Improve Machine Learning-IoT Interface
Project Description
The content of this research is to improve IoT-Machine Learning bridge. Machine learning on device data now would require human to collect the data from, preprocess the data and run the algorithms manually. This process takes time and would hinder further automation of Machine Learning-IoT connection.
In order to reduce human actions needed in this process and improve training efficiency, two challenges needed to be overcome. The first is creating a scheduler for IoT server and Machine Learning Server. Because two servers will be executing computationally expensive operations, the scheduler needs to preemptively estimate the workload, identify dependencies between multiple machine learning processes and schedule the computation and interaction between servers. The second is the need to preprocess data. Because of the variety of dataset and their dependencies, for example, multiple sensors capturing the same phenomena, the datasets need to be synchronized, batched and preprocessed for the machine learning algorithm to achieve maximum effect. Thus, once given a basic set of rules, the servers need to preprocess the data accordingly before feeding it to a machine learning algorithm.
This research will focus on: 1. Creating a scheduler for Machine Learning and IoT server to achieve low server load and high training efficiency and 2. Setting basic set of rules for data preprocessing in order to synchronize the data for better training results. To narrow down the research direction, this research will focus on a ML IoT combo of extracting cooking related information to evaluate cooking process and replicate cooking process with robots.