Contact: prefix@suffix where prefix=ssaria and suffix=cs.jhu.edu
Other Affiliations: Mathematical Institute for Data Science (MINDS), Institute for Computational Medicine, Laboratory for Computational Sensing and Robotics, Armstrong Institute for Patient Safey and Quality, Center for Population Health Information Technology, and Center for Language and Speech Processing
My interests span Bayesian and probabilistic modeling approaches for addressing challenges associated with modeling and prediction in complex, real-world temporal systems. My recent work has focused on large scale modeling with Bayesian methods, methods for counterfactual reasoning, Bayesian nonparametrics, and Gaussian Processes. I am also excited about addressing challenges related to the use of data-driven tools for decision-making.
I direct the Machine Learning and Healthcare Lab at Johns Hopkins University. We are interested in enabling new classes of diagnostic and treatment planning tools for healthcare—tools that use statistical machine learning techniques to tease out subtle information from “messy” observational datasets, and provide reliable inferences for individualizing care decisions. In order to accomplish these goals, our lab (1) identifies domains/disease areas where such approaches can make an impact, (2) identifies gaps where current technologies fail, (3) designs new statistical machine learning techniques that solve associated fundamental computational challenges, and (4) develops and deploys solutions to measure impact.
See my recent article on why I think this topic is so exciting. Also, this (undeservingly) generous article by the ACM’s XRDS Crossroads (the ACM Magazine for Students) highlights some of the work in our lab.
Prior to joining Johns Hopkins, I did my PhD at Stanford with Dr. Daphne Koller. I also spent a year at Harvard University collaborating with Dr. Ken Mandl and Dr. Zak Kohane as an NSF Computing Innovation Fellow. While in the valley, I also spent time as an early employee at Aster Data Systems, a big data startup acquired by Teradata. I am an advisor to Patient Ping. I’m also an advisor on data quality and analysis to CancerLinQ, a learning health system by the American Society of Clinical Oncology. I’m originally from Darjeeling, India. I can be bribed with good tea.
Example press on our lab’s work: NSF Science Nation, Baltimore Sun, IEEE Spectrum, Hopkins Magazine, Science, Hopkins Engineering Magazine, Healhcare IT News, Popular Science, NSF Bits and Bytes, Stanford Medicine, Pittsburgh Post-Gazette on the Frontiers meeting, Talking Machines podcast, Popular Science, and TEDxBoston.
You will likely find my FAQ below useful. Please read before you send a note. Regarding specific areas of study, we’re looking to accept students interested in probabilistic modeling, scalable inference, causal inference and sequential decision making. If you’re interested in a program that allows you to get training in both computer science and statistics, our PhD students have the flexibility to do so.
|POS: PhD, Postdoctoral and Research Scientist Openings: Email me a copy of your CV. We are especially interested in candidates with experience or strong interest in (1) large scale modeling with Bayesian methods, approximate inference, non-parametric methods, and causal inference, or (2) human-in-the-loop decision-making. We welcome candidates from all backgrounds.|
|POS: Interdisciplinary PhD program in Computational Biology. Interested students apply here.|
|POS: In 2013, I founded an interdisciplinary summer program in Computational Sciences, Systems and Engineering. Predoctoral students interested in summer internships, apply here.|
Selected Honors, Awards and Notable Events:
Selected Publications: (ML=Machine Learning, HI=Health Informatics)
[ML] A. Subbaswamy, S. Saria. Counterfactual Normalization: Proactively Addressing Dataset Shift Using Causal Mechanisms. Uncertainty in Artificial Intelligence (UAI), 2018. pdf. NEW
[ML] P. Schulam, S. Saria. Discretizing Logged Interaction Data Biases Learning for Decision-Making. arXiv preprint arXiv:1810.03025, 2018. pdf. NEW
[ML] P. Schulam, S. Saria. Reliable Decision Support Using Counterfactual Models. Neural Information Processing Systems (NIPS), 2017. pdf.
[ML] P. Schulam, S. Saria. Reliable What-If Reasoning with Counterfactual Gaussian Processes. Preprint. pdf. NEW
[ML] H. Soleimani, J. Hensman, S. Saria. Scalable Joint Models for Reliable Uncertainty-Aware Event Prediction. Transactions of Pattern Analysis and Machine Intelligence (in press), 2017. pdf. NEW
[ML] H. Soleimani, A. Subbaswamy, S. Saria. Treatment-Response Models for Counterfactual Reasoning with Continuous-time, Continuous-valued Interventions. Uncertainty in Artificial Intelligence (UAI), 2017. pdf. NEW
[ML] Y. Xu, Y. Xu, S. Saria. A Bayesian Nonparametic Approach for Estimating Individualized Treatment-Response Curves. pdf NEW
[ML] Y. Xu, Y. Xu, S. Saria. A Bayesian Nonparametric Approach for Estimating Individualized Treatment-Response Curves. Conference on Machine Learning and Health Care (MLHC), Aug. 2016. pdf.
[ML] Q. Liu, K. Henry, Y. Xu, S. Saria. Using Causal Inference to Estimate What-if Outcomes for Targeting Treatments. NIPS workshop on “What if” Reasoning, 2016. pdf.
[ML] P. Schulam, S. Saria. Integrative Analysis Using Coupled Latent Variable Models for Individualizing Prognoses. Journal of Machine Learning Research 17 (234):1−35. pdf. NEW
[ML] D. Robinson*, S. Saria*. Trading-Off Cost of Deployment Versus Accuracy in Learning Predictive Models. International Joint Conference of Artificial Intelligence (IJCAI), 2016. pdf
[ML] P. Schulam, S. Saria. A Framework for Individualizing Predictions of Disease Trajectories by Exploiting Multi-resolution Structure. Neural Information Processing Systems (NIPS), 2015. pdf
[ML] K. Dyagilev, S. Saria. Learning (Predictive) Risk Scores in the Presence of Censoring due to Interventions. Machine Learning, March 2016, Volume 102, Issue 3, pp 323-348. Online first: October 2015. pdf, ArXiv
[ML] P. Schulam, F. Wigley, S. Saria. Clustering Longitudinal Clinical Marker Trajectories from Electronic Health Data: Applications to Phenotyping and Endotype Discovery. American Association for Artificial Intelligence, January 2015. pdf
[ML] S. Saria, A. Duchi, D. Koller. Learning Deformable Motifs in Continuous Time Series data. International Joint Conference on Artificial Intelligence (IJCAI), 2011. pdf
[ML] S. Saria, D. Koller, A. Penn. Learning individual and population level traits from clinical temporal data. NIPS Predictive Models in Personalized Medicine, 2010. pdf. (Other versions: short, long)
[ML] S. Saria, U. Nodelman, D. Koller. Reasoning at the Right Time Granularity. Uncertainty in Artificial Intelligence (UAI), July 2007. pdf (Best student paper award)
[ML] V. Jojic, S. Saria, D. Koller. Convex envelopes of complexity controlling penalties: the case against premature envelopment. Artificial Intelligence and Statistics, 2011. pdf
[HI] A. Zhan*, S. Mohan*, C. Tarolli*, R.B. Schneider, J.L. Adams, S. Sharma, M.J. Elson, K.L. Spear, A.M. Glidden, M.A. Little, A. Terzis, E.R. Dorsey, S. Saria. Using Smartphones and Machine Learning to Quantify Parkinson Disease Severity: the Mobile Parkinson Disease Score. JAMA Neurology 2018. Vol. 75, Issue 7, Pages:876-880. pdf NEW
[HI] S. Saria. Individualized sepsis treatment using reinforcement learning. Nature Medicine 2018. Vol. 24. pdf NEW
[HI] K. Henry, D. Hager, P. Pronovost, S. Saria. A Targeted Real-time Early Warning Score (TREWScore) for Septic Shock. Science Translational Medicine 2015. Vol. 7, Issue 299. pdf (Cover article)
[HI] S. Saria, A. Goldenberg. Subtyping: What Is It and Its Role in Precision Medicine. IEEE Intelligent Systems, 2015. Vol. 30, Issue 4. pdf. NEW
[HI] S. Saria, A. Rajani, J. Gould, D. Koller, A. Penn. Integration of Early Physiological Responses Predicts Later Illness Severity in Preterm Infants. Science Translational Medicine September 2010. Vol. 2, Issue 48. Link (Cover article)
[HI] C. Paxton, A. Niculescu-Mizil, S. Saria. Developing Predictive Algorithms Using Electronic Medical Records: Challenges and Pitfalls. American Medical Informatics Association, 2013. pdf
[HI] S Saria, G McElvain, AK Rajani, AA Penn, DL Koller. Combining Structured and Free-text Data for Automatic Coding of Patient Outcomes. American Medical Informatics Association, 2010. (Best student paper finalist )
[HI] A. J Ma, N. Rawat, A. Reiter, C. Shrock, A. Zhan, A. Stone, A. Rabiee, S. Griffin, D. M. Needham, S. Saria. Measuring Patient Mobility in the ICU Using a Novel Noninvasive Sensor. Critical Care Medicine. Vol. 45, Issue 4. Pp. 630-636. Link NEW
[Perspective] S. Saria. A $3 Trillion Challenge to Computational Scientists: Transforming Healthcare Delivery, August 2014. IEEE Intelligent Systems. Vol. 29, Issue 4. Link (Invited article)
[Perspective] D.W. Bates, S. Saria, L. Ohno-Machado, A. Shah, G. Escobar. Big data in health care: using analytics to identify and manage high-risk and high-cost patients, July 2014. Health Affairs. Vol. 33, Issue 7. Link (Short presentation made to an audience of policy makers at the National Press Club, Washington D.C.here.)
National Science Foundation on our work in modeling complex, chronic diseases such as scleroderma. More here.
TEDxBoston talk on Better Medicine Through Machine Learning. More.
NIPS 2016 Tutorial on ML Methods for Personalization with Application to Medicine. More here.
Machine Learning and Healthcare Lab:
Peter Schulam (PhD student; Computer Science) NSF and Centennial Fellow
Peter Schulam wins the Centennial Fellowship (August 2013). Miruna Oprescu wins second prize at the JHU Summer Research Expeditions program for her work with my lab on modeling health data (Aug. 2014). Ethan Pronovost selected as one of the finalists at the Americal Medical Informatics Association HSSP for his work with my lab on measuring harms due to false alarms in the ICU (Oct. 2014). Zach Barnes won second prize at the JHU Summer Research Expeditions for his work on deploying a tool for prognosticating lung fibrosis in scleroderma (Aug 2015).
– I’m the workshop co-chair for NIPS 2017 with Ralf Herbrich and NeurIPS (formerly NIPS) with Joaquin Quinonero Candela.
– Old talks (I no longer keep my webpage up to date with talks):
Current (Spring 15): 600.476/676 Machine Learning: Data to Models
Previous (Fall 13): 600.476/676 Machine Learning in Complex Domains
Previous: 600.476/676 Machine Learning in Complex Domains, 600.775 Seminar in Machine Learning and Data-Intensive Computing
Q00. I am an international student and I want to apply to your PhD program. Are you taking students?
We get 5+ emails of this type per week through the fall. As a once international student, I understand the anxiety of being on the other side. First, these emails are not effective unless you’ve read the faculty’s papers and have something intelligent to say. So, don’t bother wasting your time. Second, let me explain a common issue with PhD admissions for international students. Typically, CS programs tend to fund their PhD students through the length of their program (5 years). This means faculty tend to be risk averse. Having sat on PhD admissions committees, most faculty find it challenging to assess the background of an international student because they don’t often know your school or your advisor. It’s also difficult to gauge whether your grades are highly competitive or not. As a result, most committees pass on international students unless they have an *obviously* strong application. If you’re serious about research and getting a PhD, and don’t have a strong research background (i.e. published papers in top conferences and strong recommendation letters), apply to the masters program. Very often, we take our strong masters students as research assistants after a semester or two. This gives you a chance to build credibility. And, very often, you can recover the cost of your masters through industry internships which pay quite a bit. Also, at a place like Hopkins, there are many faculty outside of computer science that are looking for strong programmers for a research project. That funding can tide you over until you find a lab. But, in the long run, it’s more fruitful to apply to a strong masters program with the goal of switching to a good PhD program rather than going to a PhD program with a poor fit.
No. There are a number of faculty including myself that work on machine learning problems applicable to multiple domains. Look through ML@JHU. Also, look through application areas at Human Language Center of Excellence, and IDIES.
Please take a look at my papers. If you still remain interested, please send me an email. It’s often also helpful to speak with the students in the research group to get a flavor of the problems you could get involved in.
Q2. I’m not at Hopkins currently. Can I apply to your lab for a PhD?
Yes, we are looking for creative and brilliant students to join us. However, you must formally apply to the PhD program for me to
Q3. I’m an undergraduate and I am looking for internship opportunities. Can I visit your lab?
Yes, we started a new internship program called the Summer Research Expeditions (SRE) in 2013. The program brings together faculty from multiple departments in engineering and is a great opportunity to gain exposure to multidisciplinary applications of computing.
Q4. I’m looking for postdoctoral or research scientist positions. Are there positions in your lab?
We are always looking for great people to join our group. There is flexibility in terms of the projects you can get involved with. Please send me a copy of your CV if you’d like to learn more.
Q5. I’m interested in machine learning and your work but I have never worked in medicine/biology/healthcare. Do I need a medical background to work on healthcare projects?
No. In my own work, we’ve made significant progress from bringing in a fresh machine learning perspective to existing problems in healthcare. See my recent article to get a flavor of the kinds of interesting computational problems that machine learning researchers can help solve in healthcare. You can learn most of what you need to know about the domain through your readings and interactions with your collaborators. Our healthcare expenses are upwards of 2.5 trillion dollars and we’re in desperate need of better approaches for improving outcomes and lowering cost. Our health system produces vasts amount of messy and heterogeneous data that we need smarter modelers to be looking at and gleaning insights from.
Q6. Why Hopkins?
If you’re interested in solving difficult computational problems in healthcare, Hopkins is one of the best places to join. We have more than two dozen faculty across Computer Science, Statistics, and Biomedical Engineering who are studying novel ways to improve medicine and healthcare using computational techniques. See Institute for Computational Medicine, Lab for Computational Sensing and Robotics, inHealth, ML@JHU and Institute for Data Intensive Science and Engineering for related work by faculty.