1. Efstathios D Gennatas, Jerome H Friedman, Lyle H Ungar, Romain Pirracchio, Eric Eaton, Lara G Reichmann, Yannet Interian, José Marcio Luna, Charles B Simone, Andrew Auerbach, Elier Delgado, Mark J van der Laan, Timothy D Solberg, Gilmer Valdes. Expert-augmented machine learning. Proceedings of the National Academy of Sciences 117.9 (2020): 4571-4577. https://doi.org/10.1073/pnas.1906831117
Description: One of the dogmas of Machine Learning is that the data will sort itself out and powerful algorithms will find the truth provided with enough data. In this article, I lead a team that proved that this dogma is incorrect and that Machine Learning algorithms and human experts, specifically physicians, learn complementary knowledge. Additionally, we created the first platform that effectively combines physicians' and AI knowledge for the benefit of our patients.
2. Luna JM, Gennatas ED, Ungar LH, Eaton E, Diffenderfer ES, Jensen ST, Simone CB, Friedman JH, Solberg TD, Valdes G. Building more accurate decision trees with the additive tree. Proc Natl Acad Sci U S A. 2019 Oct 01; 116(40):19887-19893. PMID: 31527280. PMCID: PMC6778203
Description: Classification and Regression Trees (CART) and Gradient Boosting are two of the most important Machine Learning algorithms developed up to date. In this theoretical work, we showed how these two algorithms exist in a continuum. Besides the theoretical implication of this work, these results allowed us to develop a single decision tree algorithm, The Additive Tree, that significantly improved over CART while maintaining its interpretability. This work, therefore, will facilitate the safe adoption of artificial intelligence in medicine and its acceptance by physicians. https://doi.org/10.1073/pnas.1816748116
3. G. Valdes, R. Scheuermann, M. Bellerive, A. Olszanski, C. Hung, T. D. Solberg. A mathematical framework for virtual IMRT QA using machine learning. Med. Phys. Vol 43, 7, 4323-4334. 2016
Description: We developed the first system that uses Machine Learning to predict Quality Assurance metrics in Radiation Oncology, Virtual IMRT QA. This algorithm could save precious time in the treatment of patients with radiation as well as allow for the introduction of Adaptative Radiation Therapy. https://doi.org/10.1118/1.4953835
4. Valdes, G., Friedman, J., Jiang, F. and Gennatas, E. "Representational Gradient Boosting: Backpropagation in the Space of Functions." IEEE Transactions on Pattern Analysis and Machine Intelligence (2021). 10.1109/TPAMI.2021.3137715
Description: We developed the the mathematical foundation for the joined optimization of different models like Neural Networks and Gradient Boosting together. This work creates the theoretical basis for the analysis of multi-modal data where the investigators can choose the best model for each type.
5. G. Valdes, Y. Interian, E. Gennatas and M. Van der Laan, "The Conditional Super Learner," in IEEE Transactions on Pattern Analysis and Machine Intelligence, doi: 10.1109/TPAMI.2021.3131976. 10.1109/TPAMI.2021.3131976
Description: Using cross validation to select the best model from a library is standard practice in machine learning. Similarly, meta learning is a widely used technique where models previously developed are combined (mainly linearly) with the expectation of improving performance with respect to individual models. In this article we consider the Conditional Super Learner (CSL) an algorithm that selects the best model candidate from a library of models, conditional on covariates. The CSL expands the idea of using cross validation to select the best model and merges it with meta learning. We propose an optimization algorithm that finds a local minimum to the problem posed and proves that it converges at a rate faster than Op(n1/4). We offer empirical evidence that: (1) CSL is an excellent candidate to substitute stacking and (2) CLS is suitable for the analysis of Hierarchical problems. Additionally, implications for global interpretability are emphasized.