1

Computer Assisted Composition with Recurrent Neural Networks

In the audio playlist linked below, we take a piece by Mozart and **1)** Fix the melody line (the first track). **2)** Fix the rhythmic (or timing) information of the remaining three tracks to that of the original midi file (last track). **3)** Select the pitches of the remaining tracks conditional on the above (second and third tracks); one example uses conditional sampling with our **particle filter**, the other using conditional probability maximisation with a **beam search**, giving two different styles.

Modelling Symbolic Music: Beyond the Piano Roll

Diffeomorphic Dimensionality Reduction

Sparse Multiscale Gaussian Process Regression

Markerless 3d Face Tracking

Learning with Transformation Invariant Kernels

Implicit Surface Modelling as an Eigenvalue Problem

Implicit Surface Modelling with a Globally Regularised Basis of Compact Support