[cs-talks] PhD Thesis Defense: Sanaz Bahargam, 6/15 @12:15pm, MCS 148
Harrington, Jacob Walter
jwharrin at bu.edu
Mon Jun 12 09:45:28 EDT 2017
Ph.D. Thesis Defense
Machine Learning Approaches to Educational Applications
Sanaz Bahargam, BU
Thursday, June 15th, 2017 at 12:15 pm in MCS 148
The recent advances in computational techniques have triggered a significant shift in the educational landscape. Perhaps the most notable outcome of this shift has been the emergence of Massive Open Online Courses (MOOCs). MOOCs have enabled educators to bring together a large number of students from all around the globe and deliver the educational content to students through a new medium. However, they have not lived up to original hopes. Consequently, there is an ongoing effort to improve these online platforms. For instance, a crucial but missing feature in MOOCs is the ability to form effective collaboration groups such that grouping facilitates learners retention and develops important skills in critical thinking and co-construction of knowledge. Moreover, online learning is most effective when delivered by customized study plans based on the difficulty of the materials and the requirements of the students. The work presented in this thesis proposes new computational solutions to improve online educational platforms. This thesis consists of two main parts: (I) forming teams in educational settings and (II) automating the process of creating a syllabus.
In the first part, we study two team-formation problems, namely Guided Team Partitioning problem (GTP) and Team Formation with Personalized Study Plans (TeamPlan). Team formation refers to the problem of partitioning a pool of candidates into multiple teams with different objectives. GTP endeavors to create teams that target a specific ability level. The TeamPlan problem has a different formulation and aims to achieve two goals simultaneously: (1) group students so that they can maximally benefit from peer interaction and (2) find a customized study plan for each group.
In the second part, we propose a time-evolving topic discovery method which, in addition to the extracted topics, is able to identify the evolution of topics over time, and their level of difficulty. Our method is based on a novel formulation of Constrained Coupled Matrix-Tensor Factorization, which adopts constraints that are well motivated for, and are necessary for high-quality topic discovery. Our approach has implications for automatic curriculum design using the extracted topics, where the notion of difficulty is necessary for the proper modeling of prerequisites and advanced concepts.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cs-talks