Unless you're an advanced user, you don't need to know how Braincel works its magic internally. But if you're curious, here's a very rough description:
A neural net consists of software-based simulated neurons or nodes connected in three or more layers, initially with random connection weights (node connections are roughly analogous to brain synapses). Historical data (such as daily stock market conditions for several months) in ranges in a spreadsheet is fed through these layers of nodes.
The net then attempts to decode the pattern, or construct the underlying "thinking" that leads to the outputs. To do this, it basically guesses on an outcome, then checks the results of this guess against the actual outcome and feeds back the resulting error information. Hits and misses strengthen or "reward" some connections and weaken or "punish" others. This process is repeated for many iterations until the net converges (reaches a state with the minimum number of errors and is considered "trained." When presented with questions (new data), it can now make forecasts and offer advice. As in statistics, forecasts and advice are probabilistic, not absolute.
Braincel Questions and Answers