Feifei Li (M.Sc.)

In the old days, physicians used statistical methods (frequency, mean and standard deviation, distribution fitting) to deal with experiment data and the regression model to testify to the medical hypothesis. These days, physicians collect massive training data to train the machine learning model, looking forward to AI that can give more accurate and efficient results. But AI is not the panacea. This seminar will provide critical thinking for machine learning methods. It will illustrate when and why they are working or not working. The difficulty of talking about computation has led to a lot of misunderstandings. The basic idea is that AI is good at some things and very bad at others, and social problems arise from situations in which people misjudge how suitable a program is for performing the task. A dialectical view of AI is to make better use of AI and avoid the risk of improper use.

This lecture will introduce the principle of Deep Learning, as an important branch of machine learning methods. It covers the fundamentals of deep learning, including neural networks, backpropagation, activation functions, and optimization techniques. After covering the basics, we will dive into popular deep learning architectures such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). We will explore the architecture of these networks in depth, and learn how to implement them using popular deep learning frameworks such as TensorFlow and PyTorch. Throughout the lecture, we will also cover practical considerations for training deep learning models, such as overfitting and regularization techniques. We will also discuss techniques for evaluating the performance of a deep learning model and tuning its hyperparameters to improve its accuracy. Prior knowledge related to python basics and data science are welcome but not necessary. This lecture will act as a preliminary role for the course of medical imaging process.