简历 - Aoxin Blog

Personal Information

Master, Beijing Jiaotong University, Control Science & Engineering, 2019.09~2026.7.
Bachelor ,Beijing Jiaotong University, Computer Science & Technology, 2019.09~2023.07.
Passed CET6/4.

Research and implementation of object detection method for remote sensing images based on deep learning：The project collects and processes the military target dataset of remote sensing images, uses yolovS for training, and optimizes the characteristics of remote sensing images, adding smaller anchors, adding attention mechanism modules, etc. Use PYQT5 to design a simple front-end and complete a graduation thesis
The inference part of the UNET segmentation model is accelerated：The project uses the mm framework to complete the training of the Unet network on the CityScapes dataset, uses the Qdrop offline quantization strategy to quantize the weight file, simulates the inference part on matlab, and finally transplants the simulation process to the OpenCL-based FPGA heterogeneous server for acceleration using C language
Super-resolution image inference implementation of OpenCL heterogeneous inference platform：On the basis of the FPGA inference framework in the laboratory, the super-resolution image inference function is added.
Convolution operation acceleration based on OpenCL：Based on OpenCL, HLS high-level synthesis is used and Xilinx’s Vitis simulation tool is used to compile to accelerate the convolution operation.

Baiyang Times，Product Development Team，Machine Learning Engineer (MLE) Intern，2023.11~2024.5
- Job Description：Job Description: Engaged in product development with the team. Contributed to the research and development of a deep learning platform and a LLM application platform during the tenure.
- Deep Learning Platform: In the development of the deep learning platform, I am responsible for the development of the algorithm part, the first part provides multiple object detection, target segmentation algorithms and corresponding datasets, and the second part is to realize the back-end of dataset preprocessing, and realize the data processing module with multiple formats of data reading, data analysis, and dozens of preprocessing methods.
- LLM Application: The first part implements the development of a Prompt-based large language model application using the COZE platform, and the second part implements the fine-tuning using the Baichuan2-based model.

An inference engine based on the PNNX model implemented in the C++ language：#########################################################