3 ms·
>> It would be far, far better if people were judged on their ability to find and rigorously test interesting questions, rather than whatever crapshoot nature s
by YeGoblynQueenne 8y ago
>> It would be far, far better if people were judged on their ability to find and rigorously test interesting questions, rather than whatever crapshoot nature spits out.
"Strong results" does not mean "positive results".
For example, two seminal works in ILP are the doctoral theses of Ehud Shapiro and Gordon Plotkin, both of which found strong negative results regarding the learning of first-order logic theories from examples. I would also point out the work of Mark. E. Gold and others on inductive inference that was mostly negative regarding the ability to learn anything above finite automata from examples.
It is true that in machine learning it is considered mandatory to show, experimentally, significant improvements in performance- personally I think that's a big mistake and a severe impediment to actual progress. In any case, in my field, you need theoretical results (theorems and proofs) for your work to pass muster.