Discovery Challenges

Discovery Challenges will take place at the Minatec site. More information on how to get to Minatec from the WTC are provided here.

Title: Lung Cancer Survival Prediction Challenge
Time: 11:00 – 13:00

Every year 1.9 million people are diagnosed with Non-Small Cell Lung Cancer (NSCLC). Only 25% of those will survive 5 years beyond their diagnosis, with prognosis depending on many factors including demographics, clinical characteristics, and genetic alterations, among others.
Survival Machine Learning models could enable us to better predict prognosis for individual patients, which in turn has real world clinical applications for improving treatment and our understanding of NSCLC. Additionally, representations learned by fitting Survival Machine Learning models on NSCLC data could be used to stratify patients and obtain clinical clusters or phenotypes that give us insights on how to better categorize NSCLCs.
In this challenge you will predict the risk of overall death using clinical EHR data from around 75,000 advanced NSCLC patients provided by Flatiron Health. The features consist of patient characteristics such as demographic information, vital sign data, and biomarkers.

Lee Cooper, Northwestern University, USA
Naghmeh Ghazaleh, Roche, Switzerland
Jonas Richiardi, Lausanne University Hospital and University of Lausanne, Switzerland
Damian Roqueiro,  Roche, Switzerland
Diego Saldana, Novartis, Switzerland
James Black, Roche, Switzerland
Selen Bozkurt, Flatiron Health, USA

Challenge Website:
Title: PRINCE Out-of-distribution Generalization Challenge
Time: 2:30-6:30pm

The goal is to learn a mapping from a set of features (X) to a label (y) that generalizes well across environments that have not been seen in training (a.k.a. out-of-distribution generalization). This challenge uses data from Criteo, an advertising company and environments correspond to different business types (e.g. Retail, Travel, Insurance, Classifieds etc). Such environments exhibit challenging  distribution shifts whilst still depending on the same underlying causal mechanism. Will you be able to infer shoppers behavior on retail merchants while seeing only travel and insurance data ?

– Eustache Diemert, Criteo AI Lab
– Ugo Tanielian, Criteo AI Lab
– Thibaud Rahier, Criteo AI Lab
– Matthieu Kirchmeyer, Criteo AI Lab
– Alain Rakotomamonjy, Criteo AI Lab
РAexandre Ram̩, Sorbonne Universit̩
Challenge Website:
Title: Expedia Group Cross-brand Lodging Recommendation

In online platforms it is often the case to have multiple brands under the same group which may target different customer profiles. For example, in the hospitality domain Expedia Group has multiple brands like Brand Expedia,, Orbitz or Vrbo. In this context, being able to provide cross-brand recommendations to travelers is an important task as it can improve traveler experience across different point of sales. In this challenge we propose a cross-brand recommendation task where the participants will be provided with traveler actions in a source brand (e.g. property clicks) and asked  to predict actions in target brands. The objective is to improve the recommendations in target brands using the data from a source brand.

Adam Woznica, Expedia Group
Ioannis Partalas, Expedia Group
– Jan Krasnodebski, Expedia Group

Challenge Website: