Travis Zhang Email & Phone Number

Ithaca, NY, US

- Improved object detection for self-driving using unsupervised learning by up to 15% (Accepted into Neurips 2022)- Developed novel object tracking pipeline using Transformers

- Host Office Hours- Create and test NLP programming assignments (HMMs and MEMMs, LSTM models, FNNs, and Transformers)

- Onboarded 10+ new members about ML basics (Convnets, Grad Descent, etc.) and mentored their first projects- Led team of 20+ members and taught data science techniques and skills- Managed AWS and Azure Cloud compute resources and GPU server for team

New York City Metropolitan Area

• Reduced time to retrieve features for Smart Shuffle ML model from weeks to hours
• Designed and implemented vector hydration and vector aggregations using Apache Flink for Smart Shuffle
• Developed SQL query to define near-real-time features for improved model recommendations

- Hosted office hours 2+ hours weekly for 300+ students- Graded student homework and exams for 3+ hours each weekHead TA (Spring 2023)

- Prepared and delivered recitation sessions for 15 students weekly- Graded student homework and exams for over 3 hours each week

New York, United States

- Developed active learning python package to improve ML model performance- Containerized and built testing suites using Docker and CI/CD pipeline to deploy product- Designed, implemented, and deployed 2 event-driven architectures (Apache Kafka and Redis) for ML product- Implemented scatter-gather for Kafka version and compared performances between the two.

- Helped organize and host academic events including research night and student-faculty luncheon- Managed academic team and lead meetings

- Mathematically formulated signaling actions to analyze how it modifies Deep Reinforcement Learning (RL) agents’ belief state- Modified ReBeL C++ code to implement custom signaling actions

- Built Transient Attribute classifier in Pytorch to help improve Generative Adversarial Network (GAN) training- Designed new GAN pipeline incorporating Transient Attribute classifier for improved cross-view image geolocalization- Applied 2 object detection networks to detect buildings in satellite imagery

- Applied transformation-invariant constraints on adversarial training using Tensorflow to improve robustness of Convolutional Neural Network (CNN)- Implemented 3 algorithms to reduce time for loss to converge for proposed methodology (up to 60% reduction)- Designed physical attack to fool Deep Reinforcement Learning (RL) Agent using Pytorch (accepted into.

Master Of Science - Ms, Computer Science

Bachelor'S Degree, Computer Science

Education record

Activities and Societies: Robotics, Science Research, Robotics Volunteering, Science Olympiad, National Honor Society, Mathworks Math.