David Ojika, an ECE PhD candidate at UF, has been awarded a $28K research grant by Amazon, his second research award this semester. The award is aimed at accelerating deep learning research using programmable logic (or FPGAs), and will help David and his team design and prototype an FPGA-accelerated deep learning architecture by leveraging Amazon F1 FPGAs.
Deep learning is a branch of machine learning that is inspired by the dynamics of biological neurons, and involves the processing of massive amount of data. This process is highly computation intensive, and often requires specialized hardware (like FPGAs) to drive performance and efficiency.
David’s project seeks to address the challenge of real-time identification of exotic physics particles (e.g. muons) detected by the Compact Muon Solenoid (CMS) detector, one of the particle detectors at CERN’s Large Hadron Collider. The muon is an unstable subatomic particle that behaves like a heavy electron; investigation of its rapid decay process is a major objective of the CMS experiment. Tremendous amounts of data are generated by the collider’s trigger system – during runtime, efficient detection algorithms are needed in order to quickly perform the required physics analyses. Through the use of deep learning algorithms and Amazon’s cloud platform, David intends to improve the performance of these analyses.
Initially, David will train a complex deep learning model suitable for the real-time identification of muons. This training procedure will be conducted in the cloud, utilizing GPUs and large samples of training data from the CMS experiment. After training is completed, the learned model will be deployed on FPGAs and evaluated through a custom-built “hardware-as-a-service” platform. Through his combination of onsite facilities and remote cloud computing resources, David hopes to present a unique solution for researchers participating in large-scale scientific studies, like the CMS experiment, that require rapid processing and storage of massive amounts of data.
Under Dr. Herman Lam for supervised teaching of the course EEL-5714 (Reconfigurable Computing), David will integrate FPGA accelerator systems together with deep learning techniques into the class curriculum, making the Amazon F1 FPGA facility accessible to graduate students with interests in deep learning and artificial intelligence (AI).
David is currently funded by Intel for his PhD studies, and is also an Intel Student Ambassador for AI. He is founding an AI-focused campus club (GatorVision) to foster a community of AI-minded students and researchers. Dr. Darin Acosta and Dr. Ann Gordon-Ross (from the Physics and ECE departments, respectively) supervise David’s research.