CNN-RNN Project
Budget: $30 – $250 USD
1. Explain how you can classify the fashion MNIST dataset using LSTM RNNs. You need to draw the LSTM RNN architecture (pref unrolled) and comment on all dimensions that you decide. The data set can be found in Kaggle. https://www.kaggle.com/zalando-research/fashionmnist
2. Implement the RNN approach you outlined in the previous question using TF or PyTortch. Implement the equivalent CNN using the same framework (CNN solution is in your notes so just include this solution).
3. Compare the performance of the LSTM RNN network vs CNN. Compare the training times between RNN and CNN that achieves the same test performance.
4. Read this paper and write a summary on how RNN and CNNs can be combined for the classification task and why you believe we get a performance benefit from such combination. Paper: https://openaccess.thecvf.com/content_cvpr_2016/papers/Wang_CNN-RNN_A_Unified_CVPR_2016_paper.pdf
2. Implement the RNN approach you outlined in the previous question using TF or PyTortch. Implement the equivalent CNN using the same framework (CNN solution is in your notes so just include this solution).
3. Compare the performance of the LSTM RNN network vs CNN. Compare the training times between RNN and CNN that achieves the same test performance.
4. Read this paper and write a summary on how RNN and CNNs can be combined for the classification task and why you believe we get a performance benefit from such combination. Paper: https://openaccess.thecvf.com/content_cvpr_2016/papers/Wang_CNN-RNN_A_Unified_CVPR_2016_paper.pdf