As the domain of machine learning concepts in increasing day by day implementation of those are very limited since they require huge resources. We aim to reduce the resources required for such concepts and to make them implementable in resource constrained device.
Report : https://github.com/yadavdeepak95/Pocket-ML/blob/master/Pocket_ML.pdf
These are the datasets that were used for analysis (links might have expired).
USPS : https://drive.google.com/file/d/1i4WRMEysEvK8emeZmOZPGQbEZ2n0nwhs/view?usp=sharing
CIFAR-10 : https://drive.google.com/file/d/1FynLNaPfUQzJR_XuLhQ8ySKOgtAhpfhG/view?usp=sharing
MNIST : https://drive.google.com/file/d/1GYzayEc3jm0DG5iHT_X7c2mW9L8BREWo/view?usp=sharing
Mnist Preprocessed : https://drive.google.com/file/d/1G-uUFi7PBdGc_5SmbbeMKLeBXl7qD88X/view?usp=sharing
MADELON : https://drive.google.com/file/d/1sWMgzko_gp9PDqNXzND1Y8jCqWPj76KP/view?usp=sharing
notMNIST : https://drive.google.com/file/d/1MNhOQlaNVlRUlijrEDOI6u8sd0iZ69yM/view?usp=sharing its a pickle file divided into 6 parts: Training(X and y) Validation (X and y) Test (X and y)
Preprocessed notMnist : https://drive.google.com/file/d/1lsZaNgL6YZzxwFYzTXPB2bd9TuEvLn6Y/view?usp=sharing
Analysis DATASHEET: https://docs.google.com/spreadsheets/d/16hGYOnnIhxAGH5N35hx52uj4TsDxReYNg7BG4OWbm8E/edit?usp=sharing
This project analyses some of recent advances in machine learning field for resource constraint devices and proposes a novel convolutional approach to Bonsai Tree model. These models try to minimize the resource requirements like RAM and storage without hurting the accuracy much. Experimental results on multiple benchmark datasets shows that our modification to bonsai tree further reduces the resource requirement than normal bonsai tree, exploits the local structure of the data and increases the explainability of convolutional networks.