Personal tools
You are here: Home Events ANC Workshop: Gavin Gray and Dragan Gasevic, Chair: Matthias Hennig

ANC Workshop: Gavin Gray and Dragan Gasevic, Chair: Matthias Hennig

— filed under:

What
  • ANC Workshop Talk
When Sep 29, 2015
from 11:00 AM to 12:00 PM
Where IF Room 4.31/4.33
Add event to calendar vCal
iCal

Gavin Gray

Shapley Relevance by Dropout

Which units in a neural network are useful for a given task? The Shapley value gives us a value for neurons collaborating in a task, but is impractical to calculate. Working from the idea of dropout, we design a new type of dropout that can maintain marginal distributions at each node. By adapting the methods used in variational autoencoders, similar to variational dropout, it is then possible to learn these
relevances.

 

Dragan Gasevic

Using learning analytics to uncover how learning unfolds over time:
Findings from a flipped computer engineering classroom

Understanding and optimizing learning and the environments in which learning happens are the main tasks of the field of learning analytics. Existing research has dedicated much attention to studies that aimed at identifying factors predicting different learning outcomes. However, there is the dearth of research that focuses on understanding how learning unfolds over a certain period of time under different instructional conditions. To address these issues, two main factors that influence students’ activities should be taken into consideration: i) instructional design of courses; and ii) students’ internal conditions (e.g., motivation, prior knowledge, and study skills). This talk will discuss findings of an on-going study conducted in the scope of a large undergraduate computer engineering course that followed a flipped classroom instructional design. The course used a wide range of interactive technologies and learning tasks designed to promote active and systematic learning during the entire course. The study results are based on the exploration of log data about activities of learners and products of students’ learning by using latent class analysis and hidden Markov models.