When Size Matters: Markov Blanket with Limited Bit Depth Conditional Mutual Information

Due to the proliferation of mobile computing and Internet of Things devices, there is an urgent need to push the machine learning frontiers to the network edge so as to fully unleash the potential of the edge big data. Since feature selection becomes a fundamental step in the data analysis process, the need to perform this preprocessing task in a reduced precision environment arises as well. To achieve this, limited bit depth conditioned mutual information is proposed within a Markov Blanket procedure. This work also shows the process of generating approximate tables and obtaining the values required to test the independence of the variables involved in the algorithm. Finally, it compares the results obtained during the whole process, from preprocessing to classification, using different numbers of bits.