View in #questions-forum on Slack
@Davide: Hi everyone ! I was wondering if time series datasets had ever been used in the field of Continual Learning. Do you know any of them?
@Soumya: There are some papers which have talked about continual learning of rnn and lstm
Even there is a paper (kind a review) by Yoshua Bengio
https://arxiv.org/pdf/1811.07017.pdf
@Davide: Thank you ! I will read it
@vlomonaco: @andcos is working on CL for Time Series, right?
@andcos: Yes, you’re right @vlomonaco 
Actually, most of the papers I have seen work either with synthetic time series (artificially generated) or with MNIST-like images taken pixel-by-pixel. There are some attempts to work with NLP (like MNLI dataset). Unfortunately, real-world time series are nowhere to be found in current literature. Of course, to the best of my knowledge 
Just to share a couple of references:
• https://link.springer.com/chapter/10.1007%2F978-3-030-30484-3_56 here there is a study of forgetting in LSTM (however, the authors did not propose a solution)
• https://ieeexplore.ieee.org/document/6707047 this is an older paper with a recurrent model which alleviates forgetting, tested on both synthetic and MNIST sequences
sorry for paywalled-link, I had these ones in my bib, but you can find open versions if you are interested
SpringerLink: A Study on Catastrophic Forgetting in Deep LSTM Networks
@Subutai_Ahmad: We put together a set of real world time series datasets here:
https://github.com/numenta/NAB
numenta/NAB
The main use case was anomaly detection, but it could be used for other things too.
Here’s our paper on it: http://www.sciencedirect.com/science/article/pii/S0925231217309864
Unsupervised real-time anomaly detection for streaming data
@Davide: thanks @andcos for sharing these articles 
. I will surely read them !
Thank you for sharing this repository @Subutai_Ahmad ! It looks very intertisting.