Overview of Udacity Artificial Intelligence Engineer Nanodegree, Term 1
After finishing Udacity Deep Learning Foundation I felt that I got a good introduction to Deep Learning, but to understand things, I must dig deeper. Besides I had a guaranteed admission to Self-Driving Car Engineer, Artificial Intelligence, or Robotics Nanodegree programs.
Step back…
Before I turn to Udacity advanced courses, I want to mention one thing at the beginning. If I could give advice to myself, I would select another introduction course on Deep Learning — Deep Learning Specialization by Andrew Ng. First of all, his way of mentoring is unique and he can explain complex things in most clear and understandable way. Second — it is cheaper than Udactiy and, I had an impression, you get more for your money.
Another very good alternative is Practical Deep Learning for Coders by Jeremy Howard. It is for free and it covers the same aspects as Udacity Deep Learning Foundations.
…Let’s continue
I decided to stay at Udacity learning platform and take one of its advanced courses. To me, the most interesting courses were Artificial Intelligence and Self Driving Car Engineer.
Self Driving Car Engineer
I found a lot of good blogs and posts about this program. My favorite is ones from Mithi — check this post, it is really brilliant.
I have an impression that the course is very interesting and demanding, but I wouldn’t have good chance to find a “Self Driving Car Engineer” job here in Germany after finishing this course. Yes, there are BMW, Daimler, Volkswagen, and Audi, but, to my knowledge, all of them either require strong working experience or prefer fresh graduates from “real” Universities. They don’t recognize online courses — they are not established (yet).
Other thoughts — from my working experience, I know that a software can behave itself differently in simulators and on the real hardware. To me develop a software for cars without a possibility of testing it extensively is somewhat questionable.
Artificial Intelligence
Surprisingly I found very few reviews on Artificial Intelligence program, which made me a bit scared. From the other side, the content was very promising to me — I wanted to learn not only modern neural networks and deep learning techniques and tools for them, but also basics. In the end, artificial intelligence is not just a deep learning, it is older and give really good results without requiring so much data as neural networks do. So I decided to enroll in the course.
Artificial Intelligence Engineer program, structure
Udacity Artificial Intelligence Engineer Nanodegree consists of 2 terms, each of them takes 3 months to complete. It is not self-paced, you have deadlines to finish the projects. It is not for everyone, but for me, it is a perfect way to learn and stay motivated.
You can take only Term 1 and skip Term 2, but not another way around. You can enroll into Term 2 only after successful completion of Term 1. You get the certificate only after finishing both terms. Each term costs 800 $, which is definitely not cheap.
You have a dedicated mentor to provide you support if you have troubles or give you a direction if you are lost. The community is really great! Either on Slack or forums, you clarify anything you need, you will never get stuck :-)
As a Nanodegree student you also have access to Udactiy Career Portal, where you get valuable information, links to resources, advise and check your online profiles and resume.
Term 1, Foundations of AI
Term 1 called “Foundations of AI” and teaches you basics and techniques that were known quite before the popularization of deep learning and neural networks. Excluding common materials (setup, introduction, etc.), it consists of 3 sections and each has a final project at the end. Each section, in turn, consists of 4 to 5 topics.
Video lecture structure and tutors
The video lectures are structured as usual for Udacity — a number of short videos per topic with explanations, small quizzes in-between and links to supporting resources at the end. Some topics also have optional mini-project at the end. I highly recommend to complete them, though it requires time.
Most of the lessons are taught by Thad Starner from Georgia Tech and Peter Norvig from Google. They are very good by explanations, but not as good as Andrew Ng — this is my personal opinion.
Section 1. Game playing agent
In section 1 you learn such algorithms and techniques as Minimax, Iterative Deepening, Alpha-Beta pruning to play games, as well as build 2 AI agents to solve Sudoku and play Isolation board game.
I found the material was not good explained and without supporting resources it would be difficult to understand it.
Section 2. Search and planning
In section 2, first, you learn different search algorithms, such as Depth-First Search, Breadth-First Search, A-Star search. Next topics are space exploration and avoiding of local optima, followed by constraint propagation and logic and reasoning. The last topic is using previously learned materials to solve a planning problem. In the final project, you create a planner to move cargos from the origin airport into destination ones.
In this section, the explanation was pretty good and I used the supporting materials mainly to dig a bit deeper as required.
Section 3. Probabilistic logic
The last section is about probability, Bayesian Networks, and Hidden Markov Models. In my opinion, it was the most interesting section from all three. Hidden Markov Models is a specialty of Thad Starner and that is reflected in the explanation quality — it is perfect. I really enjoyed by working on the final project, gesture recognition. And this section was to me, as a compensation of a weak game playing agent.
Supporting materials
The video lectures are based on “Artificial Intelligence: A Modern Approach” book, but you do not see so many formulas, rather the tutors give you an intuition, how the things are working. Anyway, I highly recommend to buy or download it — it helps a lot.
Time consumption
It took me between 10 and 15 hours per week to go through video materials, reading supporting materials, solving problems and working on projects. Before I started Term 1, I had fear to work up to 20 hours per week on the materials, luckily the time consumption was fairly acceptable.
Impressions
Through the lectures, I had a good feeling that I am really studying the foundation of the Artificial Intelligence. The materials are good structured and overall have a good quality. I definitely learned what I initially wanted.
Term 1 is a good preparation for studying the neural networks, which are the topic of Term 2 and I am quite optimistic about the next term.
Will this knowledge help to start a career in Machine Learning area — I am not sure. Projects were interesting and demanding, but they are not enough. You need practice more in solving real-world problems with learned techniques and algorithms… Or you can start applying them at your current job :-)
The Term 1 costs 800 $ and it is not cheap. I am still not sure, whether it was worth it. It is true, that you get the practical experience and good feeling, how fundamental algorithms and techniques work. But if you do not apply the knowledge at work, then it is a waste of money.
And at the end of my story — you can find all completed projects (started with aind-) at my GitHub. And here you can find more resources on AI and beyond.
Update 16. February 2018
I have completed the Term 2 of the Artificial Intelligence Nanodegree and summaries my experience here. Please take a look if you are interested in.
