dododo

as far as i can tell, the "10:10 code" is just a linear rescaling of the longitude and lattitude then encoding in base 32 (with a checksum for the last four bits).

if the above is correct, why not explain it thus?

if you do this, it becomes obvious what happens to errors; we know you can gracefully degrade the accuracy (since it's a linear encoding) by dropping the bits before the checksum.

what are the coding properties of the checksum? it doesn't look like crc32 or adler. what is it? why did you choose it?

it's more a problem with the APIs for talking to the vendor's DB. the easiest way to add user input to a DB query should be the right one. minimally, a DB query should not be trivially the same type as user input.

if f is the logistic function, then neural networks basically correspond to logistic regression.

http://en.wikipedia.org/wiki/Logistic_regression

which has been around since the 1940s... (multilayer neural networks correspond to hierarchical logistic regression--just plug them together). this should be in any reasonable stats book.

if you mean mathematically sound in terms of learning, etc, then chris bishops "neural networks and pattern recognition" is pretty good, full of sage advice and justification.

the gamma distribution is often used (for example, in neuroscience for spiking rates for the refactory period) to capture the case where you do not wish to give support to zero plus some short period after zero.

the exponential distribution is a kind of gamma distribution: Gamma(1,lambda) = Exp(lambda)

each driver decides whether or not to increase the distance to the next car. there are many such decisions that result in the total distance to the next car: i think it's a constant adjustment rather than just one decision.

another view might be that an exponential distribution on distances would lend (a lot of) support to zero distance: hopefully that's pretty rare.

it looks more like it's gamma distributed to me: you could imagine the generative process is something like a sum of exponentially distributed decisions times.