Optimization problems are to determine the best viable response from a variety of options, which can be seen frequently both in real situations and in most fields of scientific research. However, there are many complex problems that cannot be resolved with simple computer methods or that would take excessive time to resolve.
Because simple algorithms are ineffective in solving these problems, experts from around the world have worked to develop more effective strategies that can solve them in realistic deadlines. Artificial neural networks (Ann) are at the heart of some of the most promising techniques explored to date.
A new study by the Vector Institute, the University of Waterloo and the Institute of the Theoretical Physics perimeter in Canada has a variational neural receipt. This new optimization method combines recurring neural networks (RNN) with the concept of receipt. By using a configured model, this innovative technique generalizes the distribution of solutions achievable to a particular problem. Its objective is to solve problems of optimization of the real world by using a new algorithm based on the theory of CAP and the RNN of natural language treatment (NLP).
The proposed framework is based on the principle of the receipt, inspired by the metallurgical receipt, which consists in heating the material and cooling it slowly to bring it back to a lower, more resistant and more stable energy state. The simulated receipt has been developed on the basis of this process, and it seeks to identify digital solutions with optimization problems.
The largest distinctive characteristic of this optimization method is that it combines the efficiency and processing capacity of ANN with the advantages of simulated receipt techniques. The team used the RNNS algorithm which showed a particular promise for NLP applications. Although these algorithms are generally used in NLP studies to interpret human language, researchers have reused them to solve optimization problems.
Compared to more traditional digital receipt implementations, their RNN -based method has produced better decisions, increasing the effectiveness of conventional and quantum receipt procedures. With self -regressive networks, the researchers were able to code the receipt paradigm. Their strategy is resolving optimization problems at a new level by directly operating the infrastructure used to form modern neural networks, such as Tensorflow or Pytorch, accelerated by GPU and TPU.
The team carried out several tests to compare the performance of the method with traditional receipt methods based on digital simulations. On many problems of paradigmatic optimization, the proposed approach has gone beyond all techniques.
This algorithm can be used in a wide variety of real world optimization problems in the future, allowing experts in various fields to resolve difficulties faster.
Researchers wish to further assess the performance of their algorithm on more realistic problems, as well as to compare it to the performance of existing advanced optimization techniques. They also intend to improve their technique by replacing certain components or incorporating new ones.
You can see the full article here
There is also a code on GitHub:
Variational neuronal reception
Classic and quantum simulated receipt
