A form of targeting that makes use of one or more neural networks to predict enemy movement.
The first known bot to make use of neural networks was Qohnil's XBot, in May, 2002, but not much is known about it. Albert's neural network bot ScruchiPu came roughly a year later, which sparked some interest in neural networks on the RoboWiki. ScruchiPu used enemy speed and turn rate as input and output, iterating into the future one tick at a time to predict the enemy's movements, much like pattern matching.
Several more neural targeting bots were released in the following years, meeting with varied success, but none approaching the accuracy of other top targeting methods.
A more recent neural network bot is Engineer, authored by Wcsv. Engineer uses Waves to collect GuessFactors that correlate to given situations, much like a traditional GuessFactor gun, then feeds the situational attributes (as input) and resulting GuessFactors (as output) to the neural network as training data. Engineer was the first neural targeting bot to reach a RoboRumble rating of 2000+ when its initial release hit a rating of 2003 in May, 2006. (Incidentally, Engineer also uses this neural network system for its WaveSurfing.)
Another attempt at neural targeting came in 2007 from Chase-san in the form of his "Prototype" gun. Attached to an early version of DrussGT's movement, it reached a rating of 2005 in the RoboRumble in September, 2007.
The most modern example of neural targeting is Gaff, which uses waves, GuessFactors, Radial Basis Functions, and two neural networks. Darkcanuck has composed a detailed write-up of its workings at Gaff/Targeting. Gaff's gun is quite strong, on par with other top gun types according to TC2K7 results, and is probably the strongest Anti-Surfer gun of any kind, according to Anti-Surfer Challenge results.
- Temporal Difference Learning project
- Java Object Oriented Neural Engine
- JSOMap - A Java-based package for working with Self-Organizing Maps (are another type of neural network).