Hello, I was hoping someone could suggest an area within the Linux kernel that could benefit from research in machine learning. I'm particularly interested in dynamic learning where the correct solution changes over time and the learning algorithm needs to adapt. What areas of the kernel would benefit from this type of continuous learning? This need to be an area where no straightforward efficient optimization exists. Another requirement is that the learning algorithm needs some type of feedback on performance. There needs to be a way for the algorithm to assess if it is making the right choices. Ideally there should be a function that the algorithm is trying to learn that depends on the various dynamic properties of the operating system. The function selects certain operations that the OS should perform based on the current situation. Ideally, if the algorithm makes a bad choice, the system would give feedback by telling the system what would have been the best(better) choice. (A straight forward example is the weather. An algorithm makes a prediction and the next day it finds out the correct weather.) Of course, this ideal is pretty hard to get, but it would be great if some part of the kernel had a learning problem that was close. (There are nice theoretical guarantees for these types of problems.) However, even if this is not possible with the current kernel, in principal, some type of learning should be helpful. The very fact that the kernel can have parameters tuned to improve performance suggests that this tunning can be automated. Any suggestions? Thanks, Chris Mesterharm -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ IRC Channel: irc.openprojects.net / #kernelnewbies Web Page: http://www.kernelnewbies.org/
