About

Anurag Dixit is a Senior Deep Learning Software Engineer specializing in Autonomous Driving at NVIDIA with nearly 9 years of relevant experience. He is an expert in leveraging Deep Learning solutions via advanced experimental research code to develop Automotive grade high-quality software. His current role at NVIDIA involves optimizing the performance of DNNs for fast inference and improved accuracy over customer's model for NVIDIA GPUs and HW accelerators. He also designs, scopes out and implements end-to-end applications, runtime frameworks with DNNs for models, and open source community projects using Autonomous SW stack. Anurag is responsible for designing and implementing CUDA kernels, and TensorRT plugins for custom layers with safety certifications for Autonomous Driving. He is also experienced in quantization and compression of DNNs for deployment on NVIDIA DRIVE platforms. Anurag has contributed to Torch-TensorRT (Ahead of Time JIT Compiler), an Ahead of Time JIT compiler to compile Torchscript model into TensorRT for NVIDIA platforms. He focuses on performance and runtime of model inference in full and low precision mode, designs and implements plugins for custom layers or add converters for non-supported operators, and is responsible for infrastructure aspects of projects. Anurag has worked as a Machine Learning Research Assistant and Intern at Mozilla, where he designed, implemented, and tested experimental code for several benchmark variations over the existing architecture of Abstractive Text Summarization project and DeepSpeech project. He has also contributed to the DeepSpeech project by fixing and addressing multiple issues reported on Github. Anurag holds a Master’s Degree in Computer Science from the University at Buffalo and a Bachelor’s Degree in Computer Science from UIET, Panjab University. He is proficient in Deep Learning, Software Engineering, Research Science, ML, Computer Vision, NLP, infra, Corpus, RNN, and Normalization. Anurag has contributed to open source projects like Torch-TensorRT, DL4AGX, TensorRT, onnx-tensorrt, DALI, and NeMo. He has also made contributions to research papers, patents, webinars, and GTC conferences.