|60%||Project and write-up|
|20%||In class presentations of papers|
|10%||Written summaries/critiques of papers (on blackboard)|
This will be a seminar class in which we read (mostly recent) research papers that explore cutting-edge applications of machine learning (and especially statistical natural language processing) to problems in health, broadly construed. Students will take turns taking the lead presenting a paper each week, which we will discuss together critically during class meetings. The leader/presenter each week will be responsible for presenting the paper(s) to the class. This includes presenting relevant background material; this is important because in general we will have a strong bias toward reading very recent work in the class. Critical discussion will follow, and we will conclude class by talking about potential means of extending the work. Students will also be responsible for writing short summaries of the research papers we read each week.
The class will culminate in projects and accompanying papers. These will be structured as research papers (with accompanying code, when possible). Students will write up papers following the Machine Learning for Healthcare template, available here. Students may work together if they like, but then I expect a concomitant increase in complexity and a clear delineation of contributions, for fairness sake.
Projects may involve replicating a state-of-the-art paper we read, or, ideally, extending one of these or developing new ideas in the area. Some class time will be reserved for students to present their project ideas, results and findings (these will be structured as research talks). Therefore, this class will very much be research-oriented.
Students will rotate responsibility for presenting (one of the) assigned papers and leading the ensuing discussion. The presentation should provide necessary background, the core contribution of the paper, (perceived) strengths and weaknesses, and ideas for improving the work.
Additionally, prior to every class (for which there are papers listed), all students are to write brief summaries ("reviews") of one of the assigned papers for that class (their choice). These should include a concise summary of the work and enumerate at least two strong points and at least two weak points therein. Assessments are to be posted on the blackboard discussion forum. Note that is assumed students read all assigned papers, but may read only one in-depth. Furthermore, these critiques will note ways in which the work could potentially be extended/improved (which may be coupled with the weak points). The idea is basically to ensure robust discussion during class period.
In short: there are no official pre-reqs for this class, but I do have expectations (delineated below). Please do not hesitate to contact me if you have any questions; I'm happy to chat.
I expect you have some background in machine learning. Background in (statistical) NLP specifically may be helpful, but not strictly required. Students should have an interest in conducting (or learning how to conduct) research. No background in health sciences necessary! Indeed, I hope to expose students who have been exposed to machine learning to the possibilities of applying such work to problems in health.
Given the nature of the class (with discussion playing a key role), attendance and active participation are very important. Accordingly, your participation in class will account for 10% of your grade.
|09/07 W||Introductions, aims, expectations, logistics, etc.|
Visitor from the school of nursing: Tracy Magee will drop in to talk about a potential project identifying colicky babies
Inducing representations of medical concepts
Multi-layer Representation Learning for Medical Concepts. E Choi et al., 2016
Learning Low-Dimensional Representations of Medical Concepts. Y Choi et al., 2016
|Byron (E Choe et al.) |
Machine learning, social media, and public health|
Combining Search, Social Media, and Traditional Data Sources to Improve Influenza Surveillance, Santillana et al., 2015
Collective Supervision of Topic Models for Predicting Surveys with Social Media, Benton, Paul, Hancock and Dredze, 2016.
|Edward (Y Choi et al.) (from last class)|
Silvio (Benton et al.)
|09/19 M||Computational phenotyping
Comorbidity Clusters in Autism Spectrum Disorders: An Electronic Health Record Time-Series Analysis, Doshi-Velez, Ge and Kohane, 2013.
Learning probabilistic phenotypes from heterogeneous EHR data, Pivovarov et al., 2015.
|Wen (Doshi-Velez et al.)|
Yuan (Pivovarov et al.)
PlateClick: Bootstrapping Food Preferences Through an Adaptive Visual Interface, Yang et al., 2015
Yum-me: Personalized Healthy Meal Recommender System, Yang et al., 2016.
|Byron (Yang et al.)|
Srikanth (Yang et al.)
|09/26 M||Question answering I. (recent work on biomedical QA).
Special visitor Clark Freifeld: Clark Freifeld will come tell us about his work on HealthMap
Biomedical question answering using semantic relations, Hristovski et al., 2015.
An overview of the BIOASQ large-scale biomedical semantic indexing and question answering competition, Tsatsaronis et al., 2015.
cf. Answering factoid questions in the biomedical domain, Weissenborn et al, 2013.
|Edward (Hristovski et al.)|
Byron (Tsatsaronis et al.)
|09/28 W||Question answering II. (newer methods)|
A Neural Network for Factoid Question Answering over Paragraphs, Iyyer et al., 2014.
Ask Me Anything: Dynamic Memory Networks for Natural Language Processing, Kumar et al., 2015.
|Wen (Iyyer et al.)|
Ulrich (Kumar et al.)
|10/03 M||No class meeting; potential project ideas due.|
|10/05 W||Guest speaker Marzyeh Ghassemi, MIT|
|10/10 M||Columbus day; no class.|
|10/12 W||Depression and social media|
Predicting Depression via Social Media, Choudhury, Gamon, Counts and Horvitz, 2013.
Discovering Shifts to Suicidal Ideation from Mental Health Content in Social Media Choudhury et al., 2016.
|Byron (Choudhury et al. 1)|
Silvio (Choudhury et al. 2)
Intelligible Models for HealthCare: Predicting Pneumonia Risk and Hospital 30-day Readmission, Caruana et al., 2015.
Mind the Gap: A Generative Approach to Interpretable Feature Selection and Extraction, Kim, Shah and Doshi-Velez, 2015.
cf. Generalized additive models, Hastie and Tribshirani, 1986
|Srikanth (Caruana et al.)|
Edward (Kim et al.)
|10/19 W||Interpretability II|
Rationalizing Neural Predictions, Lei, Barzilay and Jaakkola, 2016.
The Mythos of Model Interpretability, Lipton, 2016.
cf. Interpretable Machine Learning: Lessons from Topic Modeling, Paul, 2016.
|Yuan (Lei et al.)|
|10/24 M||Predicting clinical events from EMRs I|
Developing Predictive Models Using Electronic Medical Records: Challenges and Pitfalls, Paxton, Niculescu-Mizil and Saria, 2013
Doctor AI: Predicting Clinical Events via Recurrent Neural Networks. Choi et al., 2016.
|Ulrich (Paxton et al.) |
Wen (Choi et al.)
|10/26 W||Predicting clinical events from EMRs II|
Clinical Tagging with Joint Probabilistic Models Halpern, Horn and Sontag, 2016
Electronic Medical Record Phenotyping Using the Anchor and Learn framework, Halpern, Horn, Choi and Sontag, 2016
|Yuan (Halpern et al. 1)|
Byron (Halpern et al. 2)
Building Bridges Across Electronic Health Record Systems Through Inferred Phenotypic Topics Chen et al., 2015
Normalising Medical Concepts in Social Media Texts by Learning Semantic Representation, Limsopatham and Collier, 2016.
|Wen (Chen et al.)|
Edward (Limsopatham and Collier)
|11/02 W||Guest speaker: Hadi Amiri.|
Biomedical information extraction|
HARVEST, A Longitudinal Patient Record Summarizer, Hirsch et al., 2016 (swapped from 11/14 previously)
Learning for Biomedical Information Extraction: Methodological Review of Recent Advances, Liu et al., 2016
|Byron (Liu et al.)|
Yuan (Hirsch et al.)
Learning to Diagnoise with LSTM Recurrent Neural Networks, Lipton et al., 2016
Deep Survival Analysis, Ranganath, Perotte, Elhadad and Blei, 2016.
| Edward (Lipton et al.) |
Silvio (Ranganath et al.)
|11/14 M||In class preliminary project presentations (on data, aims and mehods) + discussions|
Weakly Supervised Learning of Biomedical Information Extraction from Curated Data, Jain et al., 2016. (swapped from 11/07 previously)
A Multivariate Timeseries Modeling Approach to Severity of Illness Assessment and Forecasting in ICU with Sparse, Heterogeneous Clinical Data
|Sri (Ghassemi et al.)|
Srikanth (Jain et al.)
Hierarchical Span-Based Conditional Random Fields for Labeling and Segmenting Events in Wearable Sensor Data Streams, Adams et al., 2016.
Can Deep Learning Revolutionize Mobile Sensing?, Lane and Georgiev, 2015.
|Yuan (Adams et al.)|
Uli (Lane and Georgiev)
|11/23 W||Thanksgiving break; no class.|
Temporal Convolutional Neural Networks for Diagnosis from Lab Tests, Razavian and Sontag, 2016.
Deep Learning for Identifying Metastatic Breast Cancer, Wang et al., 2016.
cf. Machine Learning for Medical Diagnosis: History, State of the Art and Perspective, Kononenko, 2001.
|Srikanth (Razavian and Sontag)|
|11/30 W||Dedicated project and presentation time; no class meeting.|
A Framework for Individualizing Predictions of Disease Trajectories by Exploiting Multi-Resolution Structure, Schulam and Saria, 2015.
Unsupervised Learning of Disease Progression Models, Wang, Sontag and Wang, 2014.
|Srikanth (Schulam and Saria)|
Byron (Wang et al.)
|12/07 W||Project presentations!|