N.N. Diep1, A.A. Zhdanov2

1 JSC Intellect (Moscow, Russia)

2 JSC Lebedev Institute of Precision Mechanics and Computer Engineering Russian Academy of Sciences (Moscow, Russia)

Today neural network methods of building intelligent systems are in the spotlight. One of the problems with such systems is the determination of the network configuration. This work is devoted to this problem. However, not widely known recognizing neural networks based on neurons – threshold adders, are considered, but networks (which, for the sake of distinction, we call neuron-like), built on the basis of self-learning models of neurons developed within the framework of the theory of "Autonomous adaptive control" (AAC), and used not only for building recognition systems, but also for the construction of adaptive control systems. 

In this approach, there is also the problem of forming an effective network configuration. A solution is proposed based on an "event tree" that calculates the most probable direction of growth of a network of neurons that will be in demand by the system. As a result of using such the algorithm, the capacity of the recognition system increases with the same number of neurons - the number of images that such a system can automatically generate. 

The proposed approach increases the efficiency of the control system. Shown is an applied example with an adaptive copter control system.

Diep N.N., Zhdanov A.A. Algorithm for self-growing neural-like network in autonomous adaptive control system. Neurocomputers. 2021. V. 23. № 6. Р. 24−31. DOI: https://doi.org/10.18127/j19998554-202106-03 (in Russian).

1. Zhdanov A.A. Avtonomnyy iskusstvennyy intellekt. M.: «Binom. Laboratoriya znaniy». 2008. (in Russian).
2. Zhdanov A.A. Metod avtonomnogo adaptivnogo upravleniya. Izvestiya RAN. Teoriya i sistemy upravleniya. 1999. № 5. S. 127–134. (in Russian).
3. Zhdanov A.A. Formalnaya model neyrona i neyroseti v metodologii avtonomnogo adaptivnogo upravleniya. Sb. «Voprosy kibernetiki». Nauchnyy sovet po kompleksnoy probleme «Kibernetika» RAN. Vypusk 3. 1997. S. 258–274. (in Russian).
4. Zhdanov A.A., Ryadovikov A.V. Neuron Models in the Autonomous Adaptive Control Method. Optical Memory and Neural Network. 2000. V. 9. № 2. P. 115–132. (in Russian).
5. Kryzhanovskiy M.V. Printsipy neyronopodobnoy realizatsii sistem avtonomnogo adaptivnogo upravleniya. Diss….kand. f.-m. nauk. M.: 2004. (in Russian).
6. Zhdanov A.A., Kryzhanovskiy M.V., Preobrazhenskiy N.B. Bionicheskaya intellektualnaya adaptivnaya sistema upravleniya mobilnym robotom. Iskusstvennyy intellekt. 2002. T. 4. S. 341–350. (in Russian).
7. Zhdanov A.A. O ponyatii avtonomnogo iskusstvennogo intellekta. Sb. nauch. trudov «Iskusstvennyy intellekt v tekhnicheskikh sistemakh». M.: Gos.IFTP. 1997. S. 142–157. (in Russian).
8. Zemskikh L.V., Samarov E.K., Zhdanov A.A., Babkova V.V. Primeneniye geneticheskikh algoritmov dlya optimizatsii adaptivnoy sistemy upravleniya mobilnogo robota na parallelnom vychislitelnom komplekse. Tr. In-ta sistemnogo programmirova-niya. T. 7: «Novyye podkhody v neyronopodobnykh i osnovannykh na znaniyakh sistemakh». M.: ISP RAN. 2004. S. 79–104. (in Russian).
9. O’Keefe J., Dostrovsky J. The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Research. 1971. V. 34. P. 171–175.
10. Quiroga R.Q., Reddy L., Kreiman G., Koch C., Fried I. Invariant visual representation by single neurons in the human brain. Nature. 2005. V. 435 (7045). P. 1102–1107.