About this Course
In case you are a software program developer who needs to construct scalable AI-powered algorithms, you want to perceive tips on how to use the instruments to construct them. This course is a part of the upcoming Machine Studying in Tensorflow Specialization and can train you greatest practices for utilizing TensorFlow, a preferred open-source framework for machine studying.
WHAT YOU WILL LEARN
Deal with real-world picture knowledge
Plot loss and accuracy
Discover methods to forestall overfitting, together with augmentation and dropout
Study switch studying and the way realized options might be extracted from fashions
SKILLS YOU WILL GAIN
- Inductive Switch
- Machine Studying
Syllabus – What you’ll be taught from this course
6 hours to finish
Exploring a Bigger Dataset
Within the first course on this specialization, you had an introduction to TensorFlow, and the way, with its excessive degree APIs you could possibly do fundamental picture classification, and also you realized just a little bit about Convolutional Neural Networks (ConvNets). On this course you’ll go deeper into utilizing ConvNets will real-world knowledge, and study methods that you should use to enhance your ConvNet efficiency, significantly when doing picture classification!In Week 1, this week, you’ll get began by taking a look at a a lot bigger dataset than you’ve been utilizing so far: The Cats and Canine dataset which had been a Kaggle Problem in picture classification!
5 hours to finish
Augmentation: A method to keep away from overfitting
You’ve heard the time period overfitting various instances up to now. Overfitting is solely the idea of being over specialised in coaching — particularly that your mannequin is excellent at classifying what it’s skilled for, however not so good at classifying issues that it hasn’t seen. With a purpose to generalize your mannequin extra successfully, you’ll after all want a larger breadth of samples to coach it on. That’s not at all times doable, however a pleasant potential shortcut to that is Picture Augmentation, the place you tweak the coaching set to doubtlessly improve the variety of topics it covers. You’ll be taught all about that this week!
3 hours to finish
Constructing fashions for your self is nice, and might be very highly effective. However, as you’ve seen, you might be restricted by the info you will have available. Not all people has entry to large datasets or the compute energy that’s wanted to coach them successfully. Switch studying can assist clear up this — the place individuals with fashions skilled on massive datasets prepare them, so that you could both use them instantly, or, you should use the options that they’ve realized and apply them to your situation. That is Switch studying, and also you’ll look into that this week!
4 hours to finish
You’ve come a great distance, Congratulations! Yet one more factor to do earlier than we transfer off of ConvNets to the following module, and that’s to transcend binary classification. Every of the examples you’ve accomplished thus far concerned classifying one factor or one other — horse or human, cat or canine. When shifting past binary into Categorical classification there are some coding issues you want to consider. You’ll have a look at them this week!