1: \begin{abstract}

2: 

3: Common variable star classifiers are built only with the goal of 

4: producing the correct class labels,

5: leaving much of the multi-task capability of deep neural networks unexplored. 

6: We present a periodic light curve classifier that combines a recurrent neural

7: network autoencoder for unsupervised feature extraction and 

8: a dual-purpose estimation network for supervised classification and novelty detection. 

9: The estimation network optimizes a Gaussian mixture model in the reduced-dimension 

10: feature space, where each Gaussian component corresponds to a variable class.

11: An estimation network with a basic structure of a single hidden layer attains 

12: a cross-validation classification accuracy of $\sim99\%$, 

13: on par with the conventional workhorses, random forest classifiers.

14: With the addition of photometric features, the network is capable of detecting

15: previously unseen types of variability with 

16: precision 0.90, recall 0.96, and an $F_{1}$ score of 0.93.

17: The simultaneous training of the autoencoder and estimation network is found to 

18: be mutually beneficial, resulting in faster autoencoder convergence, and superior classification and novelty detection performance.

19: The estimation network also delivers adequate results even when optimized with 

20: pre-trained autoencoder features,

21: suggesting that it can readily extend existing classifiers to provide 

22: added novelty detection capabilities. 

23: 

24: 

25: 

26: \end{abstract}

27: