PeopleResearcher Spotlight
Spring 2023
Student Reseracher, DELOS Project
McCormick Engineering Excellence Award in Deep Learning
The Center for Deep Learning is pleased to bestow the McCormick Engineering Excellence Award in Deep Learning on Joseph Holtzman. Joseph has been a major contributor to the DELOS project since joining the team in June, 2022.
His work this quarter has centered on improving Delos’ Drift component and developing perturbation algorithms for the project’s drift stress testing module, a crucial component in the software’s overall effectiveness. After researching suitable approaches, Joseph is currently implementing an algorithm using joint distribution to capture relationships among features, and will test the new algorithm in a Forest Cover Type Dataset. Joseph also refactored DELOS’ drift detection component by introducing a burning period which improved the performance of the Drift module by 25%. The new perturbation algorithm in the drift stress testing module and refactored Drift Module will significantly strengthen DELOS.
Joey also took the initiative to create comprehensive documentation for DELOS, ensuring that the project's codebase and functionality are well-documented and easily accessible for new team members during the onboarding process. It is seldom we witness a high school student who possesses deep proficiency in software programming and who follows software engineering best practices.
This is just the latest of several contributions to the project. In earlier quarters, Joseph has implemented the Mongo DB helm chart as well as Unit Testing for DELOS’ ClientAPI and Drift components. Joseph has also succeeded in automating several functions of DELOS, including Unit Testing using GitHub actions and auto initiating microservices when the database is ready. Alongside his technical skills, Joseph has served as a strong spokesman for the project on through regular presentations of his projects on project board calls. Demonstrating an exceptional ability to pick up new material, identify code issues, and implement elegant solutions, Joseph has contributed significantly to the overall success of DELOS as a software package.
The Center deeply appreciates Joseph’s strong ongoing work on the project, and looks forward to his contributions through the end of this summer before enrolling at Stanford University this fall.
Student Researcher, REFIT Project
McCormick Engineering Excellence Award in Deep Learning
The Center for Deep Learning is pleased to bestow the McCormick Engineering Excellence Award in Deep Learning to Tony Luo. Tony has made substantial contributions to the REFIT project ever since he became part of the team in Fall 2021. As the REFIT team’s longest-serving member, he has contributed substantially to the development and testing of the current version of REFIT, while serving as a vital reference on REFIT’s functions for new team members.
Tony’s dedicated efforts for this quarter were primarily focused on enhancing REFIT through the addition of new features. Notably, he enabled Flink to delve into historical data for advanced feature engineering, a critical aspect of REFIT's AI component. Tony's exceptional work included developing a prototype to showcase this functionality. The incorporation of this feature holds immense significance for REFIT, as accessing past data is indispensable in time series forecasting to enable effective feature engineering and accurate predictions.
This latest achievement stands as a testament to Tony's exceptional contributions within REFIT. Previously, he also played a key role in implementing the Divvy bikes use case, which utilized REFIT to predict station checkouts at specific times. Additionally, Tony exhibited his expertise by developing unit tests that demonstrated the seamless streaming of real-time data via Kafka within REFIT. His ability to quickly grasp new concepts and swiftly address software challenges has greatly contributed to REFIT's overall success. Tony's remarkable dedication and proficiency have been instrumental in driving the accomplishments of the REFIT project.
The Center for Deep Learning gratefully acknowledges Tony’s contributions to REFIT’s success, and wishes him well as he graduates from Northwestern to begin a Master’s degree program at UC Berkeley.
Student Researcher, DELOS Project
Erick joined the DELOS team in November 2022, and his contributions have significantly advanced the DELOS framework’s capabilities in dealing with unstructured data.
Erick added support for processing unstructured data feature embeddings in the KSQL stream processing. This enhancement has allowed DELOS to handle unstructured data such as text, images and others in real-time, extending the scope of DELOS real-time serving. Furthermore, Erick explored vector databases to store the embeddings and identified Milvus DB as the optimal choice for integration within the DELOS framework. By integrating Milvus DB, Erick has enabled efficient storage and retrieval of the processed streams. As unstructured data processing is one of the project’s major milestones, this represents a significant step forward for DELOS as a platform.
To test these capabilities, Erick designed a client simulator capable of generating feature embeddings using Bidirectional Encoder Representations from Transformers (BERT). The simulator pushes the generated embeddings to DELOS to mimic real user behavior based on a VQnA use case (Images and Texts) he also implemented. The VQnA use case and the simulator has been crucial in validating the functionality and performance of the DELOS framework for unstructured data. This ensured DELOS’ reliability and effectiveness in real-world scenarios.
These accomplishments have opened up exciting possibilities for real-time processing within the DELOS framework. We look forward to Erick’s continued contributions to the project’s success.
Fall 2022
Joseph Holtzman
Student Researcher, DELOS Project
Joseph has been a major contributor to the DELOS project since joining the team in June. His work this quarter has centered on perturbation algorithms for the project’s drift stress testing module, a crucial component in the software’s overall effectiveness. After researching suitable approaches, Joseph identified and implemented the algorithms himself. Once the algorithms were in place, his initial test on a bank marketing dataset proved successful, with a second test currently in progress. Adding this functionality to the drift stress testing module significantly strengthens DELOS, and was undertaken with the independence and creativity that has come to characterize Joseph’s work at CDL.
This is just the latest of several contributions to DELOS. In earlier quarters, Joseph has implemented unit testing for DELOS’ ClientAPI and Drift components as well as a Mongo DB helm chart. Joseph has also succeeded in automating several functions of DELOS, including Unit Testing using GitHub actions and auto-initiating microservices when the database is ready. Demonstrating an exceptional ability to pick up new material, identify code issues, and implement elegant solutions, Joseph has contributed significantly to the overall success of DELOS as a software package.