Machine Learning Postdoctoral Fellow

GENERAL SUMMARY/ OVERVIEW STATEMENT:

Massachusetts General Hospital (MGH) Department of Neurology and Harvard Medical School (HMS) is seeking a signal processing and machine learning post-doctoral fellow for a full-time position in the Gupta Lab. The fellow will be applying diverse methodologies to uncover information in multimodal data collected from individuals with common and rare neurodegenerative diseases. Large and rapidly growing datasets include wearable sensor, video, audio, mobile device, and computer mouse data. The candidate will become immersed in a multidisciplinary and collaborative environment, consisting of neurologists, computer scientists, clinical trials experts, and drug development teams, from varied groups in academia and industry. In addition to working with large datasets, the candidate will lead analysis of novel N-of-1 study data, where both the therapy and the digital measurement strategy is custom created for an individual with a neurological disorder. The candidate will additionally contribute to an open source initiative to advance the field of digital phenotyping in neurology (https://neurobooth.mgh.harvard.edu/). The role is ideally suited for an individual with a rigorous training in signal processing and time series analysis methods (particularly wearable sensor and video data), including deep neural networks and state space models, with a particular interest in model interpretability. Individuals will be well positioned to pursue future opportunities in either academia or industry, and work will be tailored to achieve these goals.

PRINCIPAL DUTIES AND RESPONSIBILITIES:

• Use computer vision and machine learning techniques to analyze video data collected in Neurobooth

• Use signal processing and machine learning techniques to extract and learn informative features and embeddings from wearable sensor data - both from data collected during prescribed behavioral tasks and during continuous and passive monitoring at home

• Develop and implement supervised machine learning approaches for predicting disease severity and estimating disease progression

• Develop and implement unsupervised machine learning approaches for uncovering latent features and latent classes in neurodegenerative diseases

• Develop pipeline for visualizing and modeling multimodal time series data. Creatively apply a broad range of methods, including deep neural networks, HMMs, SSMs, Gaussian Processes

• Help frame cross-modal learning and data fusion problems to integrate information across multiple data types being collected

• Work on collaborative projects with other postdoctoral fellows in the group

• Engage collaborations within MGH Neurology as well as groups at Harvard SEAS, MIT CSAIL, Broad Institute, Duke CS, UMass Amherst CS, and Biogen

• Form new academic and industry collaborations

• Use coding best practices

• Make methods and coded data widely available

• Publish journal articles and present work at computer science and clinical conferences

• Participate in grant writing and preparation

• Work closely with the data collection team members, ensuring that data are being collected as expected and adjustments are being made to maximize the quality of the data

• Work with data architects and data managers to develop scalable data analysis pipelines

SKILLS & COMPETENCIES REQUIRED:

• Demonstration of the following characteristics: creative problem-solver, detail-oriented, highly organized, self-motivated, and able to work independently as well as within cross-functional teams

• Exceptional written and oral communication skills

• Ability to explain the essence of complex methods to non-technical audiences

EDUCATION:

Doctoral Degree Required

Field of Study/Additional Specialized Training:

A quantitative discipline such as computer science, engineering, math or physics

Required:

• Formal PhD training in computer science, physics, math, or related field

• Experience building neural network architectures

• Expertise with signal processing and time series analysis methods

• Proficiency with supervised and unsupervised machine learning methods

• Experience working with video and wearable sensor data

• Ability to work in Python environments

WORKING CONDITIONS:

The postdoctoral fellow will work with a team of physicians, physician-scientists, computer scientists, and clinical research coordinators. Work will primarily be performed in a computational laboratory space on the MGH main campus, although regular meetings with collaborating groups in the Boston/Cambridge area are expected. On occasion work may be performed in MGH neurology clinic space or clinical research space at other locations at MGH in order to gain clinical exposure and to inform the data collection process. Travel to conferences and meetings to present research is expected.