Achieving Low Power Classification with Classifier Ensemble
| Autor: | Min Zhang, Fanglei Hu, Hailong Jiao |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Boosting (machine learning)
Artificial neural network Computer science business.industry 020208 electrical & electronic engineering Pattern recognition 02 engineering and technology Support vector machine Statistical classification ComputingMethodologies_PATTERNRECOGNITION Polynomial kernel 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing AdaBoost Artificial intelligence business Classifier (UML) MNIST database |
| Zdroj: | ISVLSI |
| DOI: | 10.1109/isvlsi.2018.00014 |
| Popis: | Machine learning algorithms, such as Support Vector Machine (SVM) and Artificial Neural Networks, have been widely applied in many aspects of daily life. Low power/energy integrated circuit implementation of machine learning algorithms with high accuracy is however still a great challenge, which is critical for portable and wearable devices. In this paper, classifier ensemble is investigated for achieving low power classification. The classifier ensemble algorithm Adaboost is employed to combine multiple SVM classifiers with linear kernel to achieve high classification accuracy while reducing the hardware complexity. The proposed classifier ensemble is evaluated on the MNIST dataset by using a 45-nm CMOS technology. Compared to the traditional SVM classifier with second-order polynomial kernel, the proposed classifier ensemble achieves up to 45.7%, 20.3%, and 20.3% savings in total energy consumption, leakage power consumption, and area, respectively, while providing similar classification accuracy. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|