Life Science Hour Seminar Series

Name:Falk Huettmann
Title:Advancing 'open source' data mining and deep learning even further: Updating the latest paradigms of science and inference
Date:Friday, 20 October 2017
Location:Murie Life Science Bldg, Murie Auditorium.


Increasingly, the quantitative argument became a major platform to convince in the sciences, and even for public education and discussions. This type of reasoning dominates science-based Wildlife Conservation and Management decisions. However, it is still a relatively new development in the western-dominated sciences. And until recently decisions were mostly based on significance levels and such metrics (e.g. p-values, parsimonious AIC, confidence intervals and fitting details like R2). The older Bayesian school basically re-entered this scene, now empowered with Markov Chain Monte Carlo (MCMC) methods (e.g. WinBUGS) starting to challenge frequentist inference. However, already after WW2 Machine Learning (ML) evolved into a discipline of its own featuring over hundred of algorithms. This evolved due to an increased and urgent need to resolve ‘classification problems’ and real-world applications, and also due to the available computing power centered for instance around Artificial Neural Networks (ANN) or ‘The Cloud’. By now, ML algorithms have taken many genetic, industrial and pharmaceutical applications by storm. They are based on various and ‘deep’ computational algorithms such as maximum entropy, recursive binary partitioning, Classification & Regression Trees (CARTs), boosting, bagging, all freely available in R now as ‘ensembles’. These are the methods of choice, and cutting edge applications are still in the commercial and strategic realm!

In this presentation I will present and discuss this recent global development in the sciences as well as the evolution of ML and the associated science culture. Further I will show how it provided a new research frontier as well as an improved paradigm and culture-shift based on research, publications, related job descriptions and public education. Based on our own and other published long-term research worldwide, I will elaborate in which disciplines ML is still widely underused and underpublished, and why that is, considering links with culture, genealogy, administration, society, funding and education. I conclude with a renewed call for ethics, open access and repeatability in quantitative applications and for the sciences, and how ML applications can help to achieve a good stewardship of the earth within an effective (science) infrastructure.

Life Science Hour Seminars

Academic Year