crowdAI: the open, data-science challenge platform
A still from the simulation submitted by one of the competitors of “Learning to Run” (credit: JGeek/crowdAI)
crowdAI, an EPFL-developed open data science challenge platform that hosts machine learning competitions, has been awarded over $100,000 from Amazon and Nvidia for its latest challenge, “Learning to Run”.
The crowdAI project is an open platform for data-science challenges. It is developed at EPFL, and founded by Marcel Salathé, who is known for his innovative endeavors in the field of applied machine learning, from diagnosing crop diseases to estimating macro-nutrient composition through food images.
The idea behind crowdAI is to provide an open source platform for hosting open data science challenges, and inviting the community of data scientists worldwide to develop machine learning algorithms for specified data science problems. To quote the website: “crowdAI connects data science experts and enthusiasts with open data to solve specific problems, through challenges.”
The challenges span a broad spectrum, from applying deep learning to diagnose plant diseases based on images of symptomatic leaves (PlantVillage), to predicting a person’s height out of a person’s genotype, to teaching a virtual, anatomically accurate skeleton how to walk and run.
The last two challenges have been spearheaded by Stanford’s Neuromuscular Biomechanics Lab, and their ultimate aim goes beyond mere computer simulation: The aim of the challenge is to find better ways of helping children with cerebral palsy respond to muscle-relaxing surgery. This is a type of intervention that doctors often resort to as a means of improving the patient’s gait. However, it doesn’t always work.
“The key question is how to predict how patients will walk after surgery,” says Łukasz Kidziński, the Stanford bioengineering postdoc who came up with the crowdAI challenge. “That’s a big question, which is extremely difficult to approach.”
The “Learning to Walk” challenge, which has now ended, gathered a lot of interest during its lifespan, with 199 participants and 200 different submissions. But the “Learning to Run” version, still ongoing, has fared even better, with almost 1400 submissions from 391 contestants. The contest is actually part of the 2017 Neural Information Processing Systems (NIPS) conference, one of the world’s biggest conventions on artificial intelligence (the conference runs 4-9 December and is already sold out).
But that’s not all: “Learning to Run” also attracted the interest of industry giants who donated two prizes for the winners: Amazon offered $30,000 worth of AWS Cloud credits, while Nvidia is donating a DGX Station™, the world’s first personal GPU supercomputer worth $70,000, among other prizes.
Of course, this will not be the last challenge of crowdAI: “We’re in advanced talks with the Blue Brain project, two NGOs, an information retrieval conference, and various EPFL labs to run more open data science challenges,” says Marcel Salathé. “I am very excited about the potential of this platform to speed up the transition of research towards full open science, and I am very proud of what we’ve achieved so far. Our next step is to convince other universities to contribute to the platform and turn this into a multi-institutional effort.”