Current CoE-funded Projects - Center of Excellence in Data Science and Artificial Intelligence

The CoE proudly sponsors the research of faculty and partner companies. The Center’s funding allows researchers to produce innovative and visionary approaches to data science in a variety of areas and industries. We currently are funding these partnerships and are looking forward to seeing the results of the research and the positive economic impact that it will have on New York State and the world of data science.

Advancing Wavefront Sensor Performance via Deep Learning

PI Researcher: Drew Maywar
Rochester Institute of Technology

The optical wavefront — the transverse spatial variation of the optical phase — plays a significant role in the propagation of optical beams and in the characterization of optical components & systems. Wavefront sensors are therefore important instruments for the worldwide optics community and especially for our regional optics ecosystem. This project seeks to apply deep learning to advance the performance of the RAM Photonics QuantoPhase wavefront sensor, harvesting & learning aspects of raw shearing interferograms including in the realm of pre-formed interferograms to allow for previously prohibited sensor design.

Enabling Ontology Development in Graph Databases

PI Researcher: Carlos Rivero
Rochester Institute of Technology

L3Harris currently deploys ontologies that are experiencing data limitations, which compromise the performance and scalability of the applications. Thus, there is motivation to find an approach to migrate ontologies into a graph database that can provide performance, flexibility, and scalability benefits, particularly for applications involving complex and highly interconnected data. This project aims to investigate methods to enable ontology development into Neo4j, an open-source graph database, thereby empowering users to extract valuable insights from existing ontologies and expanding queries to extensive multi-domain data sets. Furthermore, the project aims to help users retrieve knowledge from existing ontologies, even when the users are not computing experts.

Controllable ASL Generation to Support Communication between Hearing Parents and DHH Children

PI Researcher: Zhen Bai
University of Rochester

Over 90% of Deaf and Hard of Hearing (DHH) children in the US are born to hearing families. This put DHH children at several risk of language deprivation that leads to inreversible and lifelong impact on linguistic, cognitive, and socio-emotional development. This project aims to design and develop advanced American Sign Langauge (ASL) generation technologies that support hearing parents to simultaneously sign and learn ASL during face-to-face interaction. The central innovation is the high controability of ASL video generation that leverages cutting-edge Generative AI technologies and nuanced needs of hearing parents in providing ASL on-the-fly to DHH children.

AI-Driven Modeling and Data Analytics for Optimizing Thermoelectric Generators in Weather-Resilient and Maritime Applications

PI Researcher: Carlos Rivero
Rochester Institute of Technology

This project will develop AI-driven tools to optimize the deployment and utilization of thermoelectric generators (TEGs) developed by SATEG Corp. By integrating experimental device outputs, environmental datasets, and calibrated computational models, we will build predictive frameworks that capture the coupling between device size, ambient conditions, and power output. Machine learning methods will enable two complementary tools: a sizing module that streamlines customer project development and a data analytics platform that transforms deployment data into insights on sea ice, ocean currents, and weather patterns. These tools will strengthen SATEG’s commercialization and advance resilient energy solutions in New York.

Enhancing Guitar Practice and Songwriting with Automatic Guitar Tablature Transcription

Headshot Photo of Zhiyao Duan PI Researcher: Zhiyao Duan
University of Rochester

Guitar is among the most popular musical instruments thanks to its versatility, convenience, and affordability. Guitarists often prefer to use guitar tablature when reading, writing, learning, and sharing music. This project aims to develop highly accurate automatic Guitar Tablature Transcription (GTT) algorithms that can convert guitar recordings into guitar tablature. The key idea is to train advanced neural network models on massive amounts of synthetic guitar data to overcome current bottlenecks data scarsity and model simplicity. The developed algorithms will be deployed to the industry partner ReWriter’s product to enhance guitar practice and songwriting experiences.