On October 28-29, 2021 in Vancouver, the Future Technolgies Conference was held remotely through a participation via the Whova platform with active participation of PhD DSc Dominik Strzałka. As part of the conference, a recorded speech was presented on the topic: In Search of Machine Learning Theory, prepared together with prof. Eugeniusz Eberbach. The result of the speech is also a scientific article published in Lecture Notes in Networks and Systems, Proceedings of the Future Technologies Conference (FTC) 2021, Volume 1, pp. 599-615.
Paper abstract:
Machine Learning (ML) and Artificial Intelligence (AI), in general, are based on search algorithms. In this paper, we use paradoxically the same search techniques again to look for the general theory of machine learning itself. In other words, we search for a unifying machine learning theory. For this purpose, we turn out for a help to the general theory of computation, called $-calculus, that is based on meta-search and super-turing/hypercomputational models. We hope that in such a way, we can unify machine learning and this should be useful in the development of new methods, algorithms, embedded devices and computer programs in the future. Firstly, we overview main machine learning areas as our training examples, and by applying our background knowledge we hand pick-up a hypothetical reasonable theory of ML. Next, we justify that this is a good generalization of ML. The open research question remains whether by applying various ML techniques we can induce automatically the optimal ML theory from the hypothesis space of possible theories. Thus, hopefully, the follow-up paper would be titled “In search of the optimal machine learning theory”.
Please also visit:
https://link.springer.com/chapter/10.1007/978-3-030-89906-6_40
