Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/43321662-C403-4226-A0E3-83CF6B2EE0AA | ||
| Year | 2021 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Kosmas Pinitas, "Dendritic application to machine learning", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2021 https://doi.org/10.26233/heallink.tuc.90049 | ||
| Appears in Collections |
The current deep learning architectures achieve remarkable performance when trained in large-scale controlled datasets. However, the predictive ability of these architectures significantly decreases when learning new classes incrementally due to their inclination to forget the knowledge acquired from previously seen data, also called catastrophic-forgetting. The Self-Organizing Maps can model the input space utilizing constrained-kmeans and thus ensure that the past knowledge is maintained. Hence, we propose the Dendritic-Self-Organizing Map algorithm consisting of a single layer of Self-Organizing Maps, which extract patterns from specific regions of the input space, and an association matrix that estimates the association between units and labels. The best-matching unit of an input pattern is selected using the maximum cosine similarity rule, while the point-wise mutual information is employed for inferencing. Our method performs unsupervised classification since we do not utilize the labels for targeted weight update. Finally, the results indicate that our algorithm outperforms several state-of-the-art continual learning algorithms on benchmark datasets such as the Split-MNIST and Split-CIFAR-10.
Copyright © DIAS 2013
