The Challenges of Model Transfer in MOOC Data Science

Speaker:        Dr. Una-May O'Reilly
                Principal Research Scientist
                AnyScale Learning For All (ALFA) Group
                MIT Computer Science and Artificial Intelligence Laboratory
                MIT

Title:          "The Challenges of Model Transfer in MOOC Data Science"

Date:           Monday, 14 November 2016

Time:           4:00pm - 5:00pm

Venue:          Lecture Theater F (near lifts 25/26), HKUST

Abstract:

A wealth of detailed observations can be collected by an online learning
platform (such as edX or Coursera) as learners use it.  This data, when
transformed into knowledge in the form of models through Machine Learning,
can inform better teaching practices and shed new light on how learning
occurs. However, while data from a course that is analyzed after the
course has been taught can yield behavioral models (e.g. stopout) that are
helpful retrospectively, it is not straightforward to develop models that
can be transferred and applied to a new course with different students and
content. We will discuss the technicalities of the challenges in
transferring models and introduce an ensemble modeling approach due to
Sebastien Dubois and Kalyan Veeramachaneni, members of the ALFA group,
that facilitates robust model transfer.


*******************
Biography:

Una-May O'Reilly is leader of the AnyScale Learning For All (ALFA) group
at Massachusetts Institute of Technology Computer Science and Artificial
Intelligence Laboratory. ALFA focuses on scalable machine learning,
evolutionary algorithms, and frameworks for knowledge mining, prediction
and analytics. The group has projects in clinical medicine knowledge
discovery, cyber security and MOOC technology. She received the EvoStar
Award for Outstanding Achievements in Evolutionary Computation in Europe
in 2013, became a ISGEC fellow in 2002 and is Vice-Chair of ACM SIGEVO.