Machine Learning Foundations: Ep #7 - Image augmentation and overfitting

Machine Learning Foundations is a free training course where you’ll learn the fundamentals of building machine learned models using TensorFlow.

In Episode 7 we’ll look at how we can use image augmentation as a technique to artificially extend your datasets to provide new information for the training of your neural network. This can potentially help you with overfitting issues!

[sivaram]

Thanks a lot for this video series. Your articulation of over fitting and under fitting is very easy to understand.

[zainabkhan]

@Laurence please make a video on time series augmentation techniques. It will be a great help.

[RajaSekharaReddyKaluri]

Thanks!! Image Data generation on fly is very much needed!

[miroslavtisma]

I think you have a mistake in the video. That lab URL is from the one of the previous episodes.

[k]

This series has been very helpful for me to understand the concepts of how neural networks operate. Thank you for your hard work on this series. One thing I'm not quite clear on is the "activation=relu" significance. Why does every Conv2D need this, what does it actually do, are there alternatives, etc? Same goes for "loss=binary_crossentropy". When would someone use this over other loss functions? Are some better suited for other things? A video explaining these "magic" references would be most useful.

[sumkh]

url of the lab is not correct i think, pl mention the correct one

[tanmaypriyaagrawal]

Hey Lawrence.
I'm absolutely wonderstruck by your amazing series.
I wanted to make my own Image Processing Model using random images from the internet. Can you tell me what additional skills I need (like image processing etc.). Also I've done the Neural Networks and Deep Learning Course by Andrew Ng. The issue is that I find it pretty tough to use the knowledge from that course to make my own project. How can I better make use of that course?

[avijitchakraborty]

What is the best way to deal with overfitting incase of Text data considering image augmentation isn't an option here ?

[JudeGussman]

NOooo, give me an excise please! I love this series! Anyone have any suggestions for a neat Kaggle dataset?

[rohirrin]

Thank you! Very interesting. Image augmentation internally does modified copys of the original image? Or only randomly change every image of your dataset?

[Sensation]

Hey there, great series! Thanks a lot for making this public! Can anyone say, how many more vids there are going to be?

[koruvappaspandu]

Hi actually i have 100 images how can i do all images at a time?

[joanam]

Thank you for sharing this interesting info. It is not only about images, it makes you wonder the importance of teaching programming and understanding its influence in the future to come. The creation and programming of artificial intelligence must be democratized because it may influence on, for example, what a machine considers human. The bias on the actual police data, for example, it's clear. If a racist or classist society provides machines of biased data, the result is biased. Technology and politics are 100% related because they are human made. Interesting video indeed, thanks for sharing your knowledge

[DavinderSingh]

Why Python! :(, I wish they did series with javascript