1: \begin{abstract}

2: Multitask algorithms typically use task similarity information as a bias to speed up 

3: and improve the performance of learning processes. 

4: Tasks are learned jointly, sharing information across them, in

5: order to construct models more accurate than those learned separately

6: over single tasks. In this contribution, we present the first multitask

7: model, to our knowledge, based on Hopfield Networks (HNs), named

8: HoMTask. We show that by appropriately building a unique HN embedding all tasks, 

9: a more robust and effective classification model can be learned. 

10: HoMTask is a transductive semi-supervised parametric HN, that

11: minimizes an energy function extended to all nodes and to all tasks under

12: study. We provide theoretical evidence that the optimal parameters automatically 

13: estimated by HoMTask make coherent the model itself with the prior knowledge 

14: (connection weights and node labels). The convergence properties of HNs are preserved, 

15: and the fixed point reached by the network dynamics gives rise to the prediction of unlabeled nodes. The proposed model improves the classification abilities of singletask HNs on a preliminary benchmark comparison, and achieves

16:   competitive performance with  state-of-the-art semi-supervised graph-based algorithms.

17: 

18: %Preliminary benchmark experiments on a real-world classification problem show that the proposed model is competitive with several state-of-the-art semi-supervised graph-based algorithms.

19: 

20: \keywords{Multitask Hopfield networks, multitask learning, multitask classification}

21: 

22: \end{abstract}

23: