Tuesday, July 28, 2015

Done! For Now...

Sometimes progress is hard to come by. Right now, for instance, I am trying to learn something new, specifically a Python web module known as flask but running it on a ipython notebook which is probably complicating matters. Maybe I should just make another sandbox to test these things with…

Anyway, on to other more interesting things while I wait for the program to complete or fail: I turned in my final for Knowledge Based AI for Georgia Tech’s Online Master's in Computer Science. It was actually quite an interesting class. We had three projects over the course of 3 weeks looking to solve visual IQ tests using cognitive techniques.

I certainly wasn’t the best in the class by a long shot, but I know I got a pretty solid B in the course and am hoping that I somehow squeaked an A, though if I don’t get one I will be close. I really enjoyed the challenge of the class and most of the time I enjoyed writing the papers too. An interesting tool the course used was a peer feedback system for the papers. Some feedback was less than useful, but there was certainly quite a bit of it that was useful.

The topics looked into how we could model AI off of human intelligence, rather than just building a logical and/or super fast machine, think Deep Blue: brute force chess playing works, but it really isn’t the way most people would approach chess, not even the masters. No one can consider the next million possible moves in a few seconds.But even if there are approaches to how to do something that make a computer behave a bit more like a human, it still isn’t the generalists that we are. Even the most strictly uni-tasked human can generally figure out how to do something new, and we are only beginning to delve into those depths of general learning. And hopefully Empathy, but I will leave it at that.

I didn’t feel as though I got my agent to be an AI much. There was quite a bit more image processing going on than trying to follow all of the different concepts we were introduced to. It might be interesting if each of those topics had a more in-depth set of materials to mess around with and code a small agent using all of those materials. That does sound like a possible project, to make mini-agents that incorporate each concept separately and only include other concepts that are needed to facilitate the actual workings of the agent. Of course that is possibly a great undertaking… and I am mostly just full of good ideas and distractions.

For instance, I was introduced to an interesting thought just now: http://nytlabs.com/projects/editor.html that could have huge effects on the osf.io wiki, by connecting it to SHARE, allowing the piecemeal typing to be sent to an agent that would use previously learned references and list them to the side of the wiki editor, so that the person writing a wiki page might see similar published items. These could also be split, sort of as the article describes, into smaller agents.

However, my agents might be grouped into science categories so that a chemist gets chemical references, and not psychological references. That could lead to boxing results a bit too closely, so there might be agents that cross disciplines so that discoveries in other disciplines might be helpful to the current project. But of course this “distraction” would have a dedicated “9-to-5” time to work on it.

Another problem with free time projects is that we just moved into our new house. It is huge, at least compared to the last, well, long while. Unpacking is nearly done and mostly waiting on Jessie to get back, but there are plenty of things to do.

One of the more fun things will be making an archery target. With all the empty boxes I have started doing triage to find the worst of them and cut them up to make the target. We also have way too much packing paper and grocery bags. Moving from Hawaii, where they were banned, to Virginia was quite a shock as to how many we accumulated over 6 months. The bags will make pretty decent fill though.

Also I have been working pretty much eight to sixteen hours extra a week on the class, no real down-time. Some projects, especially on my own will have to far away from code so I can mentally refresh and prepare for the Machine Learning class that starts at the end of August! Oof.

Maybe I should just keep the nose to the grindstone in order not to miss the fleeting down-time.

No comments:

Featured Post


John studied himself in the mirror as best he could through tears. Red, puffy eyes stared back at him, a running nose already leaked just a ...