Biological data is the key to discovering treatments for incurable diseases. But finding a cure hidden in a mountain of data can be like looking for a needle in a haystack. Unless you have a supercomputer.
Enter Silicon Valley-based Good AI lab, which aims to crowdsource the world’s unused FLOPS in order to provide medical researchers with the computing power they need.
At September’s TechCrunch Disrupt 2017, Good AI lab Founder and CEO Mohsen Hejrati announced ClusterOne, a system designed to aggregate computational resources from computers anywhere in the world. By 2024, ClusterOne hopes to amass one hundred billion cores, which by today’s measure would make it the world’s largest AI supercomputer.
Founded in 2016, Good AI lab’s mission is to build an inclusive ecosystem and infrastructure that will make AI easier and more accessible.
The path to life sciences entrepreneur
Hejrati graduated from Sharif University and took his PhD from the University of California at Irvine. After a five-month stint as a robotics software engineer at Alphabet’s self-driving car company Waymo, he decided to go solo: “I was captivated by the revolution happening in healthcare. With access to such a library of genomic data, we can literally inspect the blueprint of human life.”
Machine learning is quickly and dramatically changing how medical research is done. Doctors must explore countless hypotheses while searching through the equivalent of more than 300 Gigabytes of human genomic data. Even a large team of top researchers are incapable of processing data at this scale. With an advantage in data and speed of many orders of magnitude, machine learning systems are simply more effective at this task.

Says Hejrati, “The future of life science research rests on our ability to analyze, store, and retrieve the sequencing work of others. If we’re going to keep up with the current pace of innovation, we need a heavy dose of computational infrastructure to help us do so.”
But building such a computational infrastructure is not an easy task. Training a machine learning model involves algorithms, datasets, GPU clusters, networking, version control and output management.
Good AI’s TensorPort solution
This spring, Good AI lab released TensorPort, a distributed machine learning platform for the TensorFlow framework, which can optimize the training of machine learning models.

TensorPort needs only a machine learning model, datasets and a description of how to train the model Then it does the rest. “Our goal is to take care of logistical and technical challenges so that you can focus on your models and your data. TensorPort today is used by leading biotech companies to push medical research and development forward,” says Hejrati.
Along with Tensorflow, most other frameworks will soon be available on TensorPort.
ClusterOne lowers costs
TensorPort can make machine learning training easier, but not necessarily cheaper. Hejrati says a research team will typically spend US$10 million running machine learning models on a public cloud platform such as Amazon Web Services. “AI is exclusive because algorithms and science are still limited, knowledge and expertise is rare, and infrastructure is hard to build and costly to operate.”
ClusterOne was created to bring such costs down. Users install an app that allows them to donate their surplus computing power to academia. ClusterOne runs as an encrypted virtual machine without access to other information on computers.
This is not the first time humans have aggregated computing power to solve an unfathomable issue. Crowdsourced computational power has already supported SETI’s search for extraterrestrial life and Folding@home’s disease research.
What makes ClusterOne different is that its shared computer power is dedicated exclusively to AI-related research, and that anyone who donates to ClusterOne can also use it for AI research.
With TensorPort as a platform and ClusterOne as a big, cheap AI processor, Good AI lab is advancing its agenda: an ecosystem to support the computer science community that allows scientists and engineers to build apps, provide resources and data, and contribute their expertise.
Hejrati believes he can help the traditionally conservative health and life sciences industry better understand and exploit AI’s tremendous potential in the field: “The driving force behind this revolutionary moment in healthcare certainly doesn’t come from the industry itself. Good AI lab is not a biotech company, but we are empowering them.”
Journalist: Tony Peng | Editor: Michael Sarazen
Pingback: AI 2017 part 2 highlights/ trends | Turing Machine