Deep Normed Embeddings for Patient Representation
Abstract: We introduce a novel contrastive representation learning objective and a training scheme for clinical time series. Specifically, we project high dimensional EHR. data to a closed unit ball of low dimension, encoding geometric priors so that the origin represents an idealized perfect health state and the Euclidean norm is associated with the patient’s mortality risk. Moreover, using septic patients as an example, we show how we could learn to associate the angle between two vectors with the different organ system failures, thereby, learning a compact representation which is indicative of both mortality risk and specific organ failure. We show how the learned embedding can be used for online patient monitoring, can supplement clinicians and improve performance of downstream machine learning tasks. This work was partially motivated from the desire and the need to introduce a systematic way of defining intermediate rewards for Reinforcement Learning in critical care medicine. Hence, we also show how such a design in terms of the learned embedding can result in qualitatively different policies and value distributions, as compared with using only terminal rewards.
Submission Length: Long submission (more than 12 pages of main content)
Changes Since Last Submission: Corrected some typos in the previous revision.
For completeness we note the changes in that revision below:
We made several revisions to address all the main concerns of the reviewers.
The major chances are as follows:
1) We reproduced the results using the raw data: excluding the representation learning of previous work (Nanayakkara 2022). However, we did change the architecture to an RNN with a fixed history, so some temporal relationships are captured.
2) Several detailed ablations studies are included.
3) We restructured the mortality prediction task and considered several baseline representation learning methods. Briefly, we evaluated the methods using a linear classification protocol: a method that is common in computer vision. We excluded results of fitting random forests.
4) A section on implementation details of all tasks is included in the supplementary material.
In addition, we have corrected some typographical errors, defined loss functions, and added background material on RL.
Assigned Action Editor: ~Krzysztof_Jerzy_Geras1
Submission Number: 75
Loading