I am working on the calibration of low-cost air sensor data (a time series regression problem). My primary focus is to use some meta/ few-shot learning approach to solve this problem with a lesser amount of data. I have tried using MAML on top of LSTM/vanilla NN but the results are not convincing. Is there a different approach/paper for meta-regression? Anything that I should be doing differently? Things to avoid etc? Thank you in advance.
1
-
$\begingroup$ Is there a reason you are not considering classical time series analysis methods, i.e., ARX, ARIMA, state-space models, etc.? They are certainly what would be called "few-shot learning methods" nowadays. ;-) Also, it would be a bit easier to help you if a) you could provide a bit more details about your problem (What are the data? What are you trying to estimate?) and b) you could formulate a precise question. (Asking for general advice on X is not a precise question.) $\endgroup$– Eike P.Aug 3, 2021 at 21:04
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
In this paper the authors recast MAML and Multimodal MAML for time series regression by creating "meta-windows", which are a group of rolling windows and are considered proxies for time series tasks.
