One of the bits of M366 Block 5 that I really enjoyed was the section about Braitenberg vehicles. They are amazing! Especially compared to the big, complicated subsumption-based systems that we learned about in Block 3 – these awesome little Braitenberg vehicles have no internal processing modules, just connections between the sensors and the motors, and yet they can behave in a surprisingly complex manner.
I really like this video featuring Type 2 Braitenberg vehicles built with Lego Mindstorms. Type 2 vehicles have excitatory connections between the sensors and motors; in the Type 2a vehicle, the left sensor is connected to the left motor, and the right sensor to the right motor, whereas in the Type 2b the connections are crossed. A simple variation, but the difference in behaviour is impressive:
Normally I don’t like business-related TMA questions, because it feels like an incursion by the world of work onto the serene oasis that is my OU studies. But Question 5 of M366 TMA03 is just funny enough to make up for that:
In this final question we are asking you to put yourself in the position of a senior technical employee of a software company. Your immediate line manager has recently seen a television science programme which mentioned, briefly, the power of neural networks in handling problems that conventional software systems struggle with, and has been temporarily shaken out of her usual torpor to become excited about this. […] Your manager has a short attention span, so your case study should be no more than 1200 words.
I wonder if the assignment author had a bad experience working in industry? I work in a public sector organisation, so I can’t comment on whether software companies are good or bad places to work, but I think dealing with well-meaning yet clueless managers is probably a near-universal experience!
Anyway, now that I’ve finished my concise 1200-word report for my imaginary line-manager, I’m pretty much done with TMA03 for both M366 and MS221. As I was saying in my last post, I’ve enjoyed this M366 assignment a lot more than I expected, particularly the section about self-organising maps. There’s something about SOMs that really captures my imagination – I guess I like the idea of training a SOM to detect meaningful patterns that are just too subtle for my feeble human senses to notice. Though there’s always the danger that the SOM will find interesting-looking but ultimately meaningless patterns in the data, I guess; kind of like the way humans often read patterns into things that aren’t really there.
I’m quite glad Block 4 of M366 is over with now, because it’s the block I’ve felt least confident with. I think taking a break in the middle for M257 revision was probably a bad idea, and to be honest my understanding of the topics in Block 4 feels a bit shaky and fragmented – still, plenty of time to do TMA04 and then get on with revision for the exam in October!
M366 is a very strange course. I haven’t really enjoyed reading the course texts themselves, but the assignments are much more interesting. I spent ages fiddling around with the rabbits simulation for TMA02, and I’m finding the neural network experiments in TMA03 even more rewarding. Yes, I’m one of those weirdos who enjoys collecting and analysing data – so compiling the big network performance tables and assessing the accuracy of the perceptrons in Questions 1 & 3 was brilliant fun for me.
At the moment I’m working through Question 4, the one about self-organising maps, but unfortunately I’ve not been able to get my SOM to work properly so far – I’ve no idea why, but JavaNNS just doesn’t respond when I click the “Learn All” button to train the network. I’m probably missing something obvious, so I’ve emailed my tutor to see if he can figure out what’s going wrong – and in the meantime I guess I’m taking a bit of an enforced break. Can’t wait to get back to it, though!
It’s such a weird feeling – in the space of a week, I’ve gone from complaining that M366 feels like a second job (while I was reading through Block 4), to being so eager to get back to my TMA03 experiments that I find myself daydreaming about it during my actual job! I hope this means that I’ll enjoy the exam just as much…
UPDATE: Victory! I’ve finally managed to get my SOM for Question 4 working – reinstalling JavaNNS seems to have done the trick (but I still have no idea what caused the problem in the first place). Now, on to Question 5!
I had a really great study timetable worked out for June and July, which included a nice weekend off (for a little holiday in the Peak District), followed by a week of gentle catching-up to get back on schedule. Unfortunately for me, I came back from my holiday with a nasty cold, and ended up missing an extra week of studying. So for the last three days I’ve been trying desperately to get back up to speed with MS221, which has basically meant studying from breakfast til bedtime, with short breaks for meals. If anyone is wondering whether it’s possible to do a unit of MS221 in 24 hours, I can confirm that it is! Whether it’s possible to do it well is another matter entirely…
Anyway, despite having to rush through it, I’ve enjoyed Block C of MS221 for the most part. I was a bit bored by Taylor polynomials at first, but now that I’ve finished unit C3, I do find them quite satisfying to work with – like the presenter in the associated Algebra Workout program said, it’s a bit like “splitting the atoms” of functions, and getting to look at the building blocks that they’re composed of (or at least, that a good approximation might be composed of).
I’ve just about finished TMA03 now, so this afternoon I’ll be getting back to Block 4 of M366, and then making a start on TMA03 for that course – I’ve heard that it’s a very tough assignment, so I’m really quite worried that I won’t get it finished in time. Still, if the last few days are anything to go by, fear of failure is an excellent motivator when I’m pushed for time!