NNAISENSE, a global leader in industrial AI, has announced the launch of EvoTorch, the first open-source platform of its kind, providing industry with an evolutionary algorithm (EA) package which when combined with machine learning expertise, can solve complex operational challenges in a fraction of the time, at lower costs and greater scale.
With machine learning playing an increasingly vital role in many industrial verticals, evolutionary algorithms are an attractive solution to cascading challenges that accompany the increased complexity and size of automated processes. Evolutionary algorithms do not require a differentiable cost function, and are much more amenable to massive parallelization on modern hardware, compared to standard gradient-based alternatives. This means that a wider variety of problems, from learning robot controllers to optimizing schedules or product designs, can be tackled with greater efficiency. What has been missing is a software tool-set that makes it simple to experiment with EAs at any scale without worrying about the underlying details. Building on the popular PyTorch and Ray packages, EvoTorch provides researchers, developers and industrial manufacturers with powerful EAs that can be parallelized across CPUs or GPUs with no extra effort, speeding up optimization and minimizing costs.
Also Read: Normalyze Announces General Availability of Its Freemium Offering To Help DevSecOps
“In a pioneering project with Audi – where we worked on their autonomous parking system – we highlighted how much time could be saved by parallelizing evolutionary algorithms to achieve operational goals for industry. Using this technique, we condensed 180 years worth of simulated driving time into 24 hours,” said Jan Koutnik, Co-founder & Director of Intelligent Automation at NNAISENSE.
EAs function according to the principles of natural selection, or ‘survival of the fittest’. These algorithms start with a population of random solutions that are evaluated for fitness (propensity to solve the problem). At each iteration, the weaker solutions ‘die-out’ and the fittest 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 goal, or the trade-off between multiple conflicting goals.
“The goal of EvoTorch is to create an open-source ML community and facilitate the adoption of EAs within the field of automation by providing researchers and engineers with the tools to scale up their designs easily and quickly,” said Rupesh Srivastava, Director of Software Infrastructure at NNAISENSE. “The additional strength of this platform lies in the expert support we offer to commercial users, assisting them with effective customization, implementation and oversight. We will continuously expand our features to help users based on the data we acquire from ongoing projects, leading to a wider range of field-tested building blocks to spur on industry growth.”