Automated Machine Learning System
- 0 Collaborators
Currently, a large amount of data is being generated by various means such as sensors, cell phones, and the Internet. However, manual analysis of such data is not feasible, and it is necessary to use computational techniques to extract useful knowledge. Several of these techniques are developed in the growing area of Data Science. A key tool of the data scientist to deal with knowledge extraction is Machine Learning. However, deriving a good model from a raw data set is not a trivial task. This requires that the specialist conducts several steps of pre-processing, algorithms choice and setting hyperparameters. Such tasks are costly, leading to many decision-making that, in general, are not possible for lay users. This project aims to research and develop an Automatic Machine Learning system in three domains: preprocessing (cleaning, normalization, attribute selection), modeling (appropriate choice of Machine Learning algorithms, adjustment of its hyperparameters and final model) and post-processing (model evaluation and user report). It is hoped that the system will be able to assist both expert and lay users in the steps required to analyze data, avoiding that they spend much time with manual adjustments, focusing on problem analysis. ...learn more
Project status: Under Development
Intel Technologies
Other
Overview / Usage
This project aims to research and build an AutoML system in three domains: preprocessing (cleaning, normalization, attribute selection), modeling (appropriate choice of AM algorithms and their hyperparameters) and post-processing and report to the user). In addition, the potential of Metalearning will be investigated to use the meta-knowledge generated by other executions in recommending the activities related to each of the three modules. Through this project it is expected that the system will be able to help expert and lay users in the necessary steps for data analysis, avoiding that they spend much time with manual adjustments, focusing more on problem analysis and other priority tasks.
Methodology / Approach
- Pre-processing step with: cleaning, normalization, feature selection. 2) Modeling step: choice of Machine Learning algorithm and hyperparameters. 3) Post-processing step: evaluation of the models and report to the user. 4) Apply Metalearning at each stage to build a recommendation system for each of the three modules
Technologies Used
.....