ML@Apple | Ph.D.@GeorgiaTech
I am elated to announce that I have successfully defended by Ph.D. thesis titled Learning to Manipulate Latent Representations of Deep Generative Models last December (2020). I have had a great journey through grad school at Georgia Tech over the last ~6 years where I have learnt a lot and met some really incredible people. Looking forward to the next phase of my life now.
Our review paper titled “An Interdisciplinary Review of Music Performance Analysis” was published in the TISMIR journal as a part of the Special Collection: 20th Anniversary of ISMIR. The full paper is available online here.
Had the pleasure of attending the all-virtual 21st ISMIR conference. Even though it was a bummer to not be able to visit Montréal again, the amazing job done by the conference organizers and the warm ISMIR community made sure the virtual conference was as close as possible to the in-person event. We presented our work on the dMelodies dataset and on score-informed networks for music performance…
Blog post summarizing my recent journal article on attribute-based regularization of VAEs.
Blog post summarizing our upcoming ISMIR ‘20 paper presenting a new symbolic music-based dataset for disentanglement learning.
My paper titled “Attribute-based Regularization of Latent Spaces for Variational Auto-Encoders”, was published in Neural Computing and Applications. I explore a new supervised training method to create structured latent spaces where specific continuous-valued attributes are forced to be encoded along specific dimensions of the latent space. The full paper is available here. Alternatively, the…
Blog post summarizing my ISMIR ‘19 paper on music inpainting. The work shows that Recurrent Neural Networks can be trained using latent embeddings of a Variational Auto-Encoder-based model to perform interactive music generation tasks such as inpainting.
Had a great time attending the 20th ISMIR (International Society for Music Information Retrieval) conference at Delft, Netherlands to present my work on models for music inpainting. The full paper is available here.
Our paper titled “Explicitly conditioned melody generation: A case study with interdependent RNNs” was published in the 7th MUME (International Workshop on Musical Metacreation) workshop. This research presents a comparative analysis of RNN-based music generation models when conditioned with explicit musical information. A blog post summarizing the above paper can be found here and the full paper…
Attended the ICML workshop on Machine Learning for Music Discovery to present my work on regularizing latent spaces of VAE (Variational Auto-Encoder)-based models for automatic music generation. The proposed method can provide users with explicit control over musical attributes such as note density, rhythmic complexity, etc., and thereby, help design intuitive musical interfaces to enhance…