Computer scientists at Intel Labs in collaboration with Texas A&M University have developed an automated way of identifying the source of errors caused by software updates. Their deep learning algorithm is capable of finding performance bugs in hours instead of days.
The traditional way to find the source of errors within the software is to check the status of performance counters within the CPU. These counters are lines of code that gauge how the program is being run in the memory of the computer. By analyzing the counters the programmer can determine if the software is running correctly or if the software’s behavior goes awry.
Toady’s desktops and servers could have hundreds or thousands of performance counters, which makes it impossible to keep track of them manually. This is where the teams deep learning algorithm comes into play, the researchers were able to monitor data coming from a large number of the counters simultaneously by compressing the data. In the compressed format, the algorithm can look for patterns that deviate from the norm.
According to Dr Abdullah Muzahid, assistant professor in the Department of Computer Science and Engineering, the deep learning algorithm can also be used in developing the technology needed for autonomous driving. “The basic idea is once again the same, that is being able to detect an anomalous pattern,” said Muzahid.