My Thesis Research: Prediction Error Integration

A schematic example for one of the models that I have developed to account for prediction error integration.

Learning is essential to survive and adapt to the ever-changing environment. Historically, a large body of theoretical work views learning as a function of a feedback signal called the prediction error. The prediction error is simply the difference between the expected and the experienced outcome. If an animal hears a noise that is often followed by a food pallet and then it gets the food pallet, no learning happens. However, if it does not expect a food reward but it gets a food reward, the animal develops a positive prediction error — all of which describes the activity of the dopamine neurons.

What happens if you have multiple cues such as noise and light (and you can add as many as you can). Here, different PE-based theories diverge. While some theories assume a separate prediction error per cue, other theories assume a common prediction error to all the present cues. Both views are supported by substantial behavioral evidence.

My work is to study how does the human brain disentangle both prediction errors, and what are the neural/behavioral substrates of each signal. A final piece of my research is to develop a computational model that can account for the results supporting both views of learning and make novel predictions. All of which is still a work in progress but I have presented some initial results along with the computational model that I have developed.

My work in the prediction error integration has been presented in SfN 2019 and has won the Doctoral Students Research Grant Round 14 (DSRG).

Alhazmi F., Krishnan, A., Esber, G. (2019) A Computational Model for The Modulation of Learning by the Common and Separate Prediction Errors. Presented at Society for Neuroscience 2019. Chicago, IL.

Alhazmi, F., Kang, M., Reverte, I., Krishnan, A., Esber, G. (2019) Modulation of Learning by the Integration of Common and Separate Prediction Errors. Brooklyn College Science Day 2019. (*Best Poster Award, 3rd Place, at Brooklyn College Science Day)

Multivariate methods to analyze and understand neuroimaging literature

Neuroscience is a growing field that spans a wide array of specializations and scientific fields. In this project, we ask the question: what are the basic camps of neuroimaging research? And, more specifically, how does the semantic space of most neuroimaging literature look like? We used 10K+ published abstracts (from the Neurosynth database), and analyzed their abstracts along with the reported brain activations. The results of these projects are have been presented and published in Human Brain Mapping:

Alhazmi, F., Beaton, D., Abdi, H. (2018)  The Latent Semantic Space and Corresponding Brain Regions of the Functional Neuroimaging Literature. Human Brain Mapping. 2018;39:2764–2776.

Alhazmi, F., Krishnan, A. (2019) Multivariate meta-analytic tools to explore associations between cognitive functions & brain regions. The 25th Annual Meeting of the Organization of the Human Brain Mapping. Rome, Italy (June 9-13).