Informational and Introductory Session
When: Monday, November 21st, 9-11:30 am
Ithaca location: Weill Hall, Room 224 (light breakfast will be served)
NYC location: WebEx, link to be shared via email
This session will give basic information about the group and brief introductions to the research of many of the group's members. Those of you who plan attend via WebEx, please email amk2012@med.cornell.edu for the link.
December Seminar
To access a recording of this talk, please see the For Members page
When: Friday, December 16th, 10-11:00 am
Where: WebEx, link to be shared via email
Dr. Fabien Campagne
Training probabilistic models with semi-simulated datasets
The continued development of high-throughput assays in biology calls for probabilistic models to separate signal from noise and rank the observations made with these assays. Training deep neural networks with millions of labeled observations could generate suitable models. However, in biology, the cost of obtaining adequate labeled data is often prohibitive. We introduce signal in real datasets by simulation, to create the large semi-simulated datasets needed for training models. We focused on models to call somatic variations in sequencing data. We found that models trained with semi-simulated data are predictive when tested on a gold standard dataset. Furthermore, models trained with semi-simulation and subsequently trained with real labels outperform (AUC =0.969 [0.957-0.982], 95% confidence interval) models trained only with real labels (AUC= 0.911 [0.890-0.932]). Our results demonstrate the potential of semi-simulation to train assay-specific probabilistic models to discriminate signal from noise in high-throughput datasets.
January Seminar
When: Monday, January 9th, 10:15-11:15 am
Ithaca Location: Weill 226
NYC Location: WebEx, link to be shared via email
Dr. Haiyuan Yu - Title TBA
February Seminar
When: Monday, February 13th, 10-11:00 am
Ithaca Location: WebEx, link to be shared via email
NYC Location: WebEx, link to be shared via email
Dr. Ashish Raj - Towards computer-enabled precision radiology and neurology
Abstract:
The dawning era of precision medicine requires a tight integration of medicine with the computational sciences. In particular, imaging-facilitated informatics represents one of the largest, most promising opportunities, but there are many challenges. I will describe how research in my laboratory at Weill Cornell, the Image Data Evaluation and Analytics Laboratory (IDEAL), is furthering this goal, with a dual mission of both basic research and state of the art informatics, computational and clinical research.
In the first part of my talk I will highlight some ongoing imaging informatics, clinical and support service projects – all supported by our custom Research PACS and computing infrastructure. In the second part I will focus on original basic research on advanced medical imaging, visual computing, graph inference and computational neurology. In particular, our proposed “Network Diffusion” model of the spread of neurodegenerative diseases within the brain, and its use in computational prognostication. I will highlight recent advances in graphical models of brain activity and function. Taken together, these efforts open a way forward for precision neurology and imaging-facilitated precision health. To that end, I will suggest some future directions, in particular, the emerging science of imaging genetics.
March Seminar: (in conjunction with the Biomedical Imaging Research Seminar at WCM) Dr. Mert Sabuncu - Assistant Professor of Electrical Engineering and Biomedical Engineering, Cornell University
When: March 13th, 4-5pm
NYC location: Belfer Building, room 1501
Ithaca Location: WebEx, link to be shared via email
Title: Novel Computational Methods to Probe the Genetic Underpinnings of Brain Structure and Function
Abstract: In this talk, I will review some of our latest contributions in neuroimaging genetics – a nascent field that aims to examine the genetic underpinnings of neuroimaging measurements. First, I will present an algorithm that we call MEGHA (for Massively Expedited Genomewide Heritability Analysis). I will illustrate how MEGHA allows us to examine the heritability of volumetric measures of brain morphology using MRI and genome wide data from a large-scale dataset. Then I will present a new metric we have introduced that quantifies the heritability of multidimensional traits. I will demonstrate how this metric can be used to study the genetic basis of the shapes of neuroanatomical structures. Finally, I will review some recent work where we propose a novel strategy to compute heritability estimates based on repeat measurements. I will illustrate how this tool can be used to examine the heritability of resting-state functional connectivity.
​
April Seminar: Anita Govindjee, MS - Solutions Architect, IBM Watson Ecosystem
When: Monday, April 17th, 11:15 am-12:15 pm
NYC location: WebEx, link to be shared via email
Ithaca Location: Weill Hall, Room 224
Title: IBM Watson in Healthcare
May Seminar: Dr. Rene Vidal, PhD - Professor of Biomedical Engineering, Computer Science, Mechanical Engineering, and Electrical and Computer Engineering at Johns Hopkins University
When: May 8th, 4:15-5:15pm
NYC location: link to be shared via email
Ithaca Location: Phillips Hall, Room 233
Title: Automatic Methods for the Interpretation of Biomedical Data
Abstract: In this talk, I will overview our recent work on the development of automatic methods for the interpretation of biomedical data from multiple modalities and scales. At the cellular scale, I will present a structured matrix factorization method for segmenting neurons and finding their spiking patterns in calcium imaging videos, and a shape analysis method for classifying embryonic cardiomyocytes in optical imaging videos. At the organ scale, I will present a Riemannian framework for processing diffusion magnetic resonance images of the brain, and a stochastic tracking method for detecting Purkinje fibers in cardiac MRI. At the patient scale, I will present dynamical system and machine learning methods for recognizing surgical gestures and assessing surgeon skill in medical robotic motion and video data.
July Seminar: Byron Wallace, PhD - Assistant Professor, College of Computer and Information Science, Northeastern University
When: Friday, July 14th, 1:00 pm-2:00 pm, refreshments served at 12:45pm
NYC location: Belfer Building, Room 302-D
Ithaca Location: Zoom Meeting, link to be sent
Title: Expediting Clinical Evidence Synthesis via Machine Learning, Natural Language Processing and Crowdsourcing
Abstract: Evidence-based medicine (EBM) looks to inform patient care with the totality of the available evidence. Systematic reviews, which statistically synthesize the entirety of the biomedical literature pertaining to a specific clinical question, are the cornerstone of EBM. These reviews are critical to modern healthcare, informing everything from national health policy to bedside decision-making. But conducting systematic reviews is extremely laborious and hence expensive. Producing a single review requires thousands of expert hours. Moreover, the exponential expansion of the biomedical literature base has imposed an unprecedented burden on reviewers, thus multiplying costs. Researchers can no longer keep up with the primary literature, and this hinders the practice of evidence-based care.
I will discuss recent work on machine learning and natural language processing approaches that look to optimize the practice of EBM and thus mitigate the burden on those trying to make sense of the clinical evidence base. Specifically, I will describe methods for automatic identification of clinically salient information in full text articles (descriptions of the population, interventions and outcomes studied; collectively referred to as PICO elements). And I will describe work on automating the important step of assessing clinical trials for risks of bias. These tasks pose challenging problems from a machine learning vantage point, motivating novel approaches. For example, I will describe a new method for interpretable neural text classification which was motivated by our work on automating bias assessment for articles describing clinical trials. I will present evaluations of these methods in the context of EBM. Finally, I will highlight promising directions moving forward toward automating evidence synthesis, including hybrid crowd-sourced/machine learning systems.
Speaker Bio: Byron Wallace is an assistant professor in the College of Computer and Information Science at Northeastern University. He holds a PhD in Computer Science from Tufts University, where he was advised by Carla Brodley. He has previously held faculty positions at the University of Texas at Austin and at Brown University. His primary research is in machine learning and natural language processing methods, with an emphasis on their application in health informatics.
Wallace's work has been supported by grants from the National Institutes for Health (NIH), the National Science Foundation (NSF), and the Army Research Office (ARO). He won the Tufts University 2012 Outstanding Graduate Researcher award and his thesis work was recognized as The Runner Up for the 2013 ACM Special Interest Group on Knowledge Discovery and Data Mining (SIG KDD) Dissertation Award. He co-authored the winning submission for the Health Care Data Analytics Challenge at the 2015 IEEE International Conference on Healthcare Informatics, and his recent work with colleagues received the 2017 Distinguished Clinical Research Informatics Paper Award at the American Medical Informatics Association Joint Summits on Translational Sciences.
August Seminar: Finale Doshi-Velez, PhD - Assistant Professor of Computer Science, Harvard University
When: Monday, August 28th, 4:15-5:15 pm, refreshments served
NYC location: Zoom Meeting, link to be sent
Ithaca Location: Phillips Hall, Room 233
Title: Prediction-Constrained Training for Interpretable Models in Healthcare
Abstract: Generative models, such as mixtures or topic models, are often used to explore patterns in data. A common pipeline is to first apply such a model as a form of dimensionality reduction, and then look for enrichment for things we care about, that is, are certain clusters or topics correlated with certain outcomes? In this way, we seek to combine explanation -- clusters and topics are easy to interpret -- with outcomes. However, as the dimensionality of our data grow, it is often the case that there are many patterns in the data, and few of them relevant to outcome of interest. Thus, there has been a line of work in supervised generative models -- supervised mixtures, supervised topic models -- that try to align the discovered structure with outcomes. These supervised models are rarely used in practice, however, and in this work we describe ways in which the current set of objective functions used to optimize supervised generative models fall short of their goal. We next describe an alternative objective which addresses these shortcomings. Finally, we demonstrate early results on a task of identifying patterns that can help us determine which antidepressants will work best for patients with major depression.
Joint work with: Michael C. Hughes, Tom McCoy, Roy Perlis, Gabe Hope, Leah Weiner, Erik Sudderth
Speaker Biography: Finale Doshi-Velez is an Assistant Professor in Computer Science at the Harvard Paulson School of Engineering and Applied Sciences, where her interests lie in the intersection of machine learning and healthcare. She completed her postdoctoral fellowship at Harvard Medical School, her PhD from MIT, and her MSc as a Marshall Scholar at the University of Cambridge.
September Seminar: Jenna Wiens, PhD - Assistant Professor, Computer Science and Engineering, University of Michigan Ann Arbor
When: Friday, September 8th, 1:00-2:00 pm, refreshments served at 12:45 pm
NYC location: Belfer Building, Room 302-C/D
Ithaca Location: Zoom Meeting, link to be sent
Title: Data-Driven Models for Patient Risk Stratification - High Accuracy is Not Enough
Abstract: The increasing availability of electronic health data has led to the investigation of machine learning (ML) techniques for improving clinical decision making. In particular, patient risk stratification models could, in theory, facilitate the targeting of specific interventions to high-risk groups. However, for data-driven systems to become widely and safely adopted in clinical care, there remain several key research challenges. More specifically, many existing risk stratification models while accurate lack credibility and actionability. In this talk, I will present and motivate these challenges in the context of building models to predict patient risk of healthcare-associated infections. By adapting techniques from time-series classification and multi-task learning, I will show how one can learn accurate models for patient risk stratification for healthcare-associated infections with Clostridium difficile. Moreover, I will discuss new and ongoing research directions in ML that aim to increase the credibility and actionability of such patient risk stratification models without sacrificing accuracy.
Speaker Biography: Jenna Wiens is an Assistant Professor of Computer Science and Engineering (CSE) at the University of Michigan in Ann Arbor. Her primary research interests lie at the intersection of machine learning, data mining, and healthcare. She is particularly interested in time-series analysis and transfer/multitask learning. The overarching goal of her research agenda is to develop the computational methods needed to help organize, process, and transform patient data into actionable knowledge. Jenna received her PhD from MIT in 2014. In 2015 she was named Forbes 30 under 30 in Science and Healthcare; she received an NSF CAREER Award in 2016; and recently she was named to the MIT Tech Review's list of Innovators Under 35.
November Seminar: Richard Leahy, PhD - Dean's Professor of Electrical Engineering, Biomedical Engineering and Radiology, University of Southern California
When: November 17th, 4:15 pm, refreshments served
NYC location: Zoom, link to be sent
Ithaca Location: Phillips Hall, Room 233
Title: Discovering and exploring brain networks with spontaneous brain activity: methods and applications
December Seminar: Jiayu Zhou, PhD - Assistant Professor of Computer Science and Engineering, Michigan State University
When: December 8th, 1 pm, refreshments served
NYC location: Belfer Building 204-A/B
Ithaca Location: Phillips Hall, Room 310
Title: Multi-task Learning and its Applications to Biomedical Informatics
Abstract: The recent decade has witnessed a surging demand in data analysis, where we built machine learning models for various data analysis tasks. The multi-task learning is a machine learning paradigm that bridges related learning tasks and transfers knowledge among the tasks. Multi-task learning is currently widely used in computational medicine such as predictive modeling from electronic medical records, modeling disease progression and drug effect prediction. In the seminar, we first introduce the basics of multi-task learning. We then show how multi-task learning can benefit the study of the progression of Alzheimer’s disease. We also introduce a distributed framework for multi-task learning that allows privacy-preserving computation over distributed patient cohorts. The seminar is concluded by a discussion of future directions of multi-task learning.
Biography: Jiayu Zhou is currently an Assistant Professor in the Department of Computer Science and Engineering at Michigan State University. He received his Ph.D. degree in computer science from Arizona State University in 2014. He has a broad research interest in large-scale machine learning and data mining, and biomedical informatics. He served as technical program committee members of premier conferences such as NIPS, ICML, and SIGKDD. Jiayu’s research is supported by National Science Foundation and Office of Naval Research. His papers received the Best Student Paper Award in 2014 IEEE International Conference on Data Mining (ICDM), the Best Student Paper Award at 2016 International Symposium on Biomedical Imaging (ISBI), and Best Paper Award at 2016 IEEE International Conference on Big Data (BigData).
March Seminar: Daniel B. Neill, PhD - Associate Professor of Information Systems
Director, Event and Pattern Detection Laboratory, Carnegie Mellon University
When: March 2nd, 1 pm, refreshments served at 12:45 pm
NYC location: Belfer 302-A/B
Ithaca Location: to join via Zoom, please click here
Title: Machine Learning for Population Health and Disease Surveillance
Abstract: Over the past decade, our lab has developed a variety of new statistical and computational approaches for early and accurate detection of emerging outbreaks of disease. This talk will describe our work in addressing three distinct public health challenges: syndromic surveillance using small-area count data, drug overdose surveillance using multidimensional case data, and asyndromic surveillance using free-text emergency department chief complaint data. In the first problem setting, we monitor a set of known syndrome types (e.g., gastrointestinal illness) and identify space-time clusters of disease. In the second problem setting, we use the multiple dimensions of each case (age, race, gender, location, and drug types) to identify emerging patterns of fatal accidental overdoses affecting specific subpopulations. In the third problem setting, we identify clusters of cases that are of interest to public health but do not correspond to existing syndrome categories, such as “novel” disease outbreaks with previously unseen patterns of symptoms.
Across all three problem settings, we develop new “fast subset scan” approaches to deal with the size and complexity of real-world data. Subset scanning is a novel pattern detection approach which treats the detection problem as a search over subsets of data records and attribute values, finding those subsets which maximize an expectation-based scan statistic. One key insight is that this search over subsets can be performed very efficiently, reducing run times from years to milliseconds, using the "linear-time subset scanning" property of many commonly used likelihood ratio scan statistics. These fast subset scanning approaches enable accurate, precise, and computationally efficient detection of emerging public health threats, providing state and local health departments with the situational awareness needed for early and targeted interventions.
Biography: Daniel B. Neill is the Visiting Professor of Urban Analytics at New York University's Center for Urban Science and Progress. He is currently on approved leave from Carnegie Mellon University's Heinz College, where he has been the Dean's Career Development Professor, Associate Professor of Information Systems, and Director of the Event and Pattern Detection Laboratory (epdlab.heinz.cmu.edu). Prof. Neill holds courtesy appointments in the Machine Learning Department and Robotics Institute at CMU's School of Computer Science and is an adjunct professor in the University of Pittsburgh's Department of Biomedical Informatics. He received his M.Phil. from Cambridge University and his M.S. and Ph.D. in Computer Science from CMU. His research focuses on machine learning and event detection in massive datasets, with applications to individual, population, and community health. Prof. Neill currently serves as advisor to the board of directors of the International Society for Disease Surveillance, "AI and Health" Editor and Associate Editor of IEEE Intelligent Systems, and co-chair of the International Conference on Smart Health. Dr. Neill was the recipient of an NSF CAREER award and was named one of IEEE Intelligent Systems' "top ten AI researchers to watch".
May Seminar: Shamil Sunyaev, PhD - Professor of Medicine and Biomedical Informatics, Harvard Medical School, Distinguished Chair of Computational Genomics and Research Geneticist at Brigham & Women’s Hospital
When: May 22, 2-3pm
NYC location: Belfer Research Building, Room 302-A
Ithaca Location: Zoom, link to be sent
Title: Deleterious mutations in humans
Abstract: The evolutionary cost of gene loss is a central question in genetics and has been investigated in model organisms and human cell lines. However, estimates of the selection and dominance coefficients in humans have been elusive. We analyzed large-scale sequencing data to make genome-wide estimates of selection against heterozygous loss of gene function. Using this distribution of selection coefficients for heterozygous protein-truncating variants (PTVs), we provide corresponding Bayesian estimates for individual genes. We find that genes under the strongest selection are enriched in embryonic lethal mouse knockouts, Mendelian disease-associated genes, and regulators of transcription. There is a long-standing theoretical argument whether deleterious mutations act independently or synergistically, so that each additional deleterious allele results in a larger decrease in relative fitness. Negative selection with synergistic epistasis should produce negative linkage disequilibrium between deleterious alleles and, therefore, an underdispersed distribution of the number of deleterious alleles in the genome. Indeed, we detected underdispersion of the number of rare loss-of-function alleles in eight independent data sets from human and fly populations. Thus, selection against rare protein-disrupting alleles is characterized by synergistic epistasis, which may explain how human populations persist despite high genomic mutation rates.
Biography: Dr. Shamil Sunyaev is Professor of Biomedical Informatics at Harvard Medical School and Professor of Medicine at Brigham & Women’s Hospital and Harvard Medical School. He is a computational genomicist and geneticist. Research in his lab encompasses many aspects of population genetic variation including the origin of mutations, the effect of allelic variants on molecular function, population and evolutionary genetics, and genetics of human complex and Mendelian traits. He developed several computational and statistical methods widely adopted by the community. Sunyaev obtained a PhD in molecular biophysics from the Moscow Institute of Physics and Technology and completed his postdoctoral training in bioinformatics at the European Molecular Biology Laboratory (EMBL). He is an Associate Member at Broad Institute of MIT and Harvard.
July Seminar: Andy Swift, MD, PhD - Cardiothoracic Radiologist, Department of Infection, Immunity & Cardiovascular Disease, The University of Sheffield
When: July 17, 1-2pm
NYC location: TBA
Ithaca Location: Zoom, link to be sent
Title: Feature learning the heart: experience in pulmonary hypertension
Andy Swift is a Cardiothoracic Radiologist based in Sheffield, UK who is visiting Columbia medical center working with Dr. Graham Barr. Andy is a Wellcome Trust research fellow interested in improving the cross-sectional imaging assessment of cardiac and pulmonary diseases. He is currently exploring different image analysis and data approaches to diagnose pulmonary hypertension. Andy has developed both diagnostic and prognostic algorithms using regression in pulmonary hypertension. Andy is now working with a computer scientist Haiping Lu based in Sheffield, together they have developed a feature learning workflow for cardiac magnetic resonance images. He will present some interesting early results using this method.
April Seminar: Archana Venkataraman, PhD - John C. Malone Assistant Professor, Electrical and Computer Engineering, Johns Hopkins University
When: April 19th, 11am-12pm
NYC location: Zoom, link to be sent
Ithaca Location: Rhodes Hall, Room 310
​
Title: Generative Models to Decode Brain Pathology
This talk will highlight three ongoing projects in my lab that span a range of methodologies and clinical applications. First, I will develop a joint optimization framework to predict clinical severity from resting-state fMRI data. Our model is based on two coupled terms: a generative non-negative matrix factorization and a discriminative linear regression. This project is part of our larger effort to better characterize heterogeneous patient cohorts. Next, I will describe a spatio-temporal model to track the spread of epileptic seizures from EEG data. Unlike conventional approaches, our model relies on a latent network structure that captures the hidden state of each EEG channel; the latent variables are complemented by an intuitive likelihood model for the observed neuroimaging measures. This project takes the first steps toward noninvasive seizure localization. Finally, I will highlight a very recent initiative in my lab to manipulate emotional cues in human speech. Our long-term goal is to create a naturalistic therapy for autism.
​
Biography: Archana Venkataraman is a John C. Malone Assistant Professor in the Department of Electrical and Computer Engineering at Johns Hopkins University. She directs the Neural Systems Analysis Laboratory and is a core faculty member of the Malone Center for Engineering in Healthcare. Dr. Venkataraman’s research lies at the intersection of multimodal integration, network modeling and clinical neuroscience. Her work has yielded novel insights in to debilitating neurological disorders, such as autism, schizophrenia and epilepsy, with the long-term goal of improving patient care. Dr. Venkataraman completed her B.S., M.Eng. and Ph.D. in Electrical Engineering at MIT in 2006, 2007 and 2012, respectively. She is a recipient of the CHDI Grant on network models for Huntington's Disease, the MIT Lincoln Lab campus collaboration award, the NIH Advanced Multimodal Neuroimaging Training Grant, the National Defense Science and Engineering Graduate Fellowship, the Siebel Scholarship and the MIT Provost Presidential Fellowship.
May Seminar: Minh Do, PhD- Professor, Department of Electrical and Computer Engineering, University of Illinois, Urbana-Champaign
When: May 6th, 1:00-2:00 pm
NYC location: via Zoom
Ithaca Location: Rhodes 310
​
Title: Imaging Genomics
There is an increasing number of large-scale, publicly open datasets with multimodal imaging and genomic data, notably The Cancer Genome Atlas (TCGA) and The Cancer Imaging Archive (TCIA). Connecting genotypes to image phenotypes is crucial for a comprehensive understanding of cancer. To learn and exploit such connections, new data analysis approaches must be developed for the better integration of imaging and genomic data. First, we present a novel, general framework for predicting arbitrary genomic markers from imaging features called Discriminative Bag-of-Cells (DBC). Second, we describe several improvements in deep learning for the crucial step in accurate and efficient segmentation of cell nucleis. Third, we present a computational inference to quantify the spatial distribution of various cell types within a tumor. Finally, we propose the use of canonical correlation analysis (CCA) and a sparse variant as a discovery tool for identifying connections between gene expression and cell imaging features for joint genotype-phenotype analysis of cancer.
Speaker Biography: Minh N. Do received the B.Eng. degree in Computer Engineering from the University of Canberra, Australia, in 1997, and the Dr.Sci. degree in Communication Systems from the Swiss Federal Institute of Technology Lausanne (EPFL) in 2001. Since 2002, he has been a Professor in the Department of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign (UIUC), and held joint appointments with the Coordinated Science Laboratory, Beckman Institute, Department of Bioengineering, and Department of Computer Science. He received a Silver Medal from the 32nd International Mathematical Olympiad in 1991, University Medal from the University of Canberra in 1997, Doctorate Award from the EPFL in 2001, CAREER Award from the National Science Foundation in 2003, Xerox Award for Faculty Research from UIUC in 2007, and Young Author Best Paper Award from IEEE in 2008. He was an Associate Editor of the IEEE Transactions on Image Processing, and a member of several IEEE Technical Committees on Signal Processing. He was elected as an IEEE Fellow in 2014 for his contributions to image representation and computational imaging. He has contributed to several tech-transfer efforts, including as a co-founder and CTO of Personify and Chief Scientist of Misfit.
May Seminar: Ronilda Lacson, MD PhD- Assistant Professor of Radiology, Harvard Medical School and Brigham and Women's Hospital
When: May 24th, 2:30-3:30 pm
NYC location: Belfer Research Building (413 E69 St), BB-204A/B
Ithaca Location: Zoom, link to be sent
​
Title: Natural Language Processing and Machine Learning Applications in Medicine
A major component of natural language processing in medicine involves understanding the full meaning of textual data to provide useful information for providers and patients. Textual data is readily available in electronic health records, including radiology reports. In clinical medicine, information derived from these data sources informs providers when making clinical decisions as well as provides structured information for evaluating initiatives focused on patient safety and quality of care. This talk will describe developed and validated natural language processing (NLP), information extraction and data analytic systems that employed machine learning and traditional statistical approaches. Feature selection and combination will be described, using unsupervised and supervised algorithms. Finally, evaluation frameworks will be presented to compare various models and approaches.
​
Biography: Dr. Ronilda Lacson completed her PhD in Computer Science and MS in Medical Informatics degrees at the Massachusetts Institute of Technology. She completed her research fellowship in Biomedical Informatics at the Brigham and Women's Hospital (BWH) and Harvard Medical School (HMS) Biomedical Informatics Research Training program, after completing her MD degree from the University of the Philippines and her Internal Medicine residency training at Harbor Hospital Center in Baltimore, MD. As Principal Investigator of multiple grants, she has published extensively on implementation and evaluation of clinical decision support systems, diagnostic process safety and applications of artificial intelligence and machine learning in medicine. Her research spans a wide range of topics in biomedical informatics and includes natural language processing and machine learning to extract structured and meaningful data from clinical narratives and medical records. She is currently an Assistant Professor of Radiology at HMS and Faculty at the Center for Evidence-Based Imaging at BWH.
September Seminar: Michele Santacattarina, PhD - Postdoctoral Associate, Cornell TRIPODS Center for Data Science for Improved Decision Making and Cornell-Tech
When: September 23rd, 11 am - 12 pm
NYC location: Weill Greenberg Center (1305 York), 2nd floor, WGC-A
Ithaca Location: Weill Hall, room 224
​
Title: Kernel Optimal Orthogonality Weighting for Estimating Effects of Continuous Treatments
Abstract: Many scientific questions require estimating the effects of continuous treatments, which relationships with an outcome are usually described by dose-response curves. Outcome modeling and methods based on the generalized propensity score are the most commonly used methods to evaluate continuous effects. However, these methods may be sensitive to model misspecification. In this paper, we propose Kernel Optimal Orthogonality Weighting (KOOW), a convex optimization-based method, for estimating effects of continuous treatments. KOOW finds weights that minimize the penalized weighted functional covariance between the continuous treatment and the confounders. By minimizing this quantity while simultaneously penalizing the weights, KOOW successfully provides weights that optimally orthogonalize confounders and the continuous treatment. We describe its properties and valuate its comparative performance in a simulation study. Using data from the Women's Health Initiative observational study, we apply KOOW to evaluate the effect of red meat consumption on blood pressure.
Biography: Dr. Michele Santacatterina is a postdoctoral associate at the Cornell TRIPODS Center for Data Science for Improved Decision Making and Cornell Tech. His research focuses on the development and applications of statistical methods built upon the interaction between causal inference, machine learning, and mathematical optimization for optimal decision making using experimental and observational data. In general, he is interested in using machine learning methodologies to better understand cause-effect relationships.
Burnell Symposium: Rosalind Picard, ScD, FIEEE- MIT Media Lab
When: November 13th, 4:30-5:30 pm, panel discussion 5:30-6pm
NYC location: Cornell Tech, Bloomberg Center Room 161 (RSVP required)
Ithaca Location: Weill Hall 224
​
Title: Emotion AI: How well can we forecast and change future mental health?
Biography: is founder and director of the Affective Computing Research Group at the Massachusetts Institute of Technology (MIT) Media Lab, co-director of the Media Lab's Advancing Wellbeing Initiative, and faculty chair of MIT's Mind+Hand+Heart Initiative. She has co-founded Affectiva, Inc. providing emotion AI technology, and Empatica, Inc. creating sensors and analytics to improve health. Empatica created the first AI-based smart watch cleared by FDA in Neurology (for monitoring seizures).