The Researchers’ Symposium is an annual event held at City University London where doctoral students have the opportunity to showcase their ongoing research to a primarily non-technical audience. My abstract was selected this year for the 3rd Annual Researchers’ Symposium and I opted for presenting a poster here, titled “A Neural Network for Predicting Musical Pitch”. The abstract I submitted is the following:
“The analysis of sequential patterns is important for extracting information from music owing to its fundamentally temporal nature. Neural Networks and Markov models are two classes of models that have been considered frequently for predicting sequences of events in music. The latter, while being successful at modelling the joint probability of short musical sequences, suffer from problems pertaining to the curse of dimensionality and zero-occurrence as the sequences become longer. Here, we present a distributed model for music prediction based on a type of neural network called the Restricted Boltzmann Machine (RBM). We evaluate this model, first on sequences of musical pitch in a corpus of monophonic MIDI melodies. The results show that this model is able to make use of information present in longer sequences more effectively than previously evaluated n-gram models, outperforming them on the said corpus while also scaling gracefully in the number of free parameters required. As a case study, we also employ this model for classifying folk melodies according to origin.
Furthermore, a discussion on future extensions of the model will also be presented. Of relevance here is its extension to a larger structure known as a deep belief network which is capable of learning interesting features of data presented to it at multiple levels of abstraction. Given the encouraging results with the proposed model so far, its application to Music Information Retrieval tasks such as music transcription and segmentation will also be included in the discussion. Work is currently in progress to generalize this model for learning sequences of other musical dimensions such as note-duration, scale-degree, etc. for prediction.”
The poster I presented (created in Beamer/LaTeX) is also included below in this post. It won the Best Poster Presentation Award at the event.