Empirical Pattern Discovery

I've heard (um, read) that term from NiclasOlofsson here on WikiWiki. It seems to describe something I've been thinking about for a long time. Probably this word is also right for what NeuralNetworks do.

One example that comes to my mind was a neural network built by NASA. They trained it with pictures of landing spaceshuttles, taken from a fixed point on the space-airport, as the input map. On the output map, they "told" the neuronal network the distance from the landing point, and the pitch of the spaceshuttle. After training, the network was able to tell the distance and pitch even if it only saw a small snippet of a picture of a spaceshuttle.

Another process where this term may be appropriate is the Image Analogies project at the New York University: http://mrl.nyu.edu/projects/image-analogies/sr.html. Given training image pairs (blurred and sharp) the program infers the sharpness in blurred images.

-- ManuelSimoni


I think what I meant is that EmpiricalPatternDiscovery is a process that programmers use to avoid losing focus on their current task, preferable in an ExtremeProgramming environment. RefactorMercilessly is good if you know what to fix, but sometimes all you have to go on is that, the code smells. I suggested putting RefactoringHints into the code as a way to empirical detect smells or patters not easily discovered otherwise. One way of doing this I figured is to use FixmeComments (REFACT) and a tool to extract and summorize the content.

EmpiricalPatternDiscovery could very well be a NeuralNetworks application, but not nesseccarily since that to me implies an ExpertSystem?. As a developer I like to stay in control. But it would be cool to have an internet based system that would collect information from other systems to discover bad smell and patterns. Perhaps applied on OpenSource software.

-- NiclasOlofsson


CategoryRefactoring CategoryDiscovery

EditText of this page (last edited December 31, 2007) or FindPage with title or text search