Hi, I'm Afshin. I'm a machine learning engineer! I'm finishing my masters degree in applied science at the Univeristy of British Columbia right now. My research has been focused on infrastructure management in transportation engineering using deep-learning models.
I finished my bachelor's degree in computer engineering at the Ferdowsi University of Mashhad. For the past 4 and half years, I have been learning about and working on deep learning models in various capacities. I worked as a machine learning engineer in a FinTech startup before starting my master degree. You can find the full information on my background, experience, and education in my website or resume and CV below!
View Academic CV View Professional Resume
Served as a Teaching Assistant in different capacities for several
courses, with responsibilities ranging from teaching lab sessions,
in-class assistant, invigilation, exam and assignment grading, and
holding problem-solving sessions for students, for classes with
more than 100 students. Gained a satisfaction score of at least
90% for the classes
List of classes:
Worked on Generative Networks (DCGAN, WGAN, Conditional GAN) to
create artificial road image data, increasing crack classification
accuracy by 15%.
Designed and worked on road infrastructure object detection and
classification pipelines using images and 3D point clouds.
Researched & worked on price and trend prediction, and
decision-making models on cryptocurrency data.
Developed Deep Reinforcement Learning agents and environments to
automate the trading process, with online data processing and
agents achieving up to 3000% profits even with low training
samples using Transfer Learning
Performed various data extraction, information crawling &
transformation tasks on data using techniques to generate valuable
features from raw data.
Created Anomaly Detection and Forecasting systems for electrical
power consumption using Deep Learning and Data Analysis, as
software for the power utility company of Mashhad, with more than
92% accuracy in fraud detection and 94% accuracy in trend
prediction for power consumption.
Published 2 peer-reviewed conference and journal papers.
Teaching assistant for several courses, with responsibilities
including teaching in class, assignment creation and grading, exam
creation and grading, teaching labs, project defining and grading,
and coding assistant.
List of classes:
This project uses Deep Convolutional Generative Adversarial Networks (DCGAN), Wasserstein GAN (WGAN), WGAN with Gradient Penalty (WGAN_GP), and Wasserstein Conditional GAN (cGAN) with Gradient (C_WGAN_GP) Penalty, to generate synthetic images of pavement cracks. These images can be used to augment existing datasets, improve the robustness of machine learning models, and facilitate research in pavement maintenance and repair.
Deep Learning object detection models, including YOLOv5 and Faster R-CNN, were used to detect different objects, such as traffic signs, traffic roads, pavement lane marking, and pavement cracks, in road images. The images were gathered from Canadian roads and public datasets, and were annotated by us using CVAT to be compatible with YOLO and COCO formats.
Developed several deep reinforcement learning agents and environments on cryptocurrency trading data based on custom reward functions, with long-term and short-term benefits prioritization, and multi-currency trading. Used Transfer Learning, and data preprocessing to trade currencies with similar patterns without extensive historical data. The agents showed consistent profitability even in market volatility in various timeframes with 3000% cumulative profits in 2020-2022.
Used deep learning models, such as Autoencoders, RNNs, MLPs, and data analysis techniques, like missing value interpolation, clustering, decision trees, to develop Trend Prediction and Fraud Detection software and website. Preprocessing steps including data normalization, standard scaling, and outlier removal, are performed on the data. The trained deep learning models, data analysis techniques and preprocessing pipelines were incorporated into 2 software, sold to Mashhad Electrical Utility Company, with more than 92% accuracy in both tasks.
Published in Electric Power Systems Research", Volume 208, 2022, 107895, ISSN 0378-7796
Published in 2021 5th International Conference on Internet of Things and Applications (IoT), 2021, pp. 1-6
[Get In Touch With Me!]
My E-Mail