2019/08/27 by Iñigo Urteaga, Urteaga, Iñigo, Tristan Bertin +7
Biochemistry, Genetics and Molecular Biology · Computer Science · #Applications (stat.AP) #FOS: Computer and information sciences #Gaussian Processes and Bayesian Inference #Generative Adversarial Networks and Image Synthesis #Genetic and phenotypic traits in livestock #Machine Learning (cs.LG) #Machine Learning (stat.ML)
paper · pdf · doi:10.48550/arxiv.1908.10226
openalex publication_date 2019/08/27 · openalex created_date 2019/12/05 · openalex updated_date 2026/07/28
We present an end-to-end statistical framework for personalized, accurate,\nand minimally invasive modeling of female reproductive hormonal patterns.\nReconstructing and forecasting the evolution of hormonal dynamics is a\nchallenging task, but a critical one to improve general understanding of the\nmenstrual cycle and personalized detection of potential health issues. Our goal\nis to infer and forecast individual hormone daily levels over time, while\naccommodating pragmatic and minimally invasive measurement settings. To that\nend, our approach combines the power of probabilistic generative models (i.e.,\nmulti-task Gaussian processes) with the flexibility of neural networks (i.e., a\ndilated convolutional architecture) to learn complex temporal mappings. To\nattain accurate hormone level reconstruction with as little data as possible,\nwe propose a sampling mechanism for optimal reconstruction accuracy with\nlimited sampling budget. Our results show the validity of our proposed hormonal\ndynamic modeling framework, as it provides accurate predictive performance\nacross different realistic sampling budgets and outperforms baselines methods.\n