Comparing Deep Neural Network Features With Psychological Representations

July 1, 2020

Mentor: Shashi Kant
Project Members: Abhishek Jain, Aditya Jindal, Amartya Dash, Sahithi Macharla, Sanket Agrawal

Abstract:

Peterson et al 2016 attempted to evaluate the relation between deep representations and mental representations for making similarity judgments and see how well the former aligns with the latter. The similarity judgments, in turn, are calculated by taking a weighted inner product of the feature representations of the two objects. Results from the paper revealed that comparatively much better results were obtained when the weights were calculated using a regression technique rather than when they were considered to be constant unity. This seems to produce very favorable results and would indicate that deep learned representations can represent psychological representations. In this project, we are trying to test whether this really is this case. We start off by replicating their work, testing using other state-of-the-art models, and testing the method with different changes to testify if the method is really adapting to psychological representations. Apart from this, we have also designed a similar experiment in the NLP domain.

Poster: Link
Documentation: Link