Machine Learning Team

National Institute Of Mental Health

Investigators

Linked publications & trials

Abstract

This is the fourth annual report for the Machine Learning Team. The team comprised Francisco Pereira, Charles Zheng, Patrick McClure, Ka Chun Lam all research scientists -- and Yenho Chen (postbac IRTA) until November 2020 and January 2021, when Sebastian Bruch and Juan Antonio Lossio Ventura, respectively, joined the team. Sebastian Bruch and Yenho Chen left the team at the end of July 2021, and Nicole Kuznetsov became our new postbac IRTA in August 2021. We work very closely with the Data Science and Sharing Team (DSST), both on specific joint projects and informally across many others. Research activity: We continued research projects initiated in the previous year. These were the projects that led to a published article : 1) Martin Hebart and Chris Baker - Discover interpretable mental representations of objects from a large database of behavioral judgements, which can be used to predict human behavior on a variety of other tasks 2) Hanna Keren and Argyris Stringaris - Develop models to predict mood during a gambling task, based on participant characteristics, trial parameters, and experiment history. This project was extended to incorporate different model types, and new kinds of behavioral experiment. 3) Ana Inacio and Soohyun Lee - Determine whether the pattern of synaptic weights of four different GABAergic neuron types carries enough information to identify the input region (as part of a study of functional heterogeneity of principal neurons in layer II/III of sensory cortex) 4) Maryam Vaziri-Pashkam - Develop a method to produce interpretable representations of objects in terms of their affordances (action verbs that apply to them), grouping verbs into high-level modes of interaction, based on information extracted from text corpora. Several other projects led to submissions currently under review (or preparation): 1) Xiaoyu Ma and Zheng Li - Develop an approach for testing whether the encoding of information by prefrontal cortex neurons changes between experimental conditions, subject to confounding factors. 2) Adi Cymerblit-Sabba and Scott Young - Prediction of various experimental parameters from calcium imaging data in social and object habituation experiments, to test whether certain neural populations contain that information. 3) MLT + DSST - Develop a new method for extracting brain maps from a deep neural network trained to decode which task is being performed by a subject. The maps show where the information used by the network is present. The article introduces two different quantitative evaluation procedures for gauging the quality of the maps and compares our method with existing ones. 4) Dave Jangraw and Argyris Stringaris Help analyze data from many different experiments, imaging and behavioral, to show that rest periods lower the mood of participants, an effect we call resting-state dysphoria 5) Jingfeng Zhou and Geoff Schoenbaum Demonstrate that deep reinforcement learning can be used to produce a model of an agent performing a task, given only the same reward schedule and stimuli as an experimental subject. Validate that agent against neural data of a subject performing the same task. 6) Martin Hebart (now an external collaborator) and Chris Baker Develop an improved version of the method for extracting interpretable representations from behavior, using approximate Bayesian inference, which should perform better in smaller datasets (such as those collected by other groups in the IRP). 7) Joyce Chung and the COVID-19 survey group Develop a predictor of mental health status for participants, based on data from the subsample of participants that had also participated in prior NIH studies. This was used as part of an analysis to relate predicted mental health status to clinical symptoms and pandemic-related psychological and behavioral responses during lockdown. We also developed automated methods to quantify the emotional valence of free text produced by participants, as well as produce summaries of the main topics present in such text across time. In addition to these projects, we have ongoing research or service projects with researchers in the groups of Drs. Raznahan, Stringaris, Lee, Zarate, Pine, Baker, Conway, Martin, Atlas, Averbeck, and Bandettini, at NIMH, and Hope, at NIDA, none of which have yet led to submitted papers. Some of these have consumed months of person-effort, as we try to mix high-risk/high-impact projects with those satisfying specific needs of the PIs. We would like to highlight the main project we worked on in response to the COVID-19 pandemic. In addition to the work above, we are now developing methods to predict changes in a variety of psychiatric and behavioral outcomes, as a function of personal and environmental circumstances. The goal of this project is to identify the participant characteristics that help with resilience to pandemic stressors (or, conversely, worsen their effect). The pandemic delayed the hiring of some personnel on our team and DSST. This, in turn, forced us to delay a joint IRP-funded project with the goal of building re-usable deep neural network components (DNN) for many different feature types derived from MR imaging modalities. We were able to go back to the project in earnest in January of 2021, and carry out the first stage of our plan. This entailed developing a processing pipeline for dataset projects, as well as processing several large, public imaging datasets totaling almost 40,000 subjects. We also began developing our first DNN components, and preliminary trials indicate that they can be re-used successfully in small clinical datasets. Finally, we have a number of internal projects for developing new methods which, if they prove effective, we will bring to the attention of PIs who might have an application for them. Consultation/Service activity We carried out ad-hoc consultations on machine learning methods (which can take hours to days, if they require reading articles or finding/testing code), or co-advising of postbac and postdoc trainees (days-weeks. In addition to the PIs mentioned above, we provided consulting to people in the groups of Drs. Bandettini, Baker, Stringaris, Raznahan, Cox, Ungerleider, Li, Averbeck, Atlas, Pine, Lee, Conway, Martin, Zarate, Chung, Leibenluft, Brotman, Schmidt, Tejeda, and MacMahon, as well as Schoenbaum, Hope, and Lin (NIDA), and Horovitz, Johnson, and Nath (NINDS). Due to the pandemic and increased demand for data analysis and consulting, we reduced software development efforts. We have released our software to segment human faces and body parts in visual stimuli (e.g. photos or video frames), to go with published paper that used it. For more details, please see: A Deep Neural Network Tool for Automatic Segmentation of Human Body Parts in Natural Scenes https://arxiv.org/pdf/2009.09900 Education, Training, and Presentations Given the general fatigue with videoconferencing, and poor attendance at talks, we suspended the Machine Learning in Brain Imaging, Neuroscience and Psychology seminar, in collaboration with Javier Gomez-Castillo. We gave a presentation on our paper Mental representations of objects reflect the ways in which we interact with them at the annual meeting of the Cognitive Science Society. We also gave invited presentations at New York University, Carnegie Mellon University, Bowdoin College, and Williams College.