Jingmin Wang Email & Phone Number

• Conducted extensive experiments using Twin-Delayed Deep Deterministic Policy Gradient to generate offline RL datasets. These were used for reference for an offline RL meta-learning paper submission at the Conference on.
• Implementing/researching 6+ existing offline model-based RL algorithms and devising novel methods of evaluating them within high-dimensional, long-horizon environments
• Managing and debugging a large RL codebase consisting of over 20 algorithms and over 40,000 lines of code, reducing error rates and improving overall code coherence

Toronto, ON

• Established a unique method of optical flow motion estimation for a camera with up to 20000 exposures/second by calculating pixel intensity frame differences and using RAFT, convolutional gated recurrent unit based.
• The previous method was included in an International Solid States Circuits Conference (ISSCC 2024) paper submission.
• Worked with Coded Exposure Pixel (CEP) cameras, and engineered custom banded mask patterns for more precise light exposure control, filtering out indirect light and reducing glare
• Developed a hardware framework in Xilinx Vivado and external oscilloscopes and function generators to generate custom FPGA signals, allowing synchronization between a raster scanning projector and CEP camera
• Implemented 4+ lossless compression schemes, from Huffman encoding, Run-Length encoding, LZW encoding, and more

Toronto, Ontario, Canada

• Will be developing a reinforcement learning full lecture series and course, covering RL fundamentals to advanced topics such as decision transformers, offline RL, and RL with human feedback
• Will be delivering course content to hundreds of students

Toronto, ON

• Directing a project and leading 8-10 developers to construct a reinforcement learning and natural language processing based novel framework for real-time Forex and commodity trading.
• Devising innovative features in addition to well-known market indicators, such as predicting the change in asset prices using the risk-to-reward ratio and a classification network

Toronto, ON

• Developed the semantic segmentation submodel for PhotoML, an innovative photo culling software, using BEiT, which is a transformer (BERT) pre-trained on image data

Bachelor Of Applied Science - Basc, Electrical And Computer Engineering, 3.95/4

Full Ib Diploma, Math Analysis And Approaches Hl, Physics Hl, Chemistry Hl, Economics Hl, 45/45