The 9th Workshop on Approximate Computing

The Approximate Computing (AxC) design paradigm emerged as a promising solution to effectively enhance performance of computing systems. It cleverly leverages the intrinsic error resilience of applications to inaccuracy in their inner calculations to achieve a required trade-off between efficiency, performance, power demand, and acceptable error of returned results. Indeed, approximate results are hardly distinguishable from exact results for a vast plethora of applications, including audio, image, and video processing, data mining, information retrieval, machine learning, artificial intelligence, and even safety-critical applications.
     
The AxC’24 workshop explores the opportunity to exploit the AxC in recent application domains in an effective, dependable, and secure manner. Moreover, new techniques are moving towards considering systems as a whole, spanning from hardware to software components, broadening the potential community of interested researchers.

This year’s event will be in conjunction with the 2024 ACM/IEEE International Conference on Computer-Aided Design (ICCAD). To learn more, please visit the ICCAD Website.

 Workshop Program

9th Workshop on Approximate Computing - AxC 2024 - Program

[9:00 am - 10 am] - Keynote talk: Evolutionary approximation: From components to hardware accelerators. Prof. Vojtech Mrazek, Brno University of Technology, Czech Republic

[10:00 am - 10:15 am] - Break

[10:15 am - 11:30 am] - Session 1: Precision Meets Pragmatism: Approximating for Efficiency in AI

a) FF-INT8: INT8 Forward-Forward Algorithm for Efficient DNN Training on Resource-Constrained Devices. Jingxiao Ma, Priyadarshini Panda and Sherief Reda

b) Approximate Multipliers and Information Bottleneck Theory: A New Approach to DNN Analysis and Optimization. Salar Shakibhamedan, Nima Amirafshar, Nima Taherinejad and Axel Jantsch

c) Modular redundancy based approximation of tree ensemble classifiers. Antonio Emmanuele, Mario Barbareschi, Salvatore Pappalardo and Alberto Bosio

[11:30 am - 11:45 am] - Break

[11:45 am - 1 pm] - Session 2: Optimized Circuits and Computation: Reliability, Health, and Energy Efficiency

a) Bio-FDApx: Biometric-Feature-Driven Approximation for Real-time Health Monitoring Systems. Nishanth Chennagouni, Wei Lu, Eric Brown and Qiaoyan Yu

b) RelaX: Reliability Assessment in Approximate Circuits. Somayeh Sadeghi-Kohan, Muhammad Awais, Qazi Arbab Ahmed, Marco Platzner, Sybille Hellebrand, Thorsten Jungeblut and Hans-Joachim Wunderlich

c) AVFS Control For Preconditioned Conjugate Gradient Computations With Optimal Energy-Delay-Product. Stefan Holst, Hanieh Jafarzadeh and Hans-Joachim Wunderlich

 Invited Speakers and Talks

Evolutionary approximation: From components to hardware accelerators

Vojtech Mrazek

The aim of approximation algorithms is to introduce error into the calculation in such a way that its effect on accuracy is minimal, but its effect on energy savings is as high as possible. Identifying suitable modifications can be done analytically, but it turns out that space search algorithms can correctly find them. In this keynote, algorithms based on the principles of evolution and genetic programming will be presented. These algorithms have been successfully used for both the design of basic components and for approximating relatively large computational blocks. Ways to take advantage of bioinspired algorithms and how basic scalability issues of evaluation and representation can be addressed will be presented.

Supporters and Partners