AI Company Develops Platform to Advance Machine Learning

The rapid rise of machine learning and artificial intelligence has resulted in a mass of complex computational and operational challenges that some engineers are trying to tackle with evolutionary algorithms, which find approximate solutions to problems that cannot be solved using other techniques. To help meet this challenge, industrial Al company NNAISENSE has developed an open-source platform based on evolutionary algorithms.

The platform, called EvoTorch, provides a software tool set that enables developers to experiment with evolutionary algorithms at any scale, without worrying about underlying details. The platform, built on the popular PyTorch and Ray packages, can create evolutionary algorithms that can be parallelized across CPUs or GPUs with little additional effort.

“We started creating the algorithms about four years ago,” said Dr. Timothy Atkinson, Research Scientist at NNAISENSE, in an interview with Design News. “The idea came to use these algorithms for open-source industrial projects.”

Atkinson added that the company utilized the algorithms on a project working on an autonomous parking system with European carmaker Audi. The algorithms enabled Audi to save extensive time and effort in developing simulation algorithms.

“We were researchers,” added Dr. Jonathan Masci, Co-Founder and Chief Scientist for Deep Learning at NNAISENSE. “Developing this platform became a way to establish ourselves.

The key to the open-source platform is evolutionary algorithms, which according to NNAISENSE function according to the principles of natural selection. These algorithms start with a population of random solutions that are evaluated for fitness (propensity to solve the problem). At each iteration, the fittest or most appropriate solutions reproduce resulting in an increasingly fit population that collectively adapts to solve the problem. From the final population of solutions, it is possible to select the one that best achieves the desired trade-off between multiple conflicting goals.

With machine learning becoming more important, evolutionary algorithms become an attractive solution to cascading challenges that accompany the increased complexity and size of automated processes. According to NNAISENSE’s Atkinson, evolutionary algorithms thrive on scale and are much more amenable to massive parallelization on modern hardware. This enables various industrial problems to be tackled with greater efficiency.

The software tool-set developed by NNAISENSE contains a library of algorithms to define problems the developer is trying to solve, along with APIs and a collection of interfaces. EvoTorch builds on the user-friendly principles of PyTorch and provides easy integration to well-known monitoring libraries which makes it easy to incorporate into existing workflows. Given its ever-expanding range of evolutionary algorithms and its intuitive interface, EvoTorch can also greatly simplify the job of academics and university students developing new algorithms.

The company is backing the platform with an open-source machine learning community that gives developers the tools to scale up their designs quickly and easily. NNAISENSE plans to expand the algorithm feature set to meet developer needs for various applications.

 Spencer Chin is a Senior Editor for Design News covering the electronics beat. He has many years of experience covering developments in components, semiconductors, subsystems, power, and other facets of electronics from both a business/supply-chain and technology perspective. He can be reached at [email protected]