David Loker

Algorithmic Songwriting with ALYSIA

This paper introduces ALYSIA: Automated LYrical SongwrIting Application. ALYSIA is based on a machine learning model using Random Forests, and we discuss its success at pitch and rhythm prediction. Next, we show how ALYSIA was used to create original pop songs that were subsequently recorded and produced. Finally, we discuss our vision for the future of Automated Songwriting for both co-creative and autonomous systems.

Effects of rooting via out-groups on in-group topology in phylogeny

Users of phylogenetic methods require rooted trees, because the direction of time depends on the placement of the root. While phylogenetic trees are typically rooted by using an out-group, this mechanism is inappropriate when the addition of an out-group changes the in-group topology. We perform a formal analysis of phylogenetic algorithms under the inclusion of distant out-groups. It turns out that linkage-based algorithms (including UPGMA) and a class of bisecting methods do not modify the topology of the in-group when an out-group is included. By contrast, the popular neighbour joining algorithm fails this property in a strong sense: every data set can have its structure destroyed by some arbitrarily distant outlier. Furthermore, including multiple outliers can lead to an arbitrary topology on the in-group. The standard rooting approach that uses out-groups may be fundamentally unsuited for neighbour joining.

Effects of rooting via outgroups on ingroup topology in phylogeny

Users of phylogenetic methods require rooted trees, because the direction of time depends on the placement of the root. Phylogenetic trees are typically rooted through the use of an outgroup. However, this rooting mechanism is inappropriate when adding an outgroup yields a different topology for the ingroup. We perform a formal analysis of the response of different phylogenetic algorithms to the inclusion of distant outgroups. We prove that linkage-based algorithms, which include UP-GMA, do not modify the topology of the ingroup when an outgroup is included. A class of bisecting algorithms are similarly unaffected. These results are the first to provide formal guarantees on the use of outgroups for rooting phylogentic trees, guaranteeing that this rooting mechanism will not effect the structure of any ingroup when certain algorithms are used. By contrast, the popular neighbour joining algorithm fails this property in a strong sense. Every data set can have its structure destroyed by some arbitrarily distant outlier. Moreover, including multiple outliers can lead to an arbitrary topology on the ingroup. The standard rooting approach that uses outgroups may be fundamentally unsuited for neighbour joining.