NIPS 2017 Competition Track

This is the first NIPS edition on "NIPS Competitions". We received 23 competition proposals related to data-driven and live competitions on different aspects of NIPS. Proposals were reviewed by several high qualified researchers and experts in challenges organization. Five top-scored competitions were accepted to be run and present their results during the NIPS 2017 Competition track day. Evaluation was based on the quality of data, problem interest and impact, promoting the design of new models, and a proper schedule and managing procedure. Below, you can find the five accepted competitions. Please visit each competition webpage to read more about the competition, its schedule, and how to participate. Each competition has its own schedule defined by its organizers. The results of the competitions, including organizers and top ranked participants talks will be presented during the Competition track day at NIPS 2017. Organizers and participants will be invited to submit their contribution as a book chapter to the upcoming NIPS 2017 Competition book, within Springer Series in Challenges in Machine Learning.

The Conversational Intelligence Challenge

Competition summary
Dialogue systems and conversational agents – including chatbots, personal assistants and voice control interfaces – are becoming increasingly widespread in our daily lives. In addition to the growing real-world applications, the ability to converse is also closely related to the overall goal of AI. Recent advances in machine learning have sparked a renewed interest for dialogue systems in the research community. This NIPS Live Competition aims to unify the community around the challenging task: building systems capable of intelligent conversations. Teams are expected to submit dialogue systems able to carry out intelligent and natural conversations of news articles with humans. At the final stage of the competition participants, as well as volunteers, will be randomly matched with a bot or a human to chat and evaluate answers of a peer. We expect the competition to have two major outcomes: (1) a measure of quality of state-of-the-art dialogue systems, and (2) an open-source dataset collected from evaluated dialogues.


Mikhail Burtsev, Valentin Malykh, MIPT, Moscow
Ryan Lowe, McGill University, Montreal
Iulian Serban, Yoshua Bengio, University of Montreal, Montreal
Alexander Rudnicky, Alan W. Black, Carnegie Mellon University, Pittsburgh
Contact email:


Classifying Clinically Actionable Genetic Mutations

Competition summary
While the role of genetic testing in advancing our understanding of cancer and designing more precise and effective treatments holds much promise, progress has been slow due to significant amount of manual work still required to understand genomics. For the past several years, world-class researchers at Memorial Sloan Kettering Cancer Center have worked to create an expert-annotated precision oncology knowledge base, OncoKB. It contains several thousand annotations of which genes are clinically actionable and which are not based on clinical literature. This dataset can be used to train machine learning models to help experts significantly speed up their research. 
This competition is a challenge to develop classification models which analyze abstracts of medical articles and, based on their content accurately determine oncogenicity (4 classes) and mutation effect (9 classes) of the genes discussed in them. Participants will not only have an opportunity to work with real-world data and get to answer one of the key open questions in cancer genetics and precision medicine, but the winning model will be tested and deployed at Memorial Sloan Kettering and will have the potential to touch more than 120,000 patients it sees every year, and many more around the world.


Iker Huerga,
Alexander Grigorenko,
Anasuya Das,
Leifur Thorbergsson,
Competition Coordinators
Kyla Nemitx,
Randi Kaplan,
Jenna Sandker, 

 Webpage:  coming soon

Learning to Run

Competition summary
In this competition, you are tasked with developing a controller to enable a physiologically-based human model to navigate a complex obstacle course as quickly as possible. You are provided with a human musculoskeletal model and a physics-based simulation environment where you can synthesize physically and physiologically accurate motion. Potential obstacles include external obstacles like steps, or a slippery floor, along with internal obstacles like muscle weakness or motor noise. You are scored based on the distance you travel through the obstacle course in a set amount of time.


Lead organizer: Łukasz Kidziński <>
Coordinators: Carmichael Ong, Mohanty Sharada, Jason Fries, Jennifer Hicks
Promotion: Joy Ku
Platform administrator: Sean Carroll
Advisors: Sergey Levine, Marcel Salathé, Scott Delp


Human-Computer Question Answering Competition

Competition summary
Question answering is a core problem in natural language processing: given a question, provide the entity that it is asking about.  When top humans compete in this task, they answer questions incrementally; i.e., players can interrupt the questions to show they know the subject better than their slower competitors.  This formalism is called “quiz bowl“ and was the subject of the NIPS 2015 best demonstration.
This year, competitors can submit their own system to compete in a quiz bowl competition between computers and humans.  Entrants create systems that receive questions one word at a time and decide when to answer.  This then provides a framework for the system to compete against a top human team of quiz bowl players in a final game that will be part of NIPS 2017.


Jordan Boyd-Graber (University of Colorado)
Hal Daume III (University of Maryland)
He He (Stanford)
Mohit Iyyer (University of Maryland)
Pedro Rodriguez (University of Colorado)


Adversarial Attacks and Defences

Competition summary
Most existing machine learning classifiers are highly vulnerable to adversarial examples. An adversarial example is a sample of input data which has been modified very slightly in a way that is intended to cause a machine learning classifier to misclassify it. In many cases, these modifications can be so subtle that a human observer does not even notice the modification at all, yet the classifier still makes a mistake. Adversarial examples pose security concerns because they could be used to perform an attack on machine learning systems, even if the adversary has no access to the underlying model.

To accelerate research on adversarial examples and robustness of machine learning classifiers we organize a challenge that encourages researchers to develop new methods to generate adversarial examples as well as to develop new ways to defend against them. As a part of the challenge participants are invited to develop methods to craft adversarial examples as well as models which are robust to adversarial examples.


Alexey Kurakin,
Ian Goodfellow,
Samy Bengio,

Primary contact e-mail which will be provided to participants: 

Webpage: coming soon