Doing master thesis with the Machine Learning Group at Ericsson Research, Stockholm. Thesis is on “Deep learning for Anomaly Detection on Temporal Data”.
Part of another project on anomaly detection and predictive maintenance using sensor data from manufacturing production line.
Using LSTMs, autoencoder, Gaussian Process and Bayesian optimization.
Build a CNN based deep learning model to learn and break captchas.
The project was done as part of deep learning course at KTH with Tharidu and Dominik.
Trained the model on a dataset of 200,000 captchas on Nvidia Tesla K80 GPU an AWS EC2 instance.
Tensorflow was used as the deep learning library.
The model achieved an accuracy of ~98%.