Conference paper Open Access

Continual Learning with Echo State Networks

Andrea Cossu; Davide Bacciu; Antonio Carta; Claudio Gallicchio; Vincenzo Lomonaco


Citation Style Language JSON Export

{
  "publisher": "Zenodo", 
  "DOI": "10.5281/zenodo.5164243", 
  "language": "eng", 
  "title": "Continual Learning with Echo State Networks", 
  "issued": {
    "date-parts": [
      [
        2021, 
        8, 
        5
      ]
    ]
  }, 
  "abstract": "<p>Continual Learning (CL) refers to a learning setup where<br>\ndata is non stationary and the model has to learn without forgetting ex-<br>\nisting knowledge. The study of CL for sequential patterns revolves around<br>\ntrained recurrent networks. In this work, instead, we introduce CL in the<br>\ncontext of Echo State Networks (ESNs), where the recurrent component<br>\nis kept fixed. We provide the first evaluation of catastrophic forgetting in<br>\nESNs and we highlight the benefits in using CL strategies which are not<br>\napplicable to trained recurrent models. Our results confirm the ESN as a<br>\npromising model for CL and open to its use in streaming scenarios.</p>", 
  "author": [
    {
      "family": "Andrea Cossu"
    }, 
    {
      "family": "Davide Bacciu"
    }, 
    {
      "family": "Antonio Carta"
    }, 
    {
      "family": "Claudio Gallicchio"
    }, 
    {
      "family": "Vincenzo Lomonaco"
    }
  ], 
  "id": "5164243", 
  "type": "paper-conference", 
  "event": "European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN)"
}
23
16
views
downloads
All versions This version
Views 2323
Downloads 1616
Data volume 2.0 MB2.0 MB
Unique views 2020
Unique downloads 1515

Share

Cite as