CV

Research Scientist Intern

ByteDance Inc.

• Developed generative models (VAE and Transformers) to improve various automatic music generation pipelines.

Research Scientist Intern

Kwai Inc.

• Conducted multimodal modeling of dance videos focusing on visual beat tracking and real-time body gesture recognition.
• Created a real time digital audio effects suite in C++ (JUCE) for iOS.

Research Assistant

University of Rochester, AIR Lab

• Developing novel AI tools (algorithms and prototypes) to assist in the music making process using generative models.

Software Engineer

Metis Cyberspace Technology

• Designed algorithms for real-time remote monitoring and performance assessment of equipment onboard vessels.

PhD

University of Rochester

Electrical and Computer Engineering

• Focus Areas: Deep Learning, Music and Audio Signal Processing
• Supervised by Prof. Zhiyao Duan

BSc and MEng

National Technical University of Athens

Electrical and Computer Engineering

• Thesis: A Brain Computer Interface (BCI),using Steady State Visual Evoked Potentials (SSVEP), for the task of maze navigation.
• Supervised by Prof. A. G. Stafylopatis and Dr G. Siolas

Score Reduction for Guitar Through Reinforcement Learning

ISMIR Late Breaking & Demo

Harp 2.0: Expanding Hosted, Asynchronous, Remote Processing For Deep Learning In The Daw

ISMIR Late Breaking & Demo

In the 2.0 release we introduce support for MIDI-based and audio/MIDI labeling models

Euterpe: A Web Framework for Interactive Music Systems

Journal of the Audio Engineering Society

Euterpe is a web framework for interactive music systems.

HARP: Bringing Deep Learning to the DAW with Hosted, Asynchronous, Remote Processing

Machine Learning for Creativity and Design at NeurIPS 2023

HARP is a system for integrating deep learning into DAWs using hosted, asynchronous remote processing.

Draw and listen! a sketch-based system for music inpainting

Transactions of the International Society for Music Information Retrieval

A sketch-based system for music inpainting.

Collagenet: Fusing arbitrary melody and accompaniment into a coherent song

International Society for Music Information Retrieval

A method for fusing arbitrary melody and accompaniment into a coherent song.

BachDuet: A deep learning system for human-machine counterpoint improvisation

Proceedings of the International Conference on New Interfaces for Musical Expression

A deep learning system for human-machine counterpoint improvisation.

BachDuet: A human-machine duet improvisation system

ISMIR Late Breaking & Demo

A system for human-machine duet improvisation.

Guitar Score Reduction as a Reinforcement Learning Problem

• Framed the task of guitar score reduction as a combinatorial optimization problem and used Proximal Policy Optimization (PPO) to solve it.
• Designed novel rule-based and data-driven reward functions to guide the learning process.
• Used a transformer-based RL agent that operates on scores represented as graphs.

HARP

• Lead Developer
• HARP lets users of Digital Audio Workstations (DAWs) access large state-of-the-art deep learning models using cloud-based services without breaking the within-DAW workflow.

Euterpe: A Web Framework for Interactive Music Systems

• Enabled researchers without JavaScript expertise to easily deploy musical agents on the web.
• Supported real-time audio/MIDI synchronization and data visualization.
• Re-Implemented various deep-learning musical agents using Euterpe and gave a tutorial in ISMIR 2023.

Draw and listen!

• Built a sketch-based system for music inpainting enabling users to draw a melodic contour and hear them realized instantly.
• Derived a new melody disentanglement scheme: melody = contour + rhythm + context.
• Designed a VAE architecture that realizes the above disentanglement.

Score Following for Event Augmented Live Performances

• Implemented a modified ODTW algorithm for real-time audio-score alignment.
• Developed a UI to visualize the alignment and activate events.
• Used OSC to send event in real-time a TouchDesigner instance for triggering sound and video effects
• Deployed the system in a mini-concert with the TableTopOpera.

BachDuet

A human-machine duet improvisation system.

• Designed a RNN model for real-time musical counterpoint improvisation.
• Trained on duets extracted from Bach Chorales.
• Implemented a prototype system and demoed it live at various venues.