Last week I enrolled in the beginners machine learning course at Coursera taught by Andrew Ng and have just finished week one.
Having not done any Coursera courses before, it was interesting to see their approach to learning on-line. I particularly appreciate the fact I can get good quality teaching for no money – unless I want the certificate.
Andrew seems to know what he is talking about, although this first week was pretty much all about maths – stuff I haven’t looked at since university days when I was a maths and computer science graduate. It was a bit of a grind winding back the clock all those years, but it came back quickly enough. I feel sorry for someone who hasn’t got a maths background chugging through those lectures. Considering I feel like I have used almost none of the stuff I learned at university, it makes me wonder why I went. Mostly for social reasons I suppose – and for getting my first job.
So far linear regression seems just like something you can do with formulas (for the simple cases) and numerical analysis for the difficult ones. I have done both – and used something similar to MatLab back at university too. I wasn’t expecting there to be so much stats and maths here, although I suspect that probabilities play a large part in machine learning.
In fact, I wonder if human intuition can somehow be linked to probabilities? Typically, computers make decisions based on full information. If they don’t have it, they error out. Or a human writes some heuristic algorithm to enable the computer to make an educated guess. Humans on the other hand usually have to make decisions without access to all the informtion all the time. A visionary leader is often someone who has a “feel” about what the right thing to do is without really knowing all the facts, some kind of intuition. I wonder if that can be simulated with probabilities. I wonder if I will find out.
“I wonder if that can be simulated with probabilities.” Somewhat similar but different question: I wonder in what ways the human mind approximates probabilistic reasoning, and at what levels. ie. For very low-level perception, does the brain simulate probabilistic reasoning? For mid-level? For high-level thought?
My gut sense is that the higher up the thought level (or further out from the brain stem) the more probabilities and ‘hunches’ come into play. I doubt that my brain plays guessing games about whether it is getting enough oxygen. It has the data it needs to make a call on that, and sends chemical and electronic instructions ASAP to all the other organs in the body to fix the problem it exists. “Should I ask that girl out on a date?” is a much trickier question to answer – although secular evolutionists would tell us that the brain needs to get the answer to that one right too – although we only need to get that one right some of the time!
But it probably depends on just how chemically deterministic our brains are. Is my brain just a bone-box full of chemicals, or is there something else going on that is not quite so deterministic? I would take the latter approach of course, but do not know where a line would be drawn.
The word “hunches” does indeed bring to mind higher-level through. Thinking about perception though, which seems like a process that starts at low levels and proceeds higher, we are given a game of ambiguities. The visual system is being handed many megabytes of raw data that it needs to piece together into a cohesive whole. That process needs to interpret what starts out as extremely ambiguous things like “brown pixel” into a sensible interpretation like “picture frame”. As we move from ambiguity to near-certainty, it might be helpful to have a representation that is in some ways akin to probabilities, even if the brain doesn’t do the maths quite like we would on paper.
Yes, I’m sure the brain does make educated (and probably therefore ‘probabilistic’) guesses all the way through – and sometimes gets it wrong. I think this could be empirically shown by thinking about the visual illusion things on the interwebs whereby you can ‘trick’ your brain into thinking it is seeing something even though you know it is not what it appears.