Simulating a physical system using computer tools can hold valuable research and applications in the real world. Most of the existing tools for imitating the physical system are based on physical theory and numerical calculations. Over the past few years, computer scientists have been trying to develop techniques that complement these tools based on a large number of data analyzes.
Machine learning (ML) algorithms are one of the most effective methods of data analysis. Therefore, many computer scientists have developed ML techniques that can learn to mimic the physical system by analyzing experimental data.
Although some of these tools have achieved impressive results, it can be difficult to evaluate and compare them in any other way because of the many different methods available and the variations of their work. Repair Therefore, so far, these materials have been reviewed using different systems and temperatures.
Researchers at New York University have developed a new measurement channel that can be used to examine models for physical systems simulations . This suite, introduced in a previous article published in aRXIV, can be modified and improved to review different ML-based simulation techniques.
“We provide a minimal series of articles to guide the way towards benchmarking and evaluation,” the researchers wrote in their paper. “We offer four representative physical systems, as well as a standard time input and data processing method (based on DANA, MLP, CNN, nearby neighbors).”
The researcher’s quality results consist of four simple model assumptions and training and diagnostic constructs. There are four systems: a flow source, a wavelength analog (1D), a nerve conduction problem, and a wet source.
“These systems reflect the evolution of complexity,” the researchers explained in their paper. Spring is a continuous system with less space and less surface area than the first ship. Viewform comparison is a low frequency linear system with a relatively high (slightly) state. And we consider the creation of lower-level positions and higher-level state positions. And finally, the spring system has high performance as well as high position.
In addition to mimicking these simple physical systems, it includes a combination of methods and simulator tools. These include traditional numerical methods and data-driven ML techniques.
Using the coat, the scientist can re-evaluate its ML simulation technology, test its accuracy, functionality and accuracy. This allows them to reliably compare the performance of devices in different formats, which can be difficult to compare. A benchmark system can be created and extended to explore other functions and computational methods.
“We have found three ways to use the results of this work,” the researchers wrote in their paper. First, the developed dataset can be used to train and test new machine learning techniques in this area. They can be used, and third, to design future machine learning activities, including mimicking the trends seen in our results.
The new power distribution chamber, proposed by this research team, could help improve the evaluation of existing and emerging techniques for physical systems simulation. Currently, however, it does not cover the maintenance and installation of potential models, so it may be improved in the future.