Udacity Deep Learning Nanodegree Experience
Recently I graduated from Udacity Deep Learning Nanodegree I program .Here I am going to share my learning experience .
Deeplearning foundation teaches about Neural Network and Convolutional Neural Network ,this is part of a two part series the second part teaches RNN and further advanced Neural Net technology.As a prerequisite of this course one must have basic python programming,numpy,pandas and good to have knowledge coding in Jupyter notebook although you can pick up these as you proceed .
Syllabus for this course divided into 3 parts
- Introductions — This section talks about matrix multiplication,setting up anaconda and jupyter notebook , also about Regression as general. There is no project in this section.
- Neural Network- This is a core or backbone of this course .This section comprises of topic like Neural Network definition,Gradient Descent,Backpropagation,Forward pass,Perceptrons,Multilayer perceptrons. One of the most important topic is understanding the Backpropagation and weight updation ,it’s ok if you do not get all of the maths but conceptually one should understand what is going on .Good thing about this section is most code is done using only numpy,pandas no tensorflow involves here so you get to see what is actually going on implementing a Neural Network rather than just calling some black box API. This section also contains Youtube videos from Siraj ,but the section I like the most is Sentiment Analysis using only numpy and pandas by Andrew Trask . I like it more than the videos this is comprised of 6 mini projects. This section contains a project that need to be completed in order to graduate it is predicting the bike sharing number of NY city . The process is you will have to fill up ToDo section with code. Which is kind of fine for first timer although I’ll love to do this ground up not only todo section. Initially I experienced some hiccups submitting my project but that I was quickly resolved and Udacity also provided me a grace time to complete my course.
3. Convolutional Neural Network- This section is where I spent a lot time understanding the basic of CNN . With the videos you need to read the articles and links shared in the course to understand the theory . Tensorflow is the tool that is being used to describe the concepts and exercises.This section has one project which is Image Classification with CIFAR-10 dataset.To complete project in this section one needs GPU based machines .Udacity provides a $100 credit for AWS.Also there is option of FloydHub.I personally liked FloydHub you just upload dataset and you can run your jupyter notebook backed by GPU .
This course is of 2 months and the topics that are covered cannot be done in depth in this time frame but this will atleast give one the start needed to further explore the topic. Because the core concepts are represented using simple python and numpy,pandas helps one to understand the concept rather than just calling APIs . Overall if anyone just getting started this course is a good choice to start with .After that one can start applying the knowledge with Kaggle dataset and explore material available in internet .
Apart from the course material there’s a slack channel available where one can interact with fellow students and mentor. Which is good support in case you are stuck at something .
**As of now Udacity has stopped offering the course in two half rather there is only one course of 4 month duration.
Here’s my github repo link with projects
