Deep Learning Resources and Study Path For Aspiring Data Scientist

As a follow up of my previous post (Link Below), one frequent question I got from fellow LinkedIn are for resources on Deep Learning

First of all I would not advice someone to directly jump on to deep learning without getting decent level of understanding with Math, Statistics and some basic machine learning algorithm. This is my view though, but it is better to follow with the learning path you are comfortable with

From courses standpoint I would definitely say Andrew Ng Deep Learning course on coursera is one of the best out there while fast.ai course is also an alternative

Let us now go in to some nice resources that are free and can help get your fundamentals strong in deep learning space

CS231n Stanford lectures are one of the best resources to learn Convolution Neural Network (CNN)

You can also watch the videos of CS231n class on youtube if you are interested. Videos have additional details (compared to blog above) covering Recurrent Neural Network (RNN), Object Detection and Segmentation, Reinforcement Learning and Generative models (Pick and choose videos that you really really really want to cover)

More specifically checkout detailed post written on CNN covering each layer of the architecture

Once you are done with above click link below to get an idea on how adding layers, linear and non linear activation functions etc effect the outcome using Neural Network Playground. It is fun to try various combination against different data

Below is a nice resource to understand cost function, optimization function (First and Second order) etc. You can also play with setting different learning rate and decay for various optimizer to see how different optimization function converges

Now coming to Recurrent Neural Network (RNN) and LSTM in specific, blog below is one of the best resources

Blog above also has posts for CNN and Attention models. You can take a look at it if interested

If at all you want to know math behind Back Propagation below blog has decent information to it

In case if you want to learn about Interpretability then below is good reference

Stop at this place. You can keep learning GAN, RL etc etc etc.. Learning is a never ending process but I feel at this point you are in pretty comfort zone to enhance your skills once on the job

One big question everyone like to know, Where can I practice Deep Learning and also How to get started coding for Deep Learning?

Let us jump into that now on how to practice what we have learnt. Deep Learning requires substantial resources and infrastructure to practice and ideally regular laptop might not be sufficient for it. At this point you can sign into cloud environment like Google and utilize their 300$ credit

One alternate option I am going to talk about is Google Colab which is free for use and also comes with GPU and TPU accelerated environment

One things to remember in Google Colab is to change hardware acceleration in notebook to make use of GPU or TPU

Go to Runtime -> Runtime Type and select GPU or TPU

Now that you have environment to code and run deep learning jobs, let us look at some examples and code that you can practice on

Go to Google Codelabs (this is codelabs not colab to be clear :) ) Tensorflow learning page

Examples in codelabs are guided step by step tutorial that you can follow along with Theory and Code. These are mostly using Tensorflow. For other Deep Learning framework you can do google search and find out. Start with below first, there are example build from scratch as well as using Keras. If you find from scratch difficult read through theory and skip it

That's it. Looks Short and Simple right :) .... I hear you :) .. Happy Learning

A quick note for Aspiring data scientist. Think what is you focus area?.. To becomes a Researcher or to get into Enterprise to solve business problem or into Technology companies like google, apple etc or to stand up products that you can market

Your focus area in AI space depends on it. If you are looking to get into enterprises (non technology companies like google etc), remember 90% of AI work in enterprise is Traditional ML and only 10% in Deep Learning. If you want to be researcher your focus might be different

Draft your learning path accordingly. Learning machine learning is like optimization algorithm traversing complex search space. Find the shortest path (not shortcut) that can help you achieve your goal