Striving for world-class excellence in Data Science: Moscow Institute of Physics and Technology scoops top position at the Data Science Game 2016
Press release – Data Science Game Paris, 12 September 2016
The global event saw crème de la crème students around the world compete for the title of Data Science Champion.
The final stage of the 2nd edition of the Data Science Game, an international platform for prestigious institutes to gather and encourage their scholars to demonstrate their skillsets in the field of Data Science, was held on 10th and 11th of September at Capgemini’s Les Fontaines campus. The gathering saw eighty students compete against one another to face a coding challenge and strive to build the best algorithm to win the hackathon that included finding the best solutions to a predictive problem pertaining to Insurance, by using machine learning models.
A tribute to deep learning
To celebrate the finals, participants from around the globe were welcomed at the headquarters of Microsoft France, in Paris, during a gathering of the Data Science international community. The event culminated in an unprecedented roundtable devoted to engineering and deep learning. The roundtable was chaired by Isabelle Guyon, President of ChaLearn, and involved senior analytics figures from the three main partners of the Data Science Game: Capgemini Consulting, Microsoft, and AXA.
After the conference, students and partners enjoyed a special cocktail and networking session.
A fierce hackathon
This year once again, Data Science Game was a resounding success thanks to its partners’ support, who are all key contributors in the field of Data Science. For the final phase, the students were welcomed in the Château Les Fontaines, the training and development campus of the Capgemini Group. Microsoft provided its support by giving free access to its Azure computing clusters while the final challenge was set by AXA. A mammoth 30-hour hackathon witnessed the creation of some of the most innovative and refined solutions to the problem.
The challenge dataset contained requests for automobile insurance quotes received by AXA from different brokers and comparison websites. The participants were asked to predict whether the person who requested a given quote bought the associated insurance policy.
In the end, teams were ranked according to their prediction score. The performance measure used was log-loss, a measure that strongly penalizes predictions that are both confident (probabilities are close to 0 or 1) and wrong.
A finalist points out:
“We felt helpless after training a neural net for hours, and then noticing the importance of feature engineering in this challenge”
|Prize Capgemini, for 1st place||Russian Data Mafia||Moscow Institute of Physics and Technology||Russia|
|Prize AXA, for 2nd place||Cantab||Cambridge University||United Kingdom|
|Prize Microsoft, for 3rd place||We just want our name to be the longest one||Skoltech University||Russia|
|Prize Milliman, for 4th place||Jonquille||Université Pierre et Marie Curie||France|
|Prize Numberly, for 5th place||Brosio2BeWild||University of Padova||Italy|
|Prize QuantCube Technology, for Jury’s innovation prize||ml_noobs||Moscow State University||Russia|
Shortlisted countries included: France, Netherlands, Russia, Germany, The UK, Singapore, USA, Japan, India, and Italy.
Data Science Game qualification phase:
UPMC, Ecole Polytechnique and University of Amsterdam take the lead!
Press release – Data Science Game Paris, 16 July 2016
For the second year in a row, the Data Science Game has seen Data Science students from around the world compete for the title of Data Science Champion.
This year, 143 teams, representing more than 50 universities from 28 different countries (view the map) faced a demanding and innovative business challenge during the qualifiers.
When Computer Vision serves renewable energy production
This year, the qualification challenge of the Data Science Game has focused on concerns for the future, at the intersection between ecology and energy issues; the trial looking at optimizing the production of solar energy.
In order to map the solar energy production potential in France, the OpenSolarMap project provides satellite images of roofs of 80,000 buildings. Based on the individual contributions of users, the orientation of about 15,000 roofs has been categorised. Automatic classification of roof orientation is a true challenge for Etalab, the French public agency in charge of open data and use of data in the administration in France which provided the data.
For the contestants of the Data Science Game, the challenge was to develop an algorithm able to recognise the orientation of a roof from a satellite photograph by building on more than 10,000 photograph of roofs which have been categorized thanks to crowdsourcing.
Great success of Deep Learning
The majority of top 40 teams used Deep Learning methods. These machine learning techniques are known to be particularly efficient on Computer Vision issues and in the context of Big Data.
Thanks to these models, the top 20 teams scored very highly, with between 82 and 87% of good predictions. However good the ranking of the top 3 university contestants - , University Pierre and Marie Curie, Ecole Polytechnique and the University of Amsterdam - , the competion is far from over. For the top 20 universities, it will take far more work and energy to succeed in the final round in September.
After 24 days of competition and 831 algorithms submitted by the 143 teams, the 20 finalists are:
|1||University Pierre and Marie Curie||France|
|3||University of Amsterdam||Netherlands|
|4||Moscow Institute of Physics and Technology||Russia|
|5||Univeristy of Mannheim||Germany|
|6||Moscow State University||Russia|
|9||National University of Singapore||Singapore|
|12||University of Tsukuba||Japan|
|16||RWTH Aachen University||Germany|
|20||University of Padova||Italy|
|22||Indian Statistical Institute||India|
*two teams from Ecole Polytechnique and Columbia University actually reached the top 20 but are not among the finalists because, according to the rules, only one team can represent a university.
See you on 10 and 11 September, 2016!
On 10-11 September, the finalists are invited in Paris for the final stage of this international student hackathon. During the final, the 20 best teams will defend their university and country colors in a Big Data analysis challenge.
For the final phase, the students will be welcomed in the Château Les Fontaines, the training / development campus of of the Capgemini Group. During the competition, the contestants will be able to rely on the expertise and the knowledge of data scientists from Axa and its Data Innovation Lab, Capgemini, Microsoft and Milliman.
Will the Moscow State University, keep its title of Data Science Champion? See you during the final to find out!
Contact: Audrey Ribeiro
Data Science Game: a worldwide student challenge
Press release – Data Science Game Paris, 17 June 2016
For the second year in a row, the Data Science Game will see Data Science students from around the world compete in this prestigious competition. This year, 143 teams, representing more than 50 universities from 28 different countries (view the map) will face a real-life demanding and innovative business challenge. To stand out in this competition, students will have to imagine and implement predictive models related to Big Data issues.
On September 10 and 11, Paris will welcome an international student hackathon focused on Big Data analytics. From the 143 teams, twenty groups from around the world will defend their university’s reputation in the 2016 Data Science Game.
But to access this ultimate challenge, teams who have already enrolled first have to demonstrate their worth through an online application process from June 17 to July 10. During this first phase, co-organized with ChaLearn specialists in the organization of Machine Learning challenges, the applicants will have to find the best solutions to a predictive problem involving massive and complex data, using statistical algorithms to treat and figure out this data. Each university will be scored on the predictive power of these algorithms, and a ranking will single out the best twenty teams that will compete in the final challenge in Paris.
This year once again, Data Science Game can count on its partners’ support, who are key contributors in the field of Data Science. Thanks to Capgemini, a global leader in digital consulting and IT services, participating students will have the opportunity to stay in an exceptional historic place: « Les Fontaines » (the Capgemini Group Campus) near Paris.
John Brahim, Head of Capgemini Group’s Insights & Data Global Practice said “Data analytics are critical to our clients in this digital landscape, providing powerful insights that change the business. Competitions such as the 2016 Data Science Game will help inspire the next generation of data specialists and to provide them with the environment to experience first-hand the complexities of solving real business challenges. We hope that this will encourage them to go on to pursue a stimulating career in data analytics.”
Microsoft will provide the students its Azure cloud computing platform in order to offer them a particularly convenient environment for Big Data analysis.
Cortana Intelligence Suite is Microsoft’s fully managed big data and advanced analytics suite. With Cortana Intelligence, students can access a rich set of data science tools including Azure Machine Learning, Jupyter notebooks on R and Python as well as cognitive API’s to transform data into intelligent action. At the 2016 Data Science Game, it will be great to see how the data science community will use our platform in innovative ways to develop compelling solutions.
Herain Oberoi, Director, Product Marketing & Management, C&E, Microsoft Redmond
The teams will also benefit from the expertise of the Data Innovation Lab, created by the AXA Group in early 2014 with the objective to create value for its customers based on their data.
“At AXA, we are convinced that Big Data is a great opportunity to redefine the way we serve our customers. The Data Innovation Lab is committed to making AXA data-driven by fostering Data Science skills and expertise, building technological assets and supporting all our subsidiaries in their Big Data projects. Sponsoring the 2016 Data Science Game is aligned with this objective: it enables us to contribute to the growing data ecosystem, in particular by supporting the development of data talents, and is also a good way to get a new perspective on key insurance challenges with state-of-the-art technical solutions”.
Philippe Marie-Jeanne, AXA Group Chief Data Officer and Head of the Data Innovation Lab
These two days of competition in September represent a unique opportunity for the contestants to show their skills in the presence of data specialists. Nurtured throughout the weekend, with advice from Data Scientists working in the partner firms, the students will be in the best possible environment to learn and be excited by this competitive environment.
One question remains: who will succeed the Russian team of Moscow State University (MSU), the winner of the 2015 competition?
Contact: Audrey Ribeiro