Aditya Cherukumudi

WORK HISTORY

EDUCATION

SKILLS

ABOUT ADITYA CHERUKUMUDI

I am AI/ML Scientist with 4 yrs of experience in Machine Learning and Deep Learning (using Scipy/Keras/Tensorflow). I also have extensive experience with AWS (specifically in - Batch, ECR, ECS, EKS and S3). At Stats Perform, I’ve worked primarily on a product called AutoSTATS (a broadcast-based player tracking technology). I’ve worked on several different parts of the pipeline including a video ingestion, a neural network based automated clipping system, scoreboard detection and optical character recognition and action & actor recognition using deep learning. I’ve worked on each of these projects starting from research and creating proof-of-concepts to deploying models in production and batching processing via AWS. I have also worked on building an automated model re-training pipeline using Kubernetes (EKS on AWS) and Katib to allow models to be efficiently (re)trained while also carrying out an automated hyper parameter search.Previously, I worked as a Research Associate for the Machine Learning and Computer Vision group at Disney Research Pittsburgh. My focus was working primarily towards developing deep learning methods for modeling multi-agent trajectories in sport. During this period, I was also a core member of Audience insights group that worked on analyzing audience behavior (using computer vision) to predict TV show/movie success. During my time at Disney, I presented my research on “realtime interactive play sketching with synthesized NBA defenses” at the MIT Sloan Sports Analytics conference 2018, where my team and I finished second place in the Research Papers Competition (and was also called the “coolest” paper at the conference in 2018 by The Ringer). This work involved using deep imitation learning to predict and simulate fine-grained context specific defensive behavior in basketball. My primary contributions involved data cleaning & organization, sport specific data routing, feature engineering and network modeling & training. I also worked on designing the trajectory segmentation algorithm to implicitly extract temporal information from sketch data.Over the years, I’ve gained the experience in managing project priorities, deadlines and deliverables while contributing strongly towards not only the research efforts, but also developing products and prototypes stemming from our research. This has also given me good exposure to working effectively in multi-cultural environments. I am also extremely passionate about sports and fitness. I am a certified Animal Flow instructor and teach online fitness classes in my free time!