Cognitive computing

Body: 

The relatively recently appeared topic of cognitive computing introduces new paradigms for future computing machines that will be based on not only data processing and memorizing, but first of all on self-thinking, adaptive and decision-making units.

The history of science development teaches us that groundbreaking discoveries and solutions result from careful observation of nature. The basic example gave us Leonardo da Vinci leading the first work on flying apparatus and other issues. The expected progress in cognitive computing can take its inspiration from life science and hopefully from human-brain structures.

Despite the fact that there are many advanced ideas and solutions in the field of solid state physics and electronics, especially in spintronics, it is difficult at this moment to talk about the existence of a physical layer that implements cognitive informatics. In other words there is no fully hardware solutions. Thus, theoretical achievements at computer engineering science have reached a level of computing algebra, cognitive semantics or even sociopsychological perspective. These fields are however not fully compatible with the physical layer of actual technology.

It seems that from among many interesting solutions and directions of cognitive computing research, at least the two aspects may contribute to the beginning of creating of the physical layer of the cognitive calculations. These are: propagation of domain walls in long fibers and, from the material point of view, fiber systems as such possessing magnetic properties. The above propagation were materialized by S. Parking et al. in his race-track memory solution working at ps time scale. The fiber systems, on the other hand, made by nanolithography or electrospinning can provide wide range of possibilities, possessing a variety of spatial order and scalability, going from totally irregular to ordered systems providing new class of self-organizing phenomena emerging from growing complexity. Both research pathways require however intense efforts, and what is proposed in our manuscript is the analysis of demagnetization dynamics, and occurrence of domains, in ferromagnetic fibers of  different cross-sectioned shapes and different curvature. In this way we, at least partially, address the issue of material scale that can be used in data processing.

The question arises what an approach can be useful for future cognitive computing; that based on replication of natural objects or this based on artificial intelligence solutions that do not always refer directly to solutions found in nature. The first choice goes more into hardware architectures, the second one is mostly involved in software developments. It seems the future cognitive computing should go into such the way to manage rapidly increasing amount of data, and such a problem can be clearly solved by new hardware-base solutions beyond classical von Neumann or Harvard computing standards. This is, there is a need to go beyond actual solutions and to overcome barriers introduced by today’s classical machine learning, pattern recognition, and image processing.

Simplifying the whole thing the Neumann architecture is based on the two layers; processor (1) and memory (2), both accessed by external input/output devices. This separation leads however to limited speed of data processing because of the need to continuously provide and receive data from the memory region.

Some cognitive computing models, on the other hand, consider layered structure of the future cognitive processors imitating human-brain solutions. It postulates, going from bottom to up, the following layers: sensation (1), memory (2), perception (3), action (4), meta-cognitive functions (5), meta-inference functions (6), and higher cognitive functions (7). The hardware representation of this approach is not proposed to this day. The physical solution by Parkin et al. meets expectations for a combined data processing and memory units at a single place.

Cognitive informatics is a multi-disciplinary effort, combining the results of natural science, life science, computer science, and hopefully, in the extended form, with the solid-state physics and associated nanotechnology achievements.