talking to a intermediate-term quantitative equity manager (not a momentum strategy) yesterday trying to understand their methodology, as much as I could get him to divulge was for him to describe the process with the terms neural network & machine learning. His prior experience was working for a stat-arb shopguessing that the terms do not have any specific meaning other they sound cool. The process sounded to me like a continuous data mining of past returns through PCA / ICA or some other signal processing algorithm to create a return model that "learns" or adapts to changing conditionsthoughts?

I think a neural network is any time two patterns become associated. The patterns are themselves learned by the elements becoming associated. And then you see X and expect Y to come close by.QuoteThe theory is often summarized as "Cells that fire together, wire together.".[1] It attempts to explain "associative learning", in which simultaneous activation of cells leads to pronounced increases in synaptic strength between those cells. Such learning is known as Hebbian learning.So you could have a cell for time of day, and a cell for trading volume. The 9:00 AM cell gets wired to the trading volume cell. Then you have another cell which senses down ticks, and so on...

I studied neural networks quite heavily in my first degree in computer science, so I'd argue that neural networks have a very specific meaning, quite clearly defined at http://en.wikipedia.org/wiki/Artificial_neural_network The disadvantages are that (1) training can take a long time (2) the results are somewhat of a black box, your network will say "buy", "sell", "hold" etc., but it's not clear exactly how it reached that conclusion.Neural networks are a subset of the many tools for machine learning.