Ross King (University of Manchester)

Drug Design and Constructive Machine Learning

There is an urgent need to make drug discovery cheaper and faster, both to promote the development of treatments for diseases currently neglected for economic reasons, such as tropical and orphan diseases, and to generally increase the supply of new drugs. My colleagues and I have developed the Robot Scientist "Eve", which automates library screening, hit confirmation, and lead generation through Quantitative Structure Activity Relationship (QSAR) learning and testing (active learning). A QSAR is a function that predicts a compound's activity in an assay. Using econometric modelling we have determined the conditions where Eve outperforms standard automation. Eve's greatest limitation is that to test its QSAR hypotheses it can only select compounds from its library: it cannot synthesise new compounds. This is restrictive because even the largest compound libraries in the world have only a few million compounds, whereas the potential number of compounds that could be synthesised is probably at least 1060. Laboratory automation equipment is beginning to be become available that could synthesise almost arbitrary compounds, and these could be integrated with Eve. The problem then is the constructive machine learning one of deciding which compounds to select. I will describe our work on developing Eve, QSAR active learning, and our initial work on constructive machine learning.

Ross D. King is Professor of Machine Intelligence at the University of Manchester, UK. His main research interests are in the interface between computer science and biology/chemistry. The research achievement he is most proud of is originating the idea of a "Robot Scientist": using laboratory robotics to physically implement a closed-loop scientific discovery system. His Robot Scientist "Adam" was the first machine to hypothesise and experimentally confirm scientific knowledge. His work on this subject has been published in the top scientific journals, Science and Nature, and has received wide publicity. He is also very interested in computational aesthetics and economics.

Bob Keller (Harvey Mudd College)

Machine Learning Applied to Musical Improvisation

Development of music education software inevitably leads to questions of how to acquire musical knowledge to be made available to the student user. I will describe machine learning of patterns for accompaniment styles and grammars for improvisation, based on melodic abstraction, clustering, and chaining. I will also discuss supervised and unsupervised approaches to improvising over chord progressions using neural network. Finally, I will mention a challenging unsolved application: learning to classify idiomatic patterns in chord progressions.

The speaker gratefully acknowledges collaborations with Jim Herold, Brandy McMenamy, Sayuri Soejima, Jon Gillick, Kevin Tang, Greg Bickerman, Sam Bosley, Peter Swire, Hayden Blauzvern, and Kevin Choi.

Robert M. Keller has been professor of computer science at Harvey Mudd College since 1991. He previously held positions at Princeton University, the University of Utah, and Quintus Computer Systems. He is also a jazz musician and teaches jazz improvisation, which led to his work in software that includes a variety of machine learning aspects for helping musicians increase their improvisation skills. His prior contributions included publications in areas of declarative programming and parallel computing, with supervision of over ten PhD theses in those areas.

Doug Turnbull (Ithaca College)

Local Music Discovery

I will discuss my work on creating a better, music discovery experience. In particular, I will talk about a new personalized Internet radio system that combines computer audition, social network analysis and intelligent user interface design. The goal is to improving both how musicians grow their audiences and listeners discover new music within a local geographic region.

Doug Turnbull is an assistant professor of Computer Science at Ithaca College. His research interests include multimedia information retrieval, computational music analysis, and machine learning. Doug received a B.S.E. degree (with honors) in Computer Science from Princeton University in 2001, and Ph.D. degrees in Computer Science & Engineering from UC San Diego in 2008.

Josh Tenenbaum (MIT, tentative)