Creative Machine Learning - a new paradigm
to research and build next generation
ML systems.
“I asked him: ”How much will be two plus three?"
He answered: “Three plus two because adding is commutative!”
But he wasn't able to perform adding."
Vladimir Arnold about contemporary “smart” youth.
Paradigm of Today
Machine learning is a method of problem solving that can be implemented when and where a classic mathematical model and computational algorithm for finding a solution can not be elaborated and deployed and thus the machine is not able to handle the problem and resort to any solution.
Computer systems that use such methods of machine learning are known as artificial intelligence (AI) systems. They are designed under the hypothesis that human brain is like a huge and complex telephone circuit.
Mathematical substratum of such systems are models of artificial neurons and neural networks with multi objective optimization problems for such models being solved by a mathematical substance.
3 Tasks of Brain
There is a simple principle underlying the contemporary "AI" systems. This principle is to find the answer to the problem by fitting it in. So the mathematical foundation of such systems is just a developed and modified method of least squares.
However, human brain doesn't function like that. Real living brain is able not only to find the answer but to create the very solution from scratch. This includes solutions to the problems that human brain faces for the very first time when its resort to a solution is being made without referring to any statistical data.
Furthermore the brain is able to define (set/put) the very problem that requires the solution itself. Thus each moment brain exists it implements the following 3 tasks simultaneously: defining (setting/putting) a problem, finding a solution, making a response.
This simple view materializes as a complex and dynamic totality of biological, chemical and physical processes.
Key features of CML paradigm in comparison with other ML paradigms:
Considering both electrical and chemical processes.
​
Taking into account not only recognition but also creation.
​
Transformational shift in mathematical foundation - from probability and statistics to topology and algebraic geometry.
​
Transition from quantitative to qualitative methods.
Moving from classic statistics based AI and ML models and systems towards the next generation AI and ML is possible only through combining neuropsychological models with contemporary results in geometrical analysis and quantum field theory.
Problems that can be resolved with the implementation of CML paradigm:
Data quantity and quality- staggering decrease in the amount of data required to power ML algorithms.
​
Narrow focus - AI universality instead of narrow specialization.
​
Computational and computing power - razor-sharp decrease in the quantity of data required and thus substantial decrease in required computer power.