Explaining an increase in predicted risk for clinical alerts

Alvin Rajkomar; Andrew Dai; Claire Cui; Gerardo Flores; Greg Corrado; Michaela Hardt; Michael Howell; Moritz Hardt

arxiv: 1907.04911 · v1 · pith:CKUOPMV5new · submitted 2019-07-10 · 💻 cs.LG · cs.CY· stat.AP· stat.ML

Explaining an increase in predicted risk for clinical alerts

Michaela Hardt , Alvin Rajkomar , Gerardo Flores , Andrew Dai , Michael Howell , Greg Corrado , Claire Cui , Moritz Hardt This is my paper

Pith reviewed 2026-05-24 23:30 UTC · model grok-4.3

classification 💻 cs.LG cs.CYstat.APstat.ML

keywords explainable AIdynamical modelsrisk attributionclinical alertstemporal explanationsmachine learninghealthcare AIsequential data

0 comments

The pith

Methods lift static attribution techniques to explain risk increases in dynamical models by attributing them to specific past inputs.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper addresses how to explain why a stateful model's predicted risk has risen at a given time step, focusing on sequences of inputs rather than single static cases. It develops techniques that extend existing attribution methods to handle the dynamics of time-series predictions while addressing challenges unique to sequential data. In a clinical case study, these explanations aim to help clinicians quickly identify which past events drove an alert without reviewing the full patient record. The work includes expert evaluations to test whether the resulting attributions support effective triage decisions. If successful, the approach provides a way to make temporal risk models more interpretable in practice.

Core claim

We develop methods to lift static attribution techniques to the dynamical setting, where we identify and address challenges specific to dynamics. When the estimated risk increases, the goal of the explanation is to attribute the increase to a few relevant inputs from the past, enabling concise triage of clinical alerts.

What carries the argument

Lifting static attribution techniques to the dynamical setting to attribute risk increases to a few past inputs while addressing statefulness and sequential challenges.

If this is right

Clinicians receive concise explanations that highlight the inputs responsible for a risk increase.
The same lifted attribution methods apply to any dynamical model producing sequential risk estimates.
Challenges unique to the dynamical setting, such as handling state across time steps, are explicitly identified and mitigated.
Expert evaluation provides direct feedback on whether the explanations support real triage decisions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar attribution lifting could apply to non-clinical domains with sequential risk models, such as equipment failure prediction.
If attributions prove reliable, they might reduce the volume of alerts that require full manual review.
Future work could test whether these explanations improve actual clinical outcomes beyond expert ratings of utility.

Load-bearing premise

Concise attribution of risk increases to a few past inputs will enable clinicians to effectively triage alerts without needing the full patient history or additional context.

What would settle it

Expert clinicians reviewing the attributions find them unhelpful or misleading for deciding whether to intervene on alerts.

Figures

Figures reproduced from arXiv: 1907.04911 by Alvin Rajkomar, Andrew Dai, Claire Cui, Gerardo Flores, Greg Corrado, Michaela Hardt, Michael Howell, Moritz Hardt.

**Figure 2.** Figure 2: A patient’s risk time series predicted from RNN models on the left [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

read the original abstract

Much work aims to explain a model's prediction on a static input. We consider explanations in a temporal setting where a stateful dynamical model produces a sequence of risk estimates given an input at each time step. When the estimated risk increases, the goal of the explanation is to attribute the increase to a few relevant inputs from the past. While our formal setup and techniques are general, we carry out an in-depth case study in a clinical setting. The goal here is to alert a clinician when a patient's risk of deterioration rises. The clinician then has to decide whether to intervene and adjust the treatment. Given a potentially long sequence of new events since she last saw the patient, a concise explanation helps her to quickly triage the alert. We develop methods to lift static attribution techniques to the dynamical setting, where we identify and address challenges specific to dynamics. We then experimentally assess the utility of different explanations of clinical alerts through expert evaluation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper adapts static attribution methods to dynamical clinical risk models with a hospital case study, but the practical claim that concise explanations substitute for full patient history review rests on an assumption rather than direct evidence.

read the letter

The new element is lifting attribution techniques to a stateful, time-series setting where a model outputs a sequence of risk scores and the task is to explain why risk rose at a given step. They identify dynamics-specific issues like handling the accumulating state and then apply the adapted methods to clinical deterioration alerts. The case study is grounded in a real hospital workflow where clinicians get alerts and need quick triage help. Expert evaluation of the resulting explanations is a reasonable way to check usefulness in that domain. The work is honest about focusing on one practical setting rather than claiming a universal new framework. The main soft spot is the leap from technical attributions to clinical benefit. The abstract and setup assume that pointing to a few past inputs will let a clinician decide on intervention without needing the full recent history or extra context. Expert ratings can confirm that the explanations look reasonable, but they do not test whether those ratings translate into triage decisions that match what would happen with the complete timeline. If clinicians still pull up the full record in practice, the dynamical lift does not deliver the stated reduction in review burden. The math and citation handling look standard for this line of work; nothing stands out as circular or invented. This is worth a referee for groups doing explainable AI in medicine or sequential models. A serious editor should send it out because the problem is concrete and the methods are reproducible enough to evaluate, even if the clinical substitution claim needs tighter evidence in revision.

Referee Report

1 major / 1 minor

Summary. The paper develops methods to lift static attribution techniques to dynamical, stateful models that output sequences of risk estimates, with the goal of attributing increases in predicted risk to a small number of past inputs. It presents a clinical case study on alerting for patient deterioration, where concise explanations are intended to help clinicians triage alerts without reviewing the full history, and evaluates the resulting explanations via expert assessment.

Significance. If the lifted methods and evaluation hold, the work would offer a targeted approach to explaining temporal risk changes in clinical ML systems, addressing a practical need for concise, actionable alerts. The emphasis on dynamics-specific challenges and human-centered expert evaluation distinguishes it from purely static attribution literature.

major comments (1)

[Abstract / experimental evaluation] Abstract and experimental evaluation section: the claim that concise attributions enable clinicians to triage alerts without needing the full patient history is central to the stated clinical utility, yet the expert evaluation only assesses perceived utility of the explanations and does not test whether they can substitute for full-history review in triage decisions.

minor comments (1)

[Abstract] The abstract supplies no equations, validation details, or quantitative outcomes, making it difficult to assess technical soundness from the summary alone.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their detailed review and constructive comments. We address the major comment below.

read point-by-point responses

Referee: [Abstract / experimental evaluation] Abstract and experimental evaluation section: the claim that concise attributions enable clinicians to triage alerts without needing the full patient history is central to the stated clinical utility, yet the expert evaluation only assesses perceived utility of the explanations and does not test whether they can substitute for full-history review in triage decisions.

Authors: We agree that the expert evaluation focuses on perceived utility rather than directly testing whether the explanations can substitute for full patient history review in actual triage decisions. This represents a limitation in the strength of evidence for the clinical utility claim. In the revised manuscript, we will clarify this distinction in the abstract and experimental evaluation section, and add a discussion of this limitation along with suggestions for future work that could include controlled studies of triage performance. revision: yes

Circularity Check

0 steps flagged

No circularity: methods and evaluation are independent of inputs

full rationale

The paper describes lifting existing static attribution techniques to a dynamical setting and evaluating the resulting explanations via expert review on clinical alerts. No equations, fitted parameters, or self-citations are shown to reduce any claimed result or prediction to the inputs by construction. The core steps—identifying dynamical challenges and performing expert utility assessment—are methodological extensions and empirical checks that stand apart from the data or prior fits. This is the normal case of a self-contained applied-methods paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no free parameters, axioms, or invented entities are described.

pith-pipeline@v0.9.0 · 5708 in / 889 out tokens · 19892 ms · 2026-05-24T23:30:15.423803+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages · 3 internal anchors

[1]

Finding patients before they crash: the next major opportunity to improve patient safety

David W Bates and Eyal Zimlichman. Finding patients before they crash: the next major opportunity to improve patient safety. BMJ Qual. Saf. , 24(1):1–3, January 2015

work page 2015
[2]

Big data in health care: using analytics to identify and manage high-risk and high-cost patients

David W Bates, Suchi Saria, Lucila Ohno-Machado, Anand Shah, and Gabriel Escobar. Big data in health care: using analytics to identify and manage high-risk and high-cost patients. Health Aﬀ., 33(7):1123–1131, July 2014

work page 2014
[3]

Kim, Benjamin J

Vincent Liu, Yan S. Kim, Benjamin J. Turk, Arona Ragins, Brian A. Dummett, Carnen L. Adams, Elizabeth A. Scruth, and Patricia Kipnis. Early Detection of Impending Deterioration Outside The ICU: A Diﬀerence- in-Diﬀerences (DiD) Study, pages A7614–A7614. 2016

work page 2016
[4]

Insights into the problem of alarm fatigue with physio- logic monitor devices: a comprehensive observational study of consecutive intensive care unit patients

Barbara J Drew, Patricia Harris, Jessica K Z` egre-Hemsey, Tina Mammone, Daniel Schindler, Rebeca Salas-Boni, Yong Bai, Adelita Tinoco, Quan Ding, and Xiao Hu. Insights into the problem of alarm fatigue with physio- logic monitor devices: a comprehensive observational study of consecutive intensive care unit patients. PLoS One, 9(10):e110274, October 2014

work page 2014
[5]

Prognostic accuracy of the quick sequential organ failure assessment for mortality in patients with suspected infection: A systematic review and meta-analysis

Shannon M Fernando, Alexandre Tran, Monica Taljaard, Wei Cheng, Bram Rochwerg, Andrew J E Seely, and Jeﬀrey J Perry. Prognostic accuracy of the quick sequential organ failure assessment for mortality in patients with suspected infection: A systematic review and meta-analysis. Ann. Intern. Med., February 2018

work page 2018
[6]

How to explain individual classiﬁcation decisions

David Baehrens, Timon Schroeter, Stefan Harmeling, Motoaki Kawan- abe, Katja Hansen, and Klaus-Robert M¨ uller. How to explain individual classiﬁcation decisions. J. Mach. Learn. Res. , 11:1803–1831, August 2010

work page 2010
[7]

Visualizing higher-layer features of a deep network

Dumitru Erhan, Yoshua Bengio, Aaron Courville, and Pascal Vincent. Visualizing higher-layer features of a deep network. Technical Report 1341, University of Montreal, June 2009. Also presented at the ICML 2009 Workshop on Learning Feature Hierarchies, Montr´ eal, Canada

work page 2009
[8]

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image classiﬁcation models and saliency maps. CoRR, abs/1312.6034, 2013

work page internal anchor Pith review Pith/arXiv arXiv 2013
[9]

Axiomatic attribution for deep networks

Mukund Sundararajan, Ankur Taly, and Qiqi Yan. Axiomatic attribution for deep networks. In ICML, 2017

work page 2017
[10]

Neural machine translation by jointly learning to align and translate

Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. September 2014

work page 2014
[11]

Intelligible models for healthcare: Predicting pneumonia 13 risk and hospital 30-day readmission

Rich Caruana, Yin Lou, Johannes Gehrke, Paul Koch, Marc Sturm, and Noemie Elhadad. Intelligible models for healthcare: Predicting pneumonia 13 risk and hospital 30-day readmission. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Min- ing, KDD ’15, pages 1721–1730, New York, NY, USA, 2015. ACM

work page 2015
[12]

Development and validation of a continuous measure of patient condition using the electronic medical record

Michael J Rothman, Steven I Rothman, and Joseph Beals, 4th. Development and validation of a continuous measure of patient condition using the electronic medical record. J. Biomed. Inform., 46(5):837–848, October 2013

work page 2013
[13]

Towards a rigorous science of inter- pretable machine learning

Finale Doshi-Velez and Been Kim. Towards a rigorous science of inter- pretable machine learning. arXiv, 2017

work page 2017
[14]

In the wake of the 2003 and 2011 duty hours regulations, how do internal medicine interns spend their time? J

Lauren Block, Robert Habicht, Albert W Wu, Sanjay V Desai, Kevin Wang, Kathryn Novello Silva, Timothy Niessen, Nora Oliver, and Leonard Feldman. In the wake of the 2003 and 2011 duty hours regulations, how do internal medicine interns spend their time? J. Gen. Intern. Med. , 28(8):1042–1047, August 2013

work page 2003
[15]

How do residents spend their shift time? a time and motion study with a particular focus on the use of computers

Lena Mamykina, David K Vawdrey, and George Hripcsak. How do residents spend their shift time? a time and motion study with a particular focus on the use of computers. Acad. Med., 91(6):827–832, June 2016

work page 2016
[16]

Recurrent Neural Networks for Multivariate Time Series with Missing Values

Zhengping Che, Sanjay Purushotham, Kyunghyun Cho, David Sontag, and Yan Liu. Recurrent neural networks for multivariate time series with missing values. CoRR, abs/1606.01865, 2016

work page internal anchor Pith review Pith/arXiv arXiv 2016
[17]

Lipton, David C

Zachary C. Lipton, David C. Kale, and Randall C. Wetzel. Directly modeling missing data in sequences with rnns: Improved classiﬁcation of clinical time series. In MLHC, 2016

work page 2016
[18]

Clinical intervention prediction and understanding using deep networks

Harini Suresh, Nathan Hunt, Alistair Johnson, Leo Anthony Celi, Peter Szolovits, and Marzyeh Ghassemi. Clinical intervention prediction and understanding using deep networks. May 2017

work page 2017
[19]

Biases in electronic health record data due to processes within the healthcare system: retro- spective observational study

Denis Agniel, Isaac S Kohane, and Griﬃn M Weber. Biases in electronic health record data due to processes within the healthcare system: retro- spective observational study. BMJ, 361:k1479, April 2018

work page 2018
[20]

Possible sources of bias in primary care electronic health record data use and reuse

Robert A Verheij, Vasa Curcin, Brendan C Delaney, and Mark M McGilchrist. Possible sources of bias in primary care electronic health record data use and reuse. J Med Internet Res , 20(5):e185, May 2018

work page 2018
[21]

Sampling rate causes bias in apache ii and saps ii scores

Matti Suistomaa, Aarno Kari, Esko Ruokonen, and Jukka Takala. Sampling rate causes bias in apache ii and saps ii scores. Intensive Care Medicine, 26(12):1773–1778, Dec 2000

work page 2000
[22]

Rao, Troy D

Supriyo Chakraborty, Richard Tomsett, Ramya Raghavendra, Daniel Har- borne, Moustafa Alzantot, Federico Cerutti, Mani Srivastava, Alun Preece, Julier Simon, Raghuveer M. Rao, Troy D. Kelley, David Braines, Murat Sensoy, Christopher J. Willis, and Prudhvi Gurram. Interpretability of deep learning models: a survey of results. In IEEE Smart World Congress 20...

work page 2017
[23]

Smilkov, N

D. Smilkov, N. Thorat, B. Kim, F. Vi´ egas, and M. Wattenberg. SmoothGrad: removing noise by adding noise. ICML workshop on visualization for deep learning, June 2017

work page 2017
[24]

Zeiler and Rob Fergus

Matthew D. Zeiler and Rob Fergus. Visualizing and understanding convolu- tional networks. In David Fleet, Tomas Pajdla, Bernt Schiele, and Tinne Tuytelaars, editors, Computer Vision – ECCV 2014 , pages 818–833, Cham,

work page 2014
[25]

Springer International Publishing

work page
[26]

Striving for Simplicity: The All Convolutional Net

Jost Tobias Springenberg, Alexey Dosovitskiy, Thomas Brox, and Martin A. Riedmiller. Striving for simplicity: The all convolutional net. CoRR, abs/1412.6806, 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014
[27]

Pramod Kaushik Mudrakarta, Ankur Taly, Mukund Sundararajan, and Kedar Dhamdhere. Did the model understand the question? In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, ACL 2018, Melbourne, Australia, July 15-20, 2018, Volume 1: Long Papers , pages 1896–1906, 2018

work page 2018
[28]

”why should i trust you?”: Explaining the predictions of any classiﬁer

Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. ”why should i trust you?”: Explaining the predictions of any classiﬁer. In Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD ’16, pages 1135–1144, New York, NY, USA, 2016. ACM

work page 2016
[29]

A uniﬁed approach to interpreting model predictions

Scott M Lundberg and Su-In Lee. A uniﬁed approach to interpreting model predictions. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors, Advances in Neural Information Processing Systems 30, pages 4765–4774. Curran Associates, Inc., 2017

work page 2017
[30]

Learning to explain: An information-theoretic perspective on model interpretation

Jianbo Chen, Le Song, Martin Wainwright, and Michael Jordan. Learning to explain: An information-theoretic perspective on model interpretation. In Jennifer Dy and Andreas Krause, editors, Proceedings of the 35th In- ternational Conference on Machine Learning , volume 80 of Proceedings of Machine Learning Research, pages 883–892, Stockholmsmssan, Stockholm...

work page 2018
[31]

Infectious diseases society of America/American thoracic society consensus guidelines on the management of community-acquired pneumonia in adults

Lionel A Mandell, Richard G Wunderink, Antonio Anzueto, John G Bartlett, G Douglas Campbell, Nathan C Dean, Scott F Dowell, Thomas M File, Jr, Daniel M Musher, Michael S Niederman, Antonio Torres, Cynthia G Whitney, Infectious Diseases Society of America, and American Thoracic Society. Infectious diseases society of America/American thoracic society conse...

work page 2007
[32]

British thoracic society community acquired pneumonia guideline and the NICE pneumonia guideline: how they ﬁt together

W S Lim, D L Smith, M P Wise, and S A Welham. British thoracic society community acquired pneumonia guideline and the NICE pneumonia guideline: how they ﬁt together. BMJ Open Respir Res , 2(1):e000091, May 2015

work page 2015
[33]

Moving beyond Single-Parameter early warning scores for rapid response system activation

Matthew M Churpek and Dana P Edelson. Moving beyond Single-Parameter early warning scores for rapid response system activation. Crit. Care Med., 44(12):2283–2285, December 2016

work page 2016
[34]

Sustained eﬀectiveness of a primary-team-based rapid response system

Michael D Howell, Long Ngo, Patricia Folcarelli, Julius Yang, Lawrence Mottley, Edward R Marcantonio, Kenneth E Sands, Donald Moorman, and Mark D Aronson. Sustained eﬀectiveness of a primary-team-based rapid response system. Crit. Care Med., 40(9):2562–2568, September 2012

work page 2012
[35]

Electronic health record adoption in US hospitals: Progress continues, but challenges persist

Julia Adler-Milstein, Catherine M DesRoches, Peter Kralovec, Gregory Foster, Chantal Worzala, Dustin Charles, Talisha Searcy, and Ashish K Jha. Electronic health record adoption in US hospitals: Progress continues, but challenges persist. Health Aﬀ., 34(12):2174–2180, December 2015

work page 2015
[36]

Deep ehr: A survey of recent advances in deep learning techniques for electronic health record (ehr) analysis

Benjamin Shickel, Patrick Tighe, Azra Bihorac, and Parisa Rashidi. Deep ehr: A survey of recent advances in deep learning techniques for electronic health record (ehr) analysis. IEEE Journal of Biomedical and Health Informatics, PP(99):1–1, 2017

work page 2017
[37]

Clinical intervention prediction and understanding with deep neural networks

Harini Suresh, Nathan Hunt, Alistair Johnson, Leo Anthony Celi, Peter Szolovits, and Marzyeh Ghassemi. Clinical intervention prediction and understanding with deep neural networks. In Finale Doshi-Velez, Jim Fackler, David Kale, Rajesh Ranganath, Byron Wallace, and Jenna Wiens, editors, Proceedings of the 2nd Machine Learning for Healthcare Conference, vo...

work page 2017
[38]

Deepr: A convolutional net for medical records

Phuoc Nguyen, Truyen Tran, Nilmini Wickramasinghe, and Svetha Venkatesh. Deepr: A convolutional net for medical records. July 2016

work page 2016
[39]

Doctor AI: Predicting clinical events via recurrent neural networks

Edward Choi, Mohammad Taha Bahadori, Andy Schuetz, Walter F Stew- art, and Jimeng Sun. Doctor AI: Predicting clinical events via recurrent neural networks. In Proceedings of the 1st Machine Learning for Healthcare Conference, pages 301–318. jmlr.org, 2016

work page 2016
[40]

Deep computational phenotyping

Zhengping Che, David Kale, Wenzhe Li, Mohammad Taha Bahadori, and Yan Liu. Deep computational phenotyping. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’15, pages 507–516, New York, NY, USA, 2015. ACM

work page 2015
[41]

Lasko, Joshua C

Thomas A. Lasko, Joshua C. Denny, and Mia A. Levy. Computational phenotype discovery using unsupervised feature learning over noisy, sparse, and irregular clinical data. PLoS One, 8(6), 2013. 16

work page 2013
[42]

Learn- ing vector representation of medical objects via emr-driven nonnegative restricted boltzmann machines (enrbm)

Truyen Tran, Tu Dinh Nguyen, Dinh Phung, and Svetha Venkatesh. Learn- ing vector representation of medical objects via emr-driven nonnegative restricted boltzmann machines (enrbm). J. of Biomedical Informatics , 54(C):96–105, April 2015

work page 2015
[43]

Comorbidity clusters in autism spectrum disorders: An electronic health record time-series analysis

Finale Doshi-Velez, Yaorong Ge, and Isaac Kohane. Comorbidity clusters in autism spectrum disorders: An electronic health record time-series analysis. Pediatrics, 133(1):e54–e63, 2014

work page 2014
[44]

Mathieu Guillame-Bert, Artur Dubrawski, Donghan Wang, Marilyn Hrav- nak, Gilles Clermont, and Michael R. Pinsky. Learning temporal rules to forecast instability in continuously monitored patients. JAMIA, 24(1):47–53, 2017

work page 2017
[45]

Hughes, Huseyin Melih Elibol

Thomas McCoy Roy Perlis Finale Doshi-Velez Michael C. Hughes, Huseyin Melih Elibol. Supervised topic models for clinical interpretability. In Proceedings of the 1st Machine Learning for Healthcare Conference , 2016

work page 2016
[46]

Improving palliative care with deep learning

Anand Avati, Kenneth Jung, Stephanie Harman, Lance Downing, Andrew Ng, and Nigam H Shah. Improving palliative care with deep learning. November 2017

work page 2017
[47]

Ying Sha and May D. Wang. Interpretable predictions of clinical outcomes with an attention-based recurrent neural network. In Proceedings of the 8th ACM International Conference on Bioinformatics, Computational Biol- ogy,and Health Informatics, ACM-BCB ’17, pages 233–240, New York, NY, USA, 2017. ACM

work page 2017
[48]

Retain: An interpretable predictive model for healthcare using reverse time attention mechanism

Edward Choi, Mohammad Taha Bahadori, Jimeng Sun, Joshua Kulas, Andy Schuetz, and Walter Stewart. Retain: An interpretable predictive model for healthcare using reverse time attention mechanism. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Systems 29 , pages 3504–3512. Curran Associa...

work page 2016
[49]

Interpretable deep models for icu outcome prediction

Zhengping Che, Sanjay Purushotham, Robinder Khemani, and Yan Liu. Interpretable deep models for icu outcome prediction. 2016:371–380, 02 2017

work page 2016
[50]

Khemani, and Yan Liu

Zhengping Che, Sanjay Purushotham, Robinder G. Khemani, and Yan Liu. Distilling knowledge from deep networks with applications to healthcare domain. In NIPS Workshop on Machine Learning for Healthcare , 2015

work page 2015
[51]

Diﬀerentiation of discrete multidimensional signals

Hany Farid and Eero Simoncelli. Diﬀerentiation of discrete multidimensional signals. IEEE Transactions on Image Processing , 13(4):496–508, 4 2004

work page 2004
[52]

Farid and E

H. Farid and E. P. Simoncelli. Diﬀerentiation of discrete multidimensional signals. IEEE Transactions on Image Processing, 13(4):496–508, April 2004. 17

work page 2004
[53]

Long Short-Term memory.Neural Comput., 9(8):1735–1780, November 1997

Sepp Hochreiter and J¨ urgen Schmidhuber. Long Short-Term memory.Neural Comput., 9(8):1735–1780, November 1997

work page 1997
[54]

Johnson, Tom J

Alistair E.W. Johnson, Tom J. Pollard, Lu Shen, Li-wei H. Lehman, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and Roger G. Mark. Mimic-iii, a freely accessible critical care database. Scientiﬁc Data, 2016

work page 2016
[55]

Michael Felker, Larry A

G. Michael Felker, Larry A. Allen, Stuart J. Pocock, Linda K. Shaw, John J.V. McMurray, Marc A. Pfeﬀer, Karl Swedberg, Duolao Wang, Salim Yusuf, Eric L. Michelson, and Christopher B. Granger. Red cell distribution width as a novel prognostic marker in heart failure: Data from the charm program and the duke databank. Journal of the American College of Card...

work page 2007
[56]

Scalable and accurate deep learning for electronic health records

Alvin Rajkomar, Eyal Oren, Kai Chen, Andrew M Dai, Nissan Hajaj, Pe- ter J Liu, Xiaobing Liu, Mimi Sun, Patrik Sundberg, Hector Yee, Kun Zhang, Gavin E Duggan, Gerardo Flores, Michaela Hardt, Jamie Irvine, Quoc Le, Kurt Litsch, Jake Marcus, Alexander Mossin, Justin Tansuwan, De Wang, James Wexler, Jimbo Wilson, Dana Ludwig, Samuel L Volchen- boum, Katheri...

work page 2018
[57]

A theoretically grounded application of dropout in recurrent neural networks

Yarin Gal and Zoubin Ghahramani. A theoretically grounded application of dropout in recurrent neural networks. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Systems 29, pages 1019–1027. Curran Associates, Inc., 2016

work page 2016
[58]

Adaptive subgradient methods for online learning and stochastic optimization

John Duchi, Elad Hazan, and Yoram Singer. Adaptive subgradient methods for online learning and stochastic optimization. J. Mach. Learn. Res. , 12:2121–2159, July 2011

work page 2011
[59]

J. A. Kellum and N. Lameire. Diagnosis, evaluation, and management of acute kidney injury: a KDIGO summary (Part 1). Crit Care, 17(1):204, Feb 2013

work page 2013
[60]

Kindermans, S

P.-J. Kindermans, S. Hooker, J. Adebayo, M. Alber, K. T. Sch¨ utt, S. D¨ ahne, D. Erhan, and B. Kim. The (Un)reliability of saliency methods. NIPS workshop on Explaining and Visualizing Deep Learning , 2017

work page 2017
[61]

The mythos of model interpretability

Zachary Chase Lipton. The mythos of model interpretability. In ICML Workshop on Human Interpretability , 2016

work page 2016
[62]

Extremely Likely

Aaron Springer, Victoria Hollis, and Steve Whittaker. Dice in the black box: User experiences with an inscrutable algorithm, 2017. 18 8 Appendix 8.1 Experiments - Features For the experiments we restricted the lab tests to the following list of roughly 40 target harmonized features: blood pressure, pulse, respiratory rate, oxygen saturation, blood pressur...

work page 2017

[1] [1]

Finding patients before they crash: the next major opportunity to improve patient safety

David W Bates and Eyal Zimlichman. Finding patients before they crash: the next major opportunity to improve patient safety. BMJ Qual. Saf. , 24(1):1–3, January 2015

work page 2015

[2] [2]

Big data in health care: using analytics to identify and manage high-risk and high-cost patients

David W Bates, Suchi Saria, Lucila Ohno-Machado, Anand Shah, and Gabriel Escobar. Big data in health care: using analytics to identify and manage high-risk and high-cost patients. Health Aﬀ., 33(7):1123–1131, July 2014

work page 2014

[3] [3]

Kim, Benjamin J

Vincent Liu, Yan S. Kim, Benjamin J. Turk, Arona Ragins, Brian A. Dummett, Carnen L. Adams, Elizabeth A. Scruth, and Patricia Kipnis. Early Detection of Impending Deterioration Outside The ICU: A Diﬀerence- in-Diﬀerences (DiD) Study, pages A7614–A7614. 2016

work page 2016

[4] [4]

Insights into the problem of alarm fatigue with physio- logic monitor devices: a comprehensive observational study of consecutive intensive care unit patients

Barbara J Drew, Patricia Harris, Jessica K Z` egre-Hemsey, Tina Mammone, Daniel Schindler, Rebeca Salas-Boni, Yong Bai, Adelita Tinoco, Quan Ding, and Xiao Hu. Insights into the problem of alarm fatigue with physio- logic monitor devices: a comprehensive observational study of consecutive intensive care unit patients. PLoS One, 9(10):e110274, October 2014

work page 2014

[5] [5]

Prognostic accuracy of the quick sequential organ failure assessment for mortality in patients with suspected infection: A systematic review and meta-analysis

Shannon M Fernando, Alexandre Tran, Monica Taljaard, Wei Cheng, Bram Rochwerg, Andrew J E Seely, and Jeﬀrey J Perry. Prognostic accuracy of the quick sequential organ failure assessment for mortality in patients with suspected infection: A systematic review and meta-analysis. Ann. Intern. Med., February 2018

work page 2018

[6] [6]

How to explain individual classiﬁcation decisions

David Baehrens, Timon Schroeter, Stefan Harmeling, Motoaki Kawan- abe, Katja Hansen, and Klaus-Robert M¨ uller. How to explain individual classiﬁcation decisions. J. Mach. Learn. Res. , 11:1803–1831, August 2010

work page 2010

[7] [7]

Visualizing higher-layer features of a deep network

Dumitru Erhan, Yoshua Bengio, Aaron Courville, and Pascal Vincent. Visualizing higher-layer features of a deep network. Technical Report 1341, University of Montreal, June 2009. Also presented at the ICML 2009 Workshop on Learning Feature Hierarchies, Montr´ eal, Canada

work page 2009

[8] [8]

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image classiﬁcation models and saliency maps. CoRR, abs/1312.6034, 2013

work page internal anchor Pith review Pith/arXiv arXiv 2013

[9] [9]

Axiomatic attribution for deep networks

Mukund Sundararajan, Ankur Taly, and Qiqi Yan. Axiomatic attribution for deep networks. In ICML, 2017

work page 2017

[10] [10]

Neural machine translation by jointly learning to align and translate

Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. September 2014

work page 2014

[11] [11]

Intelligible models for healthcare: Predicting pneumonia 13 risk and hospital 30-day readmission

Rich Caruana, Yin Lou, Johannes Gehrke, Paul Koch, Marc Sturm, and Noemie Elhadad. Intelligible models for healthcare: Predicting pneumonia 13 risk and hospital 30-day readmission. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Min- ing, KDD ’15, pages 1721–1730, New York, NY, USA, 2015. ACM

work page 2015

[12] [12]

Development and validation of a continuous measure of patient condition using the electronic medical record

Michael J Rothman, Steven I Rothman, and Joseph Beals, 4th. Development and validation of a continuous measure of patient condition using the electronic medical record. J. Biomed. Inform., 46(5):837–848, October 2013

work page 2013

[13] [13]

Towards a rigorous science of inter- pretable machine learning

Finale Doshi-Velez and Been Kim. Towards a rigorous science of inter- pretable machine learning. arXiv, 2017

work page 2017

[14] [14]

In the wake of the 2003 and 2011 duty hours regulations, how do internal medicine interns spend their time? J

Lauren Block, Robert Habicht, Albert W Wu, Sanjay V Desai, Kevin Wang, Kathryn Novello Silva, Timothy Niessen, Nora Oliver, and Leonard Feldman. In the wake of the 2003 and 2011 duty hours regulations, how do internal medicine interns spend their time? J. Gen. Intern. Med. , 28(8):1042–1047, August 2013

work page 2003

[15] [15]

How do residents spend their shift time? a time and motion study with a particular focus on the use of computers

Lena Mamykina, David K Vawdrey, and George Hripcsak. How do residents spend their shift time? a time and motion study with a particular focus on the use of computers. Acad. Med., 91(6):827–832, June 2016

work page 2016

[16] [16]

Recurrent Neural Networks for Multivariate Time Series with Missing Values

Zhengping Che, Sanjay Purushotham, Kyunghyun Cho, David Sontag, and Yan Liu. Recurrent neural networks for multivariate time series with missing values. CoRR, abs/1606.01865, 2016

work page internal anchor Pith review Pith/arXiv arXiv 2016

[17] [17]

Lipton, David C

Zachary C. Lipton, David C. Kale, and Randall C. Wetzel. Directly modeling missing data in sequences with rnns: Improved classiﬁcation of clinical time series. In MLHC, 2016

work page 2016

[18] [18]

Clinical intervention prediction and understanding using deep networks

Harini Suresh, Nathan Hunt, Alistair Johnson, Leo Anthony Celi, Peter Szolovits, and Marzyeh Ghassemi. Clinical intervention prediction and understanding using deep networks. May 2017

work page 2017

[19] [19]

Biases in electronic health record data due to processes within the healthcare system: retro- spective observational study

Denis Agniel, Isaac S Kohane, and Griﬃn M Weber. Biases in electronic health record data due to processes within the healthcare system: retro- spective observational study. BMJ, 361:k1479, April 2018

work page 2018

[20] [20]

Possible sources of bias in primary care electronic health record data use and reuse

Robert A Verheij, Vasa Curcin, Brendan C Delaney, and Mark M McGilchrist. Possible sources of bias in primary care electronic health record data use and reuse. J Med Internet Res , 20(5):e185, May 2018

work page 2018

[21] [21]

Sampling rate causes bias in apache ii and saps ii scores

Matti Suistomaa, Aarno Kari, Esko Ruokonen, and Jukka Takala. Sampling rate causes bias in apache ii and saps ii scores. Intensive Care Medicine, 26(12):1773–1778, Dec 2000

work page 2000

[22] [22]

Rao, Troy D

Supriyo Chakraborty, Richard Tomsett, Ramya Raghavendra, Daniel Har- borne, Moustafa Alzantot, Federico Cerutti, Mani Srivastava, Alun Preece, Julier Simon, Raghuveer M. Rao, Troy D. Kelley, David Braines, Murat Sensoy, Christopher J. Willis, and Prudhvi Gurram. Interpretability of deep learning models: a survey of results. In IEEE Smart World Congress 20...

work page 2017

[23] [23]

Smilkov, N

D. Smilkov, N. Thorat, B. Kim, F. Vi´ egas, and M. Wattenberg. SmoothGrad: removing noise by adding noise. ICML workshop on visualization for deep learning, June 2017

work page 2017

[24] [24]

Zeiler and Rob Fergus

Matthew D. Zeiler and Rob Fergus. Visualizing and understanding convolu- tional networks. In David Fleet, Tomas Pajdla, Bernt Schiele, and Tinne Tuytelaars, editors, Computer Vision – ECCV 2014 , pages 818–833, Cham,

work page 2014

[25] [25]

Springer International Publishing

work page

[26] [26]

Striving for Simplicity: The All Convolutional Net

Jost Tobias Springenberg, Alexey Dosovitskiy, Thomas Brox, and Martin A. Riedmiller. Striving for simplicity: The all convolutional net. CoRR, abs/1412.6806, 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014

[27] [27]

Pramod Kaushik Mudrakarta, Ankur Taly, Mukund Sundararajan, and Kedar Dhamdhere. Did the model understand the question? In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, ACL 2018, Melbourne, Australia, July 15-20, 2018, Volume 1: Long Papers , pages 1896–1906, 2018

work page 2018

[28] [28]

”why should i trust you?”: Explaining the predictions of any classiﬁer

Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. ”why should i trust you?”: Explaining the predictions of any classiﬁer. In Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD ’16, pages 1135–1144, New York, NY, USA, 2016. ACM

work page 2016

[29] [29]

A uniﬁed approach to interpreting model predictions

Scott M Lundberg and Su-In Lee. A uniﬁed approach to interpreting model predictions. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors, Advances in Neural Information Processing Systems 30, pages 4765–4774. Curran Associates, Inc., 2017

work page 2017

[30] [30]

Learning to explain: An information-theoretic perspective on model interpretation

Jianbo Chen, Le Song, Martin Wainwright, and Michael Jordan. Learning to explain: An information-theoretic perspective on model interpretation. In Jennifer Dy and Andreas Krause, editors, Proceedings of the 35th In- ternational Conference on Machine Learning , volume 80 of Proceedings of Machine Learning Research, pages 883–892, Stockholmsmssan, Stockholm...

work page 2018

[31] [31]

Infectious diseases society of America/American thoracic society consensus guidelines on the management of community-acquired pneumonia in adults

Lionel A Mandell, Richard G Wunderink, Antonio Anzueto, John G Bartlett, G Douglas Campbell, Nathan C Dean, Scott F Dowell, Thomas M File, Jr, Daniel M Musher, Michael S Niederman, Antonio Torres, Cynthia G Whitney, Infectious Diseases Society of America, and American Thoracic Society. Infectious diseases society of America/American thoracic society conse...

work page 2007

[32] [32]

British thoracic society community acquired pneumonia guideline and the NICE pneumonia guideline: how they ﬁt together

W S Lim, D L Smith, M P Wise, and S A Welham. British thoracic society community acquired pneumonia guideline and the NICE pneumonia guideline: how they ﬁt together. BMJ Open Respir Res , 2(1):e000091, May 2015

work page 2015

[33] [33]

Moving beyond Single-Parameter early warning scores for rapid response system activation

Matthew M Churpek and Dana P Edelson. Moving beyond Single-Parameter early warning scores for rapid response system activation. Crit. Care Med., 44(12):2283–2285, December 2016

work page 2016

[34] [34]

Sustained eﬀectiveness of a primary-team-based rapid response system

Michael D Howell, Long Ngo, Patricia Folcarelli, Julius Yang, Lawrence Mottley, Edward R Marcantonio, Kenneth E Sands, Donald Moorman, and Mark D Aronson. Sustained eﬀectiveness of a primary-team-based rapid response system. Crit. Care Med., 40(9):2562–2568, September 2012

work page 2012

[35] [35]

Electronic health record adoption in US hospitals: Progress continues, but challenges persist

Julia Adler-Milstein, Catherine M DesRoches, Peter Kralovec, Gregory Foster, Chantal Worzala, Dustin Charles, Talisha Searcy, and Ashish K Jha. Electronic health record adoption in US hospitals: Progress continues, but challenges persist. Health Aﬀ., 34(12):2174–2180, December 2015

work page 2015

[36] [36]

Deep ehr: A survey of recent advances in deep learning techniques for electronic health record (ehr) analysis

Benjamin Shickel, Patrick Tighe, Azra Bihorac, and Parisa Rashidi. Deep ehr: A survey of recent advances in deep learning techniques for electronic health record (ehr) analysis. IEEE Journal of Biomedical and Health Informatics, PP(99):1–1, 2017

work page 2017

[37] [37]

Clinical intervention prediction and understanding with deep neural networks

Harini Suresh, Nathan Hunt, Alistair Johnson, Leo Anthony Celi, Peter Szolovits, and Marzyeh Ghassemi. Clinical intervention prediction and understanding with deep neural networks. In Finale Doshi-Velez, Jim Fackler, David Kale, Rajesh Ranganath, Byron Wallace, and Jenna Wiens, editors, Proceedings of the 2nd Machine Learning for Healthcare Conference, vo...

work page 2017

[38] [38]

Deepr: A convolutional net for medical records

Phuoc Nguyen, Truyen Tran, Nilmini Wickramasinghe, and Svetha Venkatesh. Deepr: A convolutional net for medical records. July 2016

work page 2016

[39] [39]

Doctor AI: Predicting clinical events via recurrent neural networks

Edward Choi, Mohammad Taha Bahadori, Andy Schuetz, Walter F Stew- art, and Jimeng Sun. Doctor AI: Predicting clinical events via recurrent neural networks. In Proceedings of the 1st Machine Learning for Healthcare Conference, pages 301–318. jmlr.org, 2016

work page 2016

[40] [40]

Deep computational phenotyping

Zhengping Che, David Kale, Wenzhe Li, Mohammad Taha Bahadori, and Yan Liu. Deep computational phenotyping. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’15, pages 507–516, New York, NY, USA, 2015. ACM

work page 2015

[41] [41]

Lasko, Joshua C

Thomas A. Lasko, Joshua C. Denny, and Mia A. Levy. Computational phenotype discovery using unsupervised feature learning over noisy, sparse, and irregular clinical data. PLoS One, 8(6), 2013. 16

work page 2013

[42] [42]

Learn- ing vector representation of medical objects via emr-driven nonnegative restricted boltzmann machines (enrbm)

Truyen Tran, Tu Dinh Nguyen, Dinh Phung, and Svetha Venkatesh. Learn- ing vector representation of medical objects via emr-driven nonnegative restricted boltzmann machines (enrbm). J. of Biomedical Informatics , 54(C):96–105, April 2015

work page 2015

[43] [43]

Comorbidity clusters in autism spectrum disorders: An electronic health record time-series analysis

Finale Doshi-Velez, Yaorong Ge, and Isaac Kohane. Comorbidity clusters in autism spectrum disorders: An electronic health record time-series analysis. Pediatrics, 133(1):e54–e63, 2014

work page 2014

[44] [44]

Mathieu Guillame-Bert, Artur Dubrawski, Donghan Wang, Marilyn Hrav- nak, Gilles Clermont, and Michael R. Pinsky. Learning temporal rules to forecast instability in continuously monitored patients. JAMIA, 24(1):47–53, 2017

work page 2017

[45] [45]

Hughes, Huseyin Melih Elibol

Thomas McCoy Roy Perlis Finale Doshi-Velez Michael C. Hughes, Huseyin Melih Elibol. Supervised topic models for clinical interpretability. In Proceedings of the 1st Machine Learning for Healthcare Conference , 2016

work page 2016

[46] [46]

Improving palliative care with deep learning

Anand Avati, Kenneth Jung, Stephanie Harman, Lance Downing, Andrew Ng, and Nigam H Shah. Improving palliative care with deep learning. November 2017

work page 2017

[47] [47]

Ying Sha and May D. Wang. Interpretable predictions of clinical outcomes with an attention-based recurrent neural network. In Proceedings of the 8th ACM International Conference on Bioinformatics, Computational Biol- ogy,and Health Informatics, ACM-BCB ’17, pages 233–240, New York, NY, USA, 2017. ACM

work page 2017

[48] [48]

Retain: An interpretable predictive model for healthcare using reverse time attention mechanism

Edward Choi, Mohammad Taha Bahadori, Jimeng Sun, Joshua Kulas, Andy Schuetz, and Walter Stewart. Retain: An interpretable predictive model for healthcare using reverse time attention mechanism. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Systems 29 , pages 3504–3512. Curran Associa...

work page 2016

[49] [49]

Interpretable deep models for icu outcome prediction

Zhengping Che, Sanjay Purushotham, Robinder Khemani, and Yan Liu. Interpretable deep models for icu outcome prediction. 2016:371–380, 02 2017

work page 2016

[50] [50]

Khemani, and Yan Liu

Zhengping Che, Sanjay Purushotham, Robinder G. Khemani, and Yan Liu. Distilling knowledge from deep networks with applications to healthcare domain. In NIPS Workshop on Machine Learning for Healthcare , 2015

work page 2015

[51] [51]

Diﬀerentiation of discrete multidimensional signals

Hany Farid and Eero Simoncelli. Diﬀerentiation of discrete multidimensional signals. IEEE Transactions on Image Processing , 13(4):496–508, 4 2004

work page 2004

[52] [52]

Farid and E

H. Farid and E. P. Simoncelli. Diﬀerentiation of discrete multidimensional signals. IEEE Transactions on Image Processing, 13(4):496–508, April 2004. 17

work page 2004

[53] [53]

Long Short-Term memory.Neural Comput., 9(8):1735–1780, November 1997

Sepp Hochreiter and J¨ urgen Schmidhuber. Long Short-Term memory.Neural Comput., 9(8):1735–1780, November 1997

work page 1997

[54] [54]

Johnson, Tom J

Alistair E.W. Johnson, Tom J. Pollard, Lu Shen, Li-wei H. Lehman, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and Roger G. Mark. Mimic-iii, a freely accessible critical care database. Scientiﬁc Data, 2016

work page 2016

[55] [55]

Michael Felker, Larry A

G. Michael Felker, Larry A. Allen, Stuart J. Pocock, Linda K. Shaw, John J.V. McMurray, Marc A. Pfeﬀer, Karl Swedberg, Duolao Wang, Salim Yusuf, Eric L. Michelson, and Christopher B. Granger. Red cell distribution width as a novel prognostic marker in heart failure: Data from the charm program and the duke databank. Journal of the American College of Card...

work page 2007

[56] [56]

Scalable and accurate deep learning for electronic health records

Alvin Rajkomar, Eyal Oren, Kai Chen, Andrew M Dai, Nissan Hajaj, Pe- ter J Liu, Xiaobing Liu, Mimi Sun, Patrik Sundberg, Hector Yee, Kun Zhang, Gavin E Duggan, Gerardo Flores, Michaela Hardt, Jamie Irvine, Quoc Le, Kurt Litsch, Jake Marcus, Alexander Mossin, Justin Tansuwan, De Wang, James Wexler, Jimbo Wilson, Dana Ludwig, Samuel L Volchen- boum, Katheri...

work page 2018

[57] [57]

A theoretically grounded application of dropout in recurrent neural networks

Yarin Gal and Zoubin Ghahramani. A theoretically grounded application of dropout in recurrent neural networks. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Systems 29, pages 1019–1027. Curran Associates, Inc., 2016

work page 2016

[58] [58]

Adaptive subgradient methods for online learning and stochastic optimization

John Duchi, Elad Hazan, and Yoram Singer. Adaptive subgradient methods for online learning and stochastic optimization. J. Mach. Learn. Res. , 12:2121–2159, July 2011

work page 2011

[59] [59]

J. A. Kellum and N. Lameire. Diagnosis, evaluation, and management of acute kidney injury: a KDIGO summary (Part 1). Crit Care, 17(1):204, Feb 2013

work page 2013

[60] [60]

Kindermans, S

P.-J. Kindermans, S. Hooker, J. Adebayo, M. Alber, K. T. Sch¨ utt, S. D¨ ahne, D. Erhan, and B. Kim. The (Un)reliability of saliency methods. NIPS workshop on Explaining and Visualizing Deep Learning , 2017

work page 2017

[61] [61]

The mythos of model interpretability

Zachary Chase Lipton. The mythos of model interpretability. In ICML Workshop on Human Interpretability , 2016

work page 2016

[62] [62]

Extremely Likely

Aaron Springer, Victoria Hollis, and Steve Whittaker. Dice in the black box: User experiences with an inscrutable algorithm, 2017. 18 8 Appendix 8.1 Experiments - Features For the experiments we restricted the lab tests to the following list of roughly 40 target harmonized features: blood pressure, pulse, respiratory rate, oxygen saturation, blood pressur...

work page 2017