Вы находитесь на странице: 1из 2

Competition “Ученые Будущего”, Nomination: Programming

Moscow, 2019 y.

Участники: Бийбосунов Данияр, Эсеналиев Тилек

Адрес: 1) 720010 г. Бишкек ул. Московская д. 219

2) 720000 г. Бишкек ул. Селекционная д. 1б

Контакты: +996 500 444, Email: danchegtop@gmail.com

УВК ШГ №70, 9М класс, контакты: +996 312 315 892 Email: отсутствует

Адрес: 720000 г. Бишкек ул. Боконбаева д. 149

Научный руководитель: Жолочубеков Нурсултан Жолочубекович, аспирант кафедры прикладной


информатики КГУ им. Арабаева (+996 312 623 - 409)

Контакты: +996 708 268 888 Email: nurik.ovich@mail.ru

Perceptron cluster ”Coniunctim” for solving tasks collectively


Coniunctim – perceptron cluster, created for solving tasks collectively.
Each perceptron, getting task, solving it differently. Then, their results combine, where bigger priority’s given to
continuous solution. It’s good conditions for collective work.

Methodic is based on collective mind. It’s a term that means process of solving task by collective analysis and
solutions of each member. Collective mind affects consensus and social capital. This project also include it, but
with reservation. Consensus realized as abstraction, stays in system as this phenomenon: Each past solution affect
each continuous, but for compensation injustice with continuous solutions, they’ll have bigger priority. And social
capital is also realized as abstraction introduces itself as system, where perceptron that made a mistake will have
lower priority, and perceptron that gave correct answer will slowly receive bigger priority.

Relevance, goals and results of project


Our project is very relevant now because of developing technologies of AI, ML and data science. Idea of project
was born from swarm robotics, but we made a way other from it and we created a new technology.

Target of our project was developing technology of AI, i.e. conquering new horizons of this sphere. Result of our
work – recursive bio mimicry (i.e. people copy interesting functions from nature, but here we integrated people’s
function).

Further we’ll improve our project and maybe we’ll add deep learning, and also integrate unique way of education,
where it happen not also after past experience, but also after other perceptrons.
Principle of work
In bigger picture it’s based on collective solving of tasks with perceptron cluster and algorithm, which combines
their solutions. Now in detail:
User give some signals, for example 1.00(float).
It transfers to other neurons.
For example, it’s template of our NN:

Input layer Hidden layer Output layer

Each perceptron has sigmoid activation function(1/(1+exp(-x))).

When hidden layer receive signal, It analyze his value, considering synapse’s weights (for example weight was
0.7).

Then we get, that final result is 0.7, and threshold of activ.func. is 0.66818777216.
Signal is passed to output layer, and on exit we get x = 0.7.
Other perceptrons (with different weights) also solve this task. For example, Other 2 gave result 0.8, and also 2.
Algorithm combines their solutions.
For example, combination based on average. Then we get result ~1.2. Then perceptrons educate by back
propagation method.
Important!!!
It was just example, real model works by harder algorithm and it’s math model’s written on programming
language Python.

Perspectives
Firstly we want to mark more suitable application of our project. It’s solving socio-economic problems. Firstly it
will solve basic tasks, then more complicated. It will be able to control states.

Conclusion
If we integrate our technology in some NN we will improve their capacity and we’ll open new way in developing
AI. If our upgraded NN control economics of our state (Kyrgyzstan) probably, we’ll have highest GDP. At all
creation of our project based on clear enthusiasm. I think, soon we’ll upgrade our project and put there deep
learning technology!

Reference
Wikipedia – about collective mind (30% inherited)

Вам также может понравиться