NeurIPS 2020

Representation Learning for Integrating Multi-domain Outcomes to Optimize Individualized Treatment


Meta Review

The motivation and objectives of this study are well presented. The theoretical grounding and empirical evaluation in the study are adequate. The work presents a fair novelty to the field, building on previous work by presenting a framework incorporating multi-domain outcomes. The implementation of the model preserves principles from the psychiatry and information theory and provides a cross-field work and is relevant to the NeurIPS community. There is a criticism that needs to be addressed in the discussion in the final version is how sensitive the method is to biases and what could be done to avoid the biases when using the method. NOTE FROM PROGRAM CHAIRS: The paper is accepted, however please revise and expand the Broader Impact statement in the camera-ready version. As one reviewer writes, "Latent psychological constructs are well known to be rife with bias and piping them directly into treatment policy decision making without first validating them presents the risk of propagating that bias into treatment decisions." This risk, as well as possible mitigations, should be discussed in the impact section especially given the sensitive subject matter (mental health).