Machine learning is widely used in many applications including autonomous vehicles, robotics, and medical diagnosis. Recognition of an image is one of the best examples of machine learning or artificial intelligence. Topics of the image recognition course includes fundamental components of deep learning such as convolution layer, full connection layer, pooling layer, ReLU layer, and a softmax layer. In this course, students develop and train their network with matlab by themselves.
Students are expected to
(i) gain an ability to build and learn deep neural networks,
(ii) gain an ability to use numerical computing environments using MATLAB to solve engineering problems,
(iii) gain practical skill to apply the deep learning techniques such as momentum, data arugumentation and filter setting, after taking this course.
|A faculty who has a private company experience gives a lecture.|
Object recognition, Convolutional neural network (CNN), Deep learning, matlab
|✔ 専門力||✔ 教養力||コミュニケーション力||展開力(探究力又は設定力)||✔ 展開力(実践力又は解決力)|
This class is a kind of active learning. Instructor will give some information, but students are required to develop their matlab code.
|第1回||Introduction of course, cifar-10, and sample network, GUI||To run the sample network|
|第2回||One hot encoding, softmax, MSEloss, KL-divergence, cross-entropy loss||To compare MSE loss and cross-entropy|
|第3回||Convolution and activation, ReLU, LReLU, PReLU||To compare ReLU, LReLU and PReLU|
|第4回||Pooling layer and dropout||To compare different parameter of dropout|
|第5回||Merge layer and batch normalization||To try Batch normalization|
|第6回||Optimization and data augumentation||To compare sgd and adam|
|第7回||Parameter tuning 1||To try parameter tuning|
|第8回||Parameter tuning 2||To try parameter tuning|
Assignments and report
Students need to implement the matlab code by themselves.
Student who took the course of Computational Imaging (#SCE.I501) cannot take this course.