Presented at the 15th International Work-Conference on Artificial Neural Networks, Grann Canaria, Spain On June 12, 2019
This work was published in the Springer series: Advances in Computational Intelligence
It was done in collaboration with Nithish B. Moudhgalya, Siddharth Divi, S. Sharan Sundar and in the advisory of Vineeth Vijayaraghavan.
DeepTrace is a deep learning framework comprising five variants of a model whose basic structure includes two or more of the combinations produced by Convolutional block(s), LSTM unit(s) and Dense network(s).
We also introduce a novel training methodology by using future contextual frames. However, these frames are dropped during the testing phase to verify the robustness of DeepTrace in real-world scenarios.
An optimizer is used to offset the loss incurred due to the non-provision of future contextual frames. The genericness of the framework is tested by evaluating the performance on real-world time series datasets across diverse domains. We conducted substantial experiments that show the proposed framework outperforms the existing state-of-art methods.
- All the code files are inside the folder
Final Bi-CLDNN. - The required functions are inside
Final Bi-CLDNN/Model_Functions.py. - The data will be uploaded to a drive and the link will be updated in this document soon.