE Hidden Network Inference 4D-Tensor Engine Exploiting On-Chip Model Construction Achieving 34.8-to-16.0TOPS/W for CIFAR-100 and ImageNet
• Links: https://ieeexplore.ieee.org/document/9731668
• Keywords: Pruning, On-chip weight generation
• Preliminaries:
• The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks
• What’s Hidden in a Randomly Weighted Neural Network?
• 발표자료