Cognitive Research:
Technology & Adaptive Perceptual Learning
Date: September 2017 - May 2018
Location: University of California, Los Angeles
Role: Research Assistant, Kellman Lab for Perceptual Learning
Accolades: UCLA Library Poster Grant
Citation: Kostas, S.C., Sankar, G., Simon, S., Cui, L., Massey, C.M., & Kellman, P.J (2018, May).Training Correlation Estimation with Perceptual Learning Modules. Poster presented at UCLA Research Day, Los Angeles, CA
Citation: Cui, L., Massey, C.M., & Kellman, P.J. (2018). Perceptual Learning in Correlation Estimation: The Role of Learning Category Organization. In T.T. Rogers, M. Rau, X. Zhu, & C.W. Kalish (Eds.), Proceedings of the 40th Annual Conference of the Cognitive Science Society (p. 264 - 269). Madison, WI: Cognitive Science Society.
Research Question:
How does perceptual learning (PL) in general and in different category organizations affect the accuracy of correlation estimations?
Processes:
PALMs is a teaching method by which students learn through a series of practice questions which are grouped into latent learning categories (e.g. based on visual similarity). Each question is immediately followed by accuracy feedback and an adaptive algorithm dictates the order and frequency with which questions from a given category are presented based on the accuracy and speed on all previous questions. This method is based on the hypotheses that questions can be grouped into meaningful categories based on their visual similarities and that there is an optimal inter-category sequential and temporal ordering at which students learn best. To that end, we investigated the level of learning achieved under different question orderings/timings as well as category groupings (correlation strength and trend shape) .A correlation strength arrangement is an intuitive grouping structure because it creates bins which correspond to what we were measuring, correlation. A shape arrangement is less intuitive but accounts for perceptual biases. Thus, we hypothesized that participants who receive the shape PALM will learn to ignore distracting surface features and, therefore, estimate correlations more accurately than participants who trained with the correlation strength PALM.Participants completed a pretest, training, post test, and demographics questionnaire. Training continued until the participant achieved mastery in every subcategory or bin in their module. Our preliminary findings indicated that intelligently learned question-ordering is not sufficient to achieve optimal learning because perceptual learning is heavily dependent on how well the question categories map to the cognitive needs of the student.
Responsibilities:
Once all data had been collected, I preformed data analyses in R to evaluate our findings.I, along with two other colleagues, were accepted to UCLA’s Undergraduate Research Week: Poster Conference where we presented our findings. In addition to presenting, I created our research team’s poster.
