Search and recommendations

Develop analytics and machine learning pipelines to provide users with the moments that matter in their audio and video content, utilizing NLP and computer vision. Some specific projects:- Automated annotation of videos with topics, subtopics, and paragraphs, reducing costs by 10 fold.- Identify speakers (match voice with name) using a bayesian deep learning approach.- Provide users watching a video with related topics from other videos.

Building an end-to-end recipe recommendation app based on real-time user preferences using Python, Javascript, and AWS. Aim to provide rapid comparison of recipes using word embeddings and nearest neighbors.

Led multiple projects to study the drivers of human decision-making.• Designed and conducted online experiments to determine subjective value and preference for art, clothing, and food items. Used JavaScript, R and Python.• Significantly improved prediction of behavior by modeling the semantic category of an item using cognitive science (category models) and NLP (word embedding models).• Received grants for work from the National Science Foundation and National Institute of Health.---Created a dataset of food stimuli and ratings based on a large US sample :• Worked with a team of ten to assess food choice in an online sample (N=1,075) for 138 foods with 17 ratings.• Developed a model to predict food choice using food ratings and nutritional facts in R. • Results used by other researchers to quantify unhealthy eating behaviors.

Examined the principles by which the human brain organizes visual information:• Assessed the contribution of distinct brain regions (time-series) for recognizing faces and objects. Used a model based on recursive feature elimination and the elastic-net in R.• Implemented causal inference of time-series data in R to understand information flow in the brain.

Applied machine learning models to predict human behavior and clinical outcomes:• Designed a project to determine how familiar faces are recognized accurately despite large variations over time and with the environment. Built a computational model of familiar face recognition using deep learning (OpenFace neural network in PyTorch). http://bit.do/famface• Developed a predictive algorithm to assess treatment remission for depression in R using the lasso and gradient boasted trees, outperforming predictions given by psychiatrists (31% improvement).• As a teaching assistant, created a new neuroscience lab introducing students to machine learning.

Improved the diagnosis and understanding of psychiatric disorders:• Analyzed behavioral and brain data by applying multiple methods (e.g., linear regression, support vector machines, and independent components analysis) to predict various psychiatric disorders.• Automated the subjective and painstaking method of visual inspection typically required to assess data quality of MRI brain data by quantifying various metrics of signal-to-noise in Python• Developed open-source Python code for graph network analyses of brain connectivity data as part of a team of four software developers. https://github.com/FCP-INDI/C-PAC

Built a toolbox for multivariate analysis of brain connectivity data to be fast and efficient:• Code was based on distance-based regression and implemented in R/C++ and Python to ensure a wide audience (https://github.com/czarrar/connectir). • Reduced the runtime of analyzing the connectome (all possible large-scale connections in the brain) for thousands of individuals from weeks to hours, all on a personal computer.

Studied the neural correlates of Autism Spectrum Disorder.

• Coordinated research studies (100+ participants) using MRI to identify psychiatric biomarkers.• Assessed test-retest reliability (e.g., intra-class correlation) of novel brain connectivity measures in R.• Built an online database in PHP and MySQL for participant recruitment and scheduling.