LIMITS: LIghtweight Machine learning for IoT Systems
Preliminary Beta Version
LIMITS is a python-based open source framework for automating high-level machine learning tasks targeted at resource-constrained IoT platforms. The low-level trainining of the models is performed by the coupled WEKA framework. LIMITS parses the WEKA outputs and derives an abstract model representation which is utilized for C/C++ code generation. Moreover, LIMITS can explicitly integrate the compilation toolchain of the targeted IoT platform in order to derive accurate assessments of the required memory resources for deploying the model to the considered platform.
Machine Learning Models
Currently, the following models can be utilized for data analysis and code generation [Classification/Regression Support]:
- Artificial Neural Network (ANN) [C/R] with sigmoid activation function
- M5 Regression Tree [R]
- Random Forest (RF) [C/R]
- Linear Support-Vector Machine (SVM) [C/R] based on Sequential Minimal Optimization (SMO)
The integration of additional models is planned for later releases.
Assumptions
- LIMITS works with CSV input data with a single line header. The label attribute of the data set should be defined in the first column, all other columns represent feature attributes. Examples can be found in examples, example data sets are provided at data.
- If the label is represented by a string value, LIMITS will perform classification, otherwise regression.
Quickstart
After following the setup instructions, the Command Line Interface (CLI) can be utilized for a fast setup verification:
$ ./cli.py -r ../examples/mnoA.csv -m ann,m5,rf,svm
r2 mae rmse training test
0.790+/-0.030 2.806+/-0.149 3.955+/-0.228 12.151+/-0.183 0.000+/-0.00
0.772+/-0.030 2.773+/-0.081 4.022+/-0.206 0.584+/-0.005 0.000+/-0.00
0.834+/-0.014 2.428+/-0.092 3.435+/-0.130 2.218+/-0.029 0.056+/-0.00
0.552+/-0.030 4.351+/-0.147 5.666+/-0.192 10.667+/-2.033 0.000+/-0.00
Related Publications
- B. Sliwa, C. Wietfeld, Towards Data-driven Simulation of End-to-end Network Performance Indicators, In 2019 IEEE 90th Vehicular Technology Conference (VTC-Fall), 2019
- B. Sliwa, C. Wietfeld, Empirical Analysis of Client-based Network Quality Prediction in Vehicular Multi-MNO Networks, In 2019 IEEE 90th Vehicular Technology Conference (VTC-Fall), 2019
- B. Sliwa, R. Falkenberg, T. Liebig, N. Piatkowski, C. Wietfeld, Boosting Vehicle-to-Cloud Communication by Machine Learning-Enabled Context Prediction, In IEEE Transactions on Intelligent Transportation Systems, 2019
- B. Sliwa, T. Liebig, R. Falkenberg, J. Pillmann, C. Wietfeld, Machine Learning Based Context-Predictive Car-to-Cloud Communication Using Multi-Layer Connectivity Maps for Upcoming 5G Networks, In 2019 IEEE 88th Vehicular Technology Conference (VTC-Fall), 2019
- B. Sliwa, T. Liebig, R. Falkenberg, J. Pillmann, C. Wietfeld, Efficient Machine-type Communication using Multi-metric Context-awareness for Cars used as Mobile Sensors in Upcoming 5G Networks, In 2019 IEEE 87th Vehicular Technology Conference (VTC-Spring), 2019, Best Student Paper