We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Discovering Invariances in Healthcare Neural Networks

Abstract: We study the invariance characteristics of pre-trained predictive models by empirically learning transformations on the input that leave the prediction function approximately unchanged. To learn invariance transformations, we minimize the Wasserstein distance between the predictive distribution conditioned on the data instances and the predictive distribution conditioned on the transformed data instances. To avoid finding degenerate or perturbative transformations, we further regularize by adding a similarity term between the data and its transformed values. Applying the proposed technique to clinical time series data, we discover variables that commonly-used LSTM models do not rely on for their prediction, especially when the LSTM is trained to be adversarially robust.
Comments: An improved theorem 1
Subjects: Machine Learning (cs.LG); Quantitative Methods (q-bio.QM); Machine Learning (stat.ML)
Cite as: arXiv:1911.03295 [cs.LG]
  (or arXiv:1911.03295v2 [cs.LG] for this version)

Submission history

From: Mohammad Taha Bahadori [view email]
[v1] Fri, 8 Nov 2019 14:48:05 GMT (290kb,D)
[v2] Tue, 14 Jan 2020 03:40:38 GMT (314kb,D)

Link back to: arXiv, form interface, contact.