Deep features for CBIR with scarce data using Hebbian learning

Lagani G., Bacciu D., Gallicchio C., Falchi F., Gennaro C., Amato G.

Computer Science - Machine Learning  Semi-Supervised  Neural Networks  Content Based Image Retrieval  Deep Learning  Content based image retrieval  Deep learning  Neural and Evolutionary Computing (cs.NE)  Computer Vision and Pattern Recognition (cs.CV)  FOS: Computer and information sciences  Hebbian Learning  Bio-inspired  Neural networks  Semi-supervised  Hebbian learning  Bio-Inspired  Computer Science - Neural and Evolutionary Computing  Machine Learning (cs.LG)  Computer Science - Computer Vision and Pattern Recognition 

Features extracted from Deep Neural Networks (DNNs) have proven to be very effective in the context of Content Based Image Retrieval (CBIR). Recently, biologically inspired Hebbian learning algorithms have shown promises for DNN training. In this contribution, we study the performance of such algorithms in the development of feature extractors for CBIR tasks. Specifically, we consider a semi-supervised learning strategy in two steps: first, an unsupervised pre-training stage is performed using Hebbian learning on the image dataset; second, the network is fine-tuned using supervised Stochastic Gradient Descent (SGD) training. For the unsupervised pre-training stage, we explore the nonlinear Hebbian Principal Component Analysis (HPCA) learning rule. For the supervised fine-tuning stage, we assume sample efficiency scenarios, in which the amount of labeled samples is just a small fraction of the whole dataset. Our experimental analysis, conducted on the CIFAR10 and CIFAR100 datasets, shows that, when few labeled samples are available, our Hebbian approach provides relevant improvements compared to various alternative methods.

Metrics