Robust data analysis pipelines often require human intervention for important decision-making steps. Unfortunately, inherent differences between the people making those decisions introduces an irreproducibility into the analysis. It also makes it difficult to precisely communicate the details of the analysis procedure, and the need for direct human input hinders automation and experimental throughput. In this talk, we will cover the use of probabilistic inference to automate the types of decision-making steps that traditionally require human insight, and discuss how the Bayesian approach to probability enables the development of powerful, quantitative representations of the human decision-making process. A general overview of the use of Bayesian probabilistic inference to select the best model of some data will be explained at the non-expert level. Recent advances in modeling the intrinsic ‘shape’ of data will then be discussed. In analogy to a human spotting a peak in a spectrum, this Bayesian approach will be used in step-by-step example calculations to automatically determine whether the presence of a peak in 1D and in higher dimensional NMR spectra are supported by the data; open-source code for these examples will be made available on GitHub.
