About Me
Haoyu Sun who got MAI (Master in AI) degree in Illinois Institute of Technology (IIT), he was born in 1996 in Inner Mongolia, and work with Dr. Yan Yan (Tom Yan) who is currently a Gladwin Development Chair Assistant Professor in the Department of Computer Science at Illinois Institute of Technology.
The main courses are: deep learning (computer vision), machine learning, NLP, probability graph models and bioinformatic etc. You can feel free to check the AI core course in IIT website .
GPA > 3.65 and the research directions are subdivided into: image detection and segmentation, bioinformatics analysis, and reliable intelligent decision-making framework.
Projects
Use a less complex linear transformer to achieve classification tasks with comparable accuracy to transformers.
In this project, we combine the existing traditional models and apply the new model to the MRI image detection of Alzheimer’s disease to achieve the identification of different degrees of pathological stages. This project compares the characteristics of the old and new model structures and conducts in-depth experimental tests. On the basis of testing, we integrate the idea of improving the prediction accuracy of the model, and design the block-CNN structure to improve the accuracy of the traditional CNN model structure. At the same time, the attention mechanism of linear complexity is applied to the computer vision classification task, which improves the efficiency and accuracy of the model.
Research on the solution of domain drift problem in computer vision.
This project mainly solves the domain drift problem of training and test sets and real-world engineering data, reduces the training cost of model engineering application and the workload of feature marking, and has far-reaching engineering application significance.
Reinforcement learning intelligent decision framework.
Using machine learning models and reinforcement learning to build a framework to predict pavement technical conditions, disease development prediction and intelligent recommendation of maintenance project implementation plans, the project has begun to be promoted in Xinjiang Autonomous Region of China and is ready to enter the product development stage
Experience
Responsible for the course: Intro to AI Assisting faculty with classroom instruction:
1, leading discussion sections.
2, meeting with students during office hours.
3, grading assignments/exams, proctoring examinations, and providing feedback on assignments.
Research Institute of Highway Ministry of Transport (RoadMainT Co Ltd)
Research And Development Engineer
3/2022 - 1/2023
1, Research and development pavement crack recognition based on computer vision (Team role: algorithm design):
• Established the DensNet121 network to identify transverse cracks, longitudinal cracks and cracks. According to the actual production demand we modified the activation function of the last layer, except this, the structure of network and loss function are partially modified.
• Established the Mask-RCNN based on mm detection framework to identify the cracks and large area repairs through the front image taken by driving recorder.
• Tested the different performance of different model on road image recognition.
2, Research and development based on the classification and quantity of highway pavement diseases to predict highway structural diseases (Team role: Leader):
• Established the classification algorithm based on the highway pavement diseases to predict the probability of structural disease in each road section. Then based on the result of predict, according to the actual production, visualization results are used to provide scientific decision-making for highway management departments.
Research Institute of Highway Ministry of Transport (RoadMainT Co Ltd)
Research And Development Engineer
7/2019 - 3/2022
Predict on pavement performance prediction and optimal allocation model of funds based on machine learning (Team role: leader):
• Use the dataset of traffic volume, capital, climate, pavement performance and department production design a pavement performance prediction model (GM-LR model), It combines grey prediction and linear regression model, It has been applied in Xinjiang, Inner Mongolia and other provinces and has good results.
• Established the multi-objective decision-making model according to the data of distribution of concept road maintenance funds and the actual benefit.
Education
Illinois Institute of Technology
Masters in AI
09/2021 - 12/2023
Deep Learning, Machine Learning, NLP, Bioinformatics Analysis, Algorithm Design, Databases, Advanced AI.
Inner Mongolia University of Technology
Bachelor of Information and Computing Science
09/2015 - 7/2019
Mathematical statistics and probability theory 、mathematical analysis、Advanced algebra、 discrete mathematics、ordinary differential equation、Convex optimization、MYSQL、Python、R.
Publishes
With the rapid development of global road transportation, countries worldwide have completed the construction of road networks. However, the ensuing challenge lies in the maintenance of existing roads. It is well-known that countries allocate limited budgets to road maintenance projects, and road management departments face difficulties in making scientifically informed maintenance decisions. Therefore, integrating various artificial intelligence decision-making techniques to thoroughly explore historical maintenance data and adapt them to the context of road maintenance scientific decision-making has become an urgent issue. This integration aims to provide road management departments with more scientific tools and evidence for decision-making. The framework proposed in this paper primarily addresses the following four issues: 1) predicting the pavement performance of various routes, 2) determining the prioritization of maintenance routes, 3) making maintenance decisions based on the evaluation of the effects of past maintenance, and considering comprehensive technical and management indicators, and 4) determining the prioritization of maintenance sections based on the maintenance effectiveness and recommended maintenance effectiveness. By tackling these four problems, the framework enables intelligent decision-making for the optimal maintenance plan and maintenance sections, taking into account limited funding and historical maintenance management experience.
Pavement performance prediction method based on GM-ANN model (Chinese Version:基于GM-ANN模型的路面性能预测方法)
Hai Zhang, Haoyu Sun, Yuqiang Wang. (张海,孙浩宇,王宇强)
English:
Pavement performance prediction is the core technical difficulty in highway pavement management system, which has been limited by various factors such as analysis methods, insufficient data volume and low data dimension, resulting in a large deviation between the predicted value and the measured value of several existing performance prediction models. The pavement disease is analyzed by gray system (GM) in time series, and then the prediction value of various diseases is used as input, and the pavement condition index (PCI) is indirectly predicted with the help of artificial neural network (ANN), and a hybrid model based on artificial intelligence (GM-ANN) is constructed. The results show that the hybrid prediction model based on GM-ANN has better accuracy and operability, which can provide a more accurate and reliable reference for big data maintenance decisions in practical engineering applications.
Chinese version:
路面性能预测是公路路面管理系统中的核心技术难点,一直以来受限于分析手段、数据体量不足及数据维度低等多种因素影响,导致现有几类性能预测模型的预测值与实测值偏差较大.将路面病害以时间序列进行灰色系统(GM)分析,后以各类病害预测值为输入,借助人工神经网络(ANN)间接对路面状况指数(PCI)进行预测,构建基于人工智能的混合模型(GM-ANN).最后选用地区随机路段进行实例验证,并同常用模型预测结果进行对比分析.结果显示:基于GM-ANN的混合预测模型更具较好的精度及可操作性,在实际工程应用中,可为大数据养护决策提供更准确、可靠的参考依据.
Patents
一种道路标牌及安防设施缺失的检测方法、介质及系统 (English:A detection method, medium and system for the absence of road signs and security facilities)
张海,孙浩宇,王宇强 (Hai Zhang, Haoyu Sun, Yuqiang Wang.)
202211318065 · Issued Apr 6, 2023
Honors, awards and certificate
Licenses & certifications
About me
Know more about me? Fell free to contact me!