Approximate Computing

Approximate computing (AxC) is a paradigm shift that trades off computational accuracy for improved energy efficiency and performance, particularly beneficial in resource-constrained environments like embedded systems and mobile devices. Current research focuses on applying AxC to various machine learning models, including convolutional neural networks (CNNs), vision transformers (ViTs), and spiking neural networks (SNNs), often incorporating techniques like probabilistic approximation and low-precision arithmetic within hardware architectures. This approach holds significant promise for enabling more efficient deployment of AI and other computationally intensive applications in power-limited settings, impacting both the design of energy-efficient hardware and the development of more sustainable computing systems.

Papers

June 21, 2022

Enabling Capsule Networks at the Edge through Approximate Softmax and Squash Operations
Alberto Marchisio, Beatrice Bussolino, Edoardo Salvati, Maurizio Martina, Guido Masera, Muhammad Shafique
Extreme Edge Softmax Function Capsule Network Computational Resource Complex Deep Approximate Computing

March 16, 2022

Hardware Approximate Techniques for Deep Neural Network Accelerators: A Survey
Giorgos Armeniakos, Georgios Zervakis, Dimitrios Soudris, Jörg Henkel
Deep Neural Network Timely Survey DNN Accelerator Approximate Computing Deep Learning Accelerator Hardware Approximation Approximate DNN Accelerator

December 2, 2021

Is Approximation Universally Defensive Against Adversarial Attacks in Deep Neural Networks?
Ayesha Siddique, Khaza Anuarul Hoque
Deep Neural Network Adversarial Attack Adversarial Robustness Universal Approximation Approximate Computing Approximate Multiplier Approximate DNN Accelerator

Approximate Computing

Papers

Enabling Capsule Networks at the Edge through Approximate Softmax and Squash Operations

Hardware Approximate Techniques for Deep Neural Network Accelerators: A Survey

Is Approximation Universally Defensive Against Adversarial Attacks in Deep Neural Networks?