Projects - ALVIN YAU

GitHub: https://github.com/alvinsgithub

New! Hackster: http://www.hackster.io/alvin/

Twitter Lenses – Nishio Laboratory, Osaka University, Japan

Slides

Google Docs: Link

Research Paper

Available upon request.

Demo

CHALEARN Virtual Lab

: Login

: Create New Challenge

: Create New Challenge

Project Purpose

What affects your health, the economy, climate changes, and which actions would have desired effects? ChaLearn has launched a series of challenges in machine learning to evaluate new algorithms capable of unraveling cause-effect relationships. Upcoming challenges will focus on high impact problems, including development projects of food supply/famine and the spread of crop disease http://ai-d.org/pdfs/Guyon.pdf. We need help to prepare our web portal to address such problems.

Brief Description

An environment to test cutting edge methods of causal discovery on data generated by complex systems

Specifics

Most existing platforms to test machine learning algorithms provide competitors with a pre-recorded dataset usually split into a labeled training set and an unlabeled test set. Only the organizers have access to the test data labels (truth values). The goal is simply for competitors to design a predictor (learning machine) using the training data and predict the test data labels. The competitors are then ranked according to the prediction accuracy. This paradigm has limitations, particularly to evaluate methods of causal discovery, because unraveling cause-effect relationships usually requires performing experiments on a system, which can be thought of as a form of data generation on-the-fly rather than working on pre-recorded data. In this project, we seek to emulate real situations in which experiments are performed. The new platform that will be developed will be designed to interface with realistic simulators of epidemiology, system biology, weather forecast, and econometrics. This will allow researchers in causal discovery to have access not just to passive data but to data generating models of complex systems. The new platform will replace and upgrade the causality workbench virtual lab http://www.causality.inf.ethz.ch/workbench.php by providing a universal interface between challenge portals and a backend serving data generated on-the-fly by simulators of complex systems.