Until now, I have taken Machine Leaning By Dr. Andrew Ng of Coursera, and the famous [Deep Learning from scratch](https: //). I read www.oreilly.co.jp/books/9784873117584/) and wrote the code according to the book.
However, it was unclear what to start when creating and implementing a new Deep Learning network. At that time, I found the following course and took it, which was better than I expected. Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning
--There is almost no mathematical explanation --Background knowledge is required for machine learning --The content is quite introductory --Working with Jupyter Notebook --There is also a link to Google's colab etc., and you can work on it, so you do not need to build an environment --You can also submit assignments from Jupyter Notebook on Coursera
--Tensorflow / Keras is now available --You can now add layers of the network and try and error yourself.
Week 1. A New Programming Paradigm
Explains the difference between machine learning and regular programming. Pretty basic content.
In the task, I solved the regression problem using Tensorflow.
Week 2. Introduction to Computer Vision
Solve classification problems using Fashion MNIST data. This week, we will not use CNN and solve the image data as a single line vector with Dense Neural net. In this process, I learned how to configure a network using Keras.
You will also learn how to call a callback after epoch is finished. As a result, learning can be completed when the accuracy exceeds a certain level.
In the task, I solved the usual MNIST classification problem.
Week 3. Enhancing Vision with Convolutional Neural Networks
Solve the Fashion MNIST classification problem on Week 2. with CNN. Learn a very brief description of CNN and how to actually implement it in Tensorflow / Keras. I deepened my understanding by visualizing the results of Convolution. In addition, I was able to experience how it affects accuracy and the speed of learning and inference by trial and error, such as changing the number of Convolution layers and excluding Max Pooling.
Week 4. Using Real-world Images
Apply what you did in Week 3 to a real image (M pixels, color images, objects not in the center). Specifically, the classification of "people" and "horse" images. The following was especially helpful for visualization of the output of the intermediate layer.
As written in reviews, the content and exercise are certainly easy. Coursera's Dr. Andrew Ng's Machine Leaning And Deep Learning from scratch / 9784873117584 /) Isn't it a good next step after learning?
Recommended Posts