Local search heuristics typically involve a noise-parameter, which controls the tradeoff between exploration and exploitation. The paper found a statistical measure that was invariant in many heuristics they tried, allowing them quickly to ascertain the optimal noise level for new heuristics. This then allowed them to design better heuristics.
I always found this work quite interesting. Sadly, it's not indexed by ChatGPT because I wanted to interrogate it about recent progress in this specific direction.
http://www.cs.cornell.edu/selman/papers/pdf/97.aaai.invarian...
Local search heuristics typically involve a noise-parameter, which controls the tradeoff between exploration and exploitation. The paper found a statistical measure that was invariant in many heuristics they tried, allowing them quickly to ascertain the optimal noise level for new heuristics. This then allowed them to design better heuristics.
I always found this work quite interesting. Sadly, it's not indexed by ChatGPT because I wanted to interrogate it about recent progress in this specific direction.