Open-Source AI Training Platform Aims To Break Deep Learning 'Bottleneck'

Citing lack of software infrastructure as "a fundamental bottleneck in achieving artificial intelligence's (AI)immense potential," Determined AI announced last week that it is open sourcing its Deep Learning Training Platform under the Apache 2.0 license.

Tech giants like Google and Facebook have the resources to build proprietary, AI-native internal infrastructure, the founders of Determined AI argued in a blog post announcing the open source move. "For everyone else who doesn't have access to this infrastructure, building practical applications powered by AI remains prohibitively expensive, time-consuming, and difficult," the blog states.

"At Determined, we have been laser-focused on removing these formidable infrastructure barriers to AI success," wrote Evan Sparks, Neil Conway, and Ameet Talwalkar.

The trio met while studying computer science at UC Berkeley and founded their San Francisco-based company in 2017, according to an article on

"We've spent the last three years working closely with cutting-edge teams applying deep learning to a variety of industries," they wrote in the blog. "These early customers spoke with a clear voice: without better infrastructure, training deep learning models at scale remains extremely difficult, as organizations move from research to production."

The list of features in the platform includes:

  • High-performance distributed training, which Determined AI claims is twice as fast as the stock Horovod distributed deep learning training framework configuration.
  • Hyperparameter search that integrates tightly with the job scheduler and is parallel by default, "helping users achieve more accurate models 100x faster than standard search methods and 10x faster than Bayesian Optimization methods."
  • Deep learning tools for individuals and teams, helping users with experiment tracking, log management, metrics visualization, reproducibility, and dependency management

The platform is already being used by model developers in a pharmaceutical drug discovery, industrial IoT, autonomous vehicles, and advertising technology (AdTech), according to the company blog.

"By adopting Determined AI's software platform, our team of deep learning engineers has been able to rapidly deliver new, advanced, Industrial IoT products to our customers," said Ben Chehebar, chief product officer at San Francisco-based Compology, in a statement. "We're delivering new AI features 10 times faster than before."

Ben Mabey, Interim CTO of Salt Lake City-based Recursion Pharmaceuticals stated: "With AI at the forefront of Recursion's vision for biopharmaceuticals, we use Determined to manage hundreds of on-premises GPUs, as well as dynamically scale to using GPUs on Google Cloud Platform. Using Determined's native support for distributed training, we were able to reduce the training time for a key computer vision model from three days to three hours, without changing our model code."

The Determined AI platform is currently available on GitHub.