We don’t meet often, but every time we talk, she reminds me that I had failed to recognize her the first time we had met after graduating together from school. Yes, I could claim in my defence that I was seeing her for the first time in over six years. While that might be a valid excuse for most people, it doesn’t apply to me, since I normally claim to have superior long-term memory. If I’ve seen you somewhere before, I ought to recognize you. The only times I don’t I’m pretending, since I don’t want to embarrass you (and myself) by recognizing you while you don’t recognize me (see this incident for an example of this).
The reason for my failure that cold Bangalore evening in December 2006 was that my Bayesian system had failed me. Let me explain, in the process giving you an insight into my Bayesian system which I use to recognize you when I meet you.
About a month or two back, I was at a friend’s wedding, which is where I hit upon this term “Bayesian recognition” to explain this phenomenon  (which I’ve been practicing for ages). Now, this friend whose wedding I was attending was one year my junior at two different schools. As you might expect at an event where you and the host share more than one social network, there were a lot of familiar faces. Some people I knew fairly well, and could easily recognize. But the others had to go through a “Bayesian search”.
So when I saw someone who was one of three people I know – let’s say X, Y and Z. In order to determine which of these this person is, I would ask myself two questions – firstly, what were the prior odds that the person I saw could be each of X, Y or Z. Secondly, what were the odds of each of X, Y and Z being there at that event. Note that the latter is important. For example, if someone at the event looks like you and I know (for example) that you are currently in another country, despite the strong resemblance I can discount the possibility that that person is you, and go ahead with my search.
Note that this differs from “frequentist recognition”, where I only look at the person’s face and try and understand who he/she most resembles, without any thought to the odds that that person is there. Frequentist recognition can lead to a large number of false positives, and after a few rounds of embarrassment, you start giving up on recognizing, and many a possible reunion thus gets missed. Bayesian recognition, on the other hand, restricts your field of search (to the people who you give good odds of being there), prevents you from being distracted and increases your chances of making a good recognition.
So why did Bayesian recognition fail me when I met this former classmate back in 2006? The problem was her company. She had come for this Deep Purple concert with another friend of mine, who was my classmate in another school (and who I had been in touch with, and so easily recognized). I had no clue that these two were friends (it turned out they didn’t know each other that well – they had come there with a common friend). So when this girl (the one I didn’t recognize) popped up with “Hey SK! Do you remember me?” I assumed that she was someone I knew from the same school as the other girl I was meeting, and that wrongly restricted my search space. And so my mind was trying to map her to my friends from school 1, while she happened to be a friend from school 2. And my search returned a blank, and my legendary long-term memory skills were embarrassed.
I must mention here, though, that this is possibly the only time that my Bayesian recognition model has acted up, and refused to recognize someone I know. There have been 2-3 false positives, but this has been the only negative. And when you consider the sample size to be all the people I have recognized in different places, this is small indeed.
Oh, and after failing to recognize her then, I’ve kept in touch with this friend.
Fully agree. That’s how I track characters in the novels I read. Don’t remember names, just the context & the circumstances. So people popping up where they shouldn’t (!) causes confusion!
@karthiks whatnonsense. you obviously do frequentiest recognition and then tune it up to do bayesian.
My 2 cents – this is post facto rationalization + oversimplifying clearly what is a complex process – you should read at least the first instance in this – http://robinlea.com/pub/wife-hat.pdf – I got here from watching a fantastic movie called Awakenings (http://www.imdb.com/title/tt0099077/)
so what do you think your search space size is ?