INTELlinext

Using machine learning to predict PC user behavior.

To accurately predict the apps that PC users will use (and for how long), we use Hidden Markov and LSTM models. We're able to predict the most likely next three apps with 70% accuracy, and the usage duration within a margin of error of 45 seconds. These results hold promise for preloading applications users are likely to use in the background, reducing user waiting time. View Whitepaper

We want to improve percieved system experience hampered by loading times. How?

To collect the data necessary for our prection models, we utilize the Intel System Usage Reporting (SUR) SDK to gather data on usage habits. Deployed on 8M+ systems worldwide, SUR utilizes the Intel Energy Server (ESRV) to analyze and store important datapoints related to usage. We develop four efficient, low-impact Input Libraries (ILs) with the SUR SDK to get a picture of what apps a user is likely to use. View Paper