Is it possible measure the evolution of AI ?
30
Jul
Now it seems that yes, it is necessary to establish more precise tests and exactly the accuracy
of these tests, but researchers of the Computer Vision Lab, of the Federal Institute of Technology of Zurich have created a test system (benchmark) that tests the performance of neural network´s platforms used to perform common Artificial Intelligence tasks, such as answering questions, finding certain data, and other smartphone applications already used in common neural network tasks.
The idea is to measure the performance of AI systems, such as and done on today’s computers and on applications useful in smartphones, which use Artificial Intelligence.
The basic plan + and measure the performance in the systems, as already done for some answers: time, locators, flights, restaurants, etc. and growing in complexity, which will make manufacturers and application vendors make their models more sophisticated.
The application calls AI Benchmark and allows, among other things, the comparison of the speed of AI models to run on different Android smartphones, punctuating performance, already found in the Google Play app.
The device that has the dedicated AI chip dedicated to some devices (at least with Android OS 4.1 installed already will be fast enough to do these functions.
Some applications may be more complex, for example sorting images with easy recognition and the ability to segment and improve photos, indicating that this type of application can grow and evolve to more complex applications.
