Quick Look at Foundations of AI term of Udacity Artificial Intelligence Nano Degree.
Info graphic below gives a birds eye view of the program planned for the “Foundation of AI” term of Udacity AI Nano Degree.
You can download the info graphic from this gist.
https://user-images.githubusercontent.com/4714062/37488795-f7b29b6e-2852-11e8-9654-f842c0774017.png
For a couple of reasons, I was waiting for this course to open up. It did on March 1st.
- Passionate industry leaders like Peter Norvig and Sebastian Thrun are involved in designing this course.
- There are different definitions of AI provided by leaders in the industry and academia. Eager to see what this course would offer.
After the first quick run of the complete course, if I had to summarize this course in one phrase, it would be “It is all about developing Rational Agents.”
One key observation is, that this course goes at warp speed. If you know the math, you will enjoy this course. Else you may not like the pace, and you might get lost. It is not about teaching you all the concepts in detail. It is about artfully connecting the dots and helping you understand how to apply AI to solve real-world problems.
The course starts off with developing a game of Sudoku and then picks up speed. It explains what constitutes a Task Environment, how to represent the states of the world, ways of searching for a solution by the problem-solving agent. Make a case for moving beyond the classical search. Quickly jumps to explaining the difference between a planning agent and the problem-solving agent.
I was binge watching this course past week. The way all the concepts are weaved together smoothly makes it compelling.
Thrun himself, explains about probabilities, introducing Bayes theorem,and using examples explains Bayes Nets starting from the point of calculating the joint probability, conditional independence, conditional dependency, what happens an event is observed? , d-separation, finally moving toward Bayesian inference and also introducing the role of sampling, mcmc and Gibbs sampling. All this masterfully tying up in one stretch.
Finally, Thad Starner talks about “pattern recognition through time.” He jumps right into Dynamic Time Warping and switching it to HMMs for their easy of training. Calculating the transition probabilities and building the lattice for American Sign Languages. Goes on explaining how to build the HMM Topologies for phrase level recognition.
Before this course, I knew dynamic programming, tree search, graph search, propositional logic, first-order logic, Bayes Nets, Hidden Markov Models in different contexts and never related it to AI. Still had a question “What is AI? How is it different from ML and DL? Never got that big picture, Never connected the dots.
After this course, I hope AI would make more sense. I am expecting to be in a position to develop Rational Agents to solve enterprise problems. Hopefully develop agents (software tools and solutions) which not only solve the problem but also can keep learning from the environment and keep improving.
AI is powerful, and no doubt this course will provide the guidance. It is also true that this course needs lot more commitment on my part.
