A Refinement Algorithm for Deep Learning via Error-Driven Propagation of Target Outputs
| Autor: | Edmondo Trentin, Vincenzo Laveglia |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Relation (database)
Artificial neural network business.industry Computer science Deep learning Process (computing) Order (ring theory) 02 engineering and technology 010501 environmental sciences 01 natural sciences Backpropagation 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Artificial intelligence Layer (object-oriented design) business Algorithm MNIST database 0105 earth and related environmental sciences |
| Zdroj: | Artificial Neural Networks in Pattern Recognition ISBN: 9783319999777 ANNPR |
| Popis: | Target propagation in deep neural networks aims at improving the learning process by determining target outputs for the hidden layers of the network. To date, this has been accomplished via gradient-descent or relying on autoassociative networks applied top-to-bottom in order to synthesize targets at any given layer from the targets available at the adjacent upper layer. This paper proposes a different, error-driven approach, where a regular feed-forward neural net is trained to estimate the relation between the targets at layer \(\ell \) and those at layer \(\ell -1\) given the error observed at layer \(\ell \). The resulting algorithm is then combined with a pre-training phase based on backpropagation, realizing a proficuous “refinement” strategy. Results on the MNIST database validate the feasibility of the approach. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|