Hacker News Reader

Comment by nerdy | Hacker News Reader

AI is very interesting but not very accessible because it's so specialized. I have a fairly strong programming background but feel like I'd need to study theory for a significant amount of time to even get my feet wet with AI.

If you have (condensed, especially) AI resources that you think would help bridge that gap, please share! Toy-scale project ideas would also be appreciated.

View on HN

Buy and work through "Artificial Intelligence: A Modern Approach". It's a huge book and the de facto standard for pretty much every AI 101+ course. Some of the stuff may not interest you some might but it covers a broad range (from logic based agents to Bayesian networks). It's systematic and has excellent references and further reading notes for each chapter. The focus is not on the currently sexy "data science" aspects though (however you will find plenty of material that is relevant).

The edX class from Berkeley is pretty fun and hands on. It uses Pacman as a running example and essentially teaches the agents stuff from AIAMA:

https://www.edx.org/course/artificial-intelligence-uc-berkel...

The Stanford class by Thrun and Norvig himself (one of the authors of AIAMA) is also good but I prefer the edX one:

https://www.udacity.com/course/intro-to-artificial-intellige...

Edit: changed to direct links for the courses

The AIMA book is sort of a Good Old-Fashioned AI (GOFAI) book that focuses a lot on agents and planning. The jobs this article is talking about are really machine learning ones-- taking large volumes of data and extracting knowledge, so as to build recommender systems and such. For that, Kevin Murphy's book, "Machine Learning: A Probabilistic Approach" is without a doubt the best book out there, both in terms of explaining things from the ground up and being the most comprehensive/up-to-date source.

There's still quite a bit of material on Bayesian networks (with the dreadfull dentist example :D), neural networks and support vector machines but overall you're right the focus is on agents. The relevant chapters are great staring points though and as always filled with great reference material for further reading.

+ I'm pretty sure if you apply for an AI job somewhere and it's labaled AI and not "data science" they'll expect that you know the material in AIAMA.

Murphy's book is actually subtitled "A Probabilistic Perspective" -- "Machine Learning: A Probabilistic Approach" is a different book by a different author.

+1 for the Stanford course. Great intro to AI and super easy to follow - I've done it after my uni class elsewhere, and it helped to internalise what I've learned there.

reading the book still requires a significant amount of time

'AI' is not programming, it's mathematics (well the current 'flavor' of statistics-based AI, that is - the 1980's style AI people I used to work with were philosophers, legal scholars and the like). Anyway, there is no 'bridging the gap' - you need to start from a good foundation of statistics (and the 'prerequisites' - algebra, calculus, linear algebra) and in the end, the technicalities of the software and the theory come together naturally.

(source: have tried to 'bridge the gap' for 2 year, including taking MSc courses, before admitting to myself that it's a lost battle. Am now starting to build a solid math foundation before revisiting ML applications.)

This is exactly my experience as well. I'm an alright programmer, but it is insufficient, because machine learning and AI are a table resting on four legs: linear algebra, calculus, statistics, and programming. I've also found myself going back to build up those foundations.

There was a tutorial link trending on HN a few months back, I can't seem to find it. But these links are helpful (though not condensed) as well:

https://github.com/ChristosChristofidis/awesome-deep-learnin...

https://github.com/owainlewis/awesome-artificial-intelligenc...

These links can be helpful too:

https://news.ycombinator.com/item?id=7783550

https://news.ycombinator.com/item?id=9432952

https://news.ycombinator.com/item?id=2645671

https://news.ycombinator.com/item?id=5508261

https://news.ycombinator.com/item?id=9004689

Here are two tutorials I've written on natural language understanding:

https://honnibal.wordpress.com/2013/09/11/a-good-part-of-spe...

https://honnibal.wordpress.com/2013/12/18/a-simple-fast-algo...

Reply on news.ycombinator.com