A spreadsheet method for studying neural networks
David Walter, Michael McMillan · 2002
A unified framework and method for studying small neural networks (up to 75 neurons) using a computer spreadsheet is described. Neural networks actually resemble spreadsheets in several ways. A neural network consists of many simple computational units, highly interconnected and operating in parallel. Each unit has a numerical value (an output), which it communicates to other units along connections of varying strength. Similarly, a spreadsheet contains several thousand cells, arranged in rows and columns, appearing to perform in parallel. The numerical values of certain cells (i.e. their outputs) become parameters for calculating the values of others linked to them via suitable formulas. Just as the units of most networks are identical to each other, the formulas of spreadsheet cells are often highly repetitive, except for the relative location of cells which they reference. The authors do not advocate that all artificial neural networks be implemented on a spreadsheet. However, the spreadsheet is a valuable research tool and learning aid.>