Semi-supervised deep rule-based approach for image classification

Abstract

In this paper, a semi-supervised learning approach based on a deep rule-based (DRB) classifier is introduced. With its unique prototype-based nature, the semi-supervised DRB (SSDRB) classifier is able to generate human interpretable IF…THEN… rules through the semi-supervised learning process in a selforganising and highly transparent manner. It supports online learning on a sample-by-sample basis or on a chunk-by-chunk basis. It is also able to perform classification on out-of-sample images. Moreover, the SSDRB classifier can learn new classes from unlabelled images in an active way becoming dynamically self-evolving. Numerical examples based on large-scale benchmark image sets demonstrate the strong performance of the proposed SSDRB classifier as well as its distinctive features compared with the "state-of-the-art" approaches.

Toward Anthropomorphic Machine Learning

“…As one of the recent advances in the field of machine learning, the recently introduced Deep Rule-Based (DRB) systems [11], [12] provide a possible solution to the problem. The DRB system is a principally new generic approach that combines the advantages of both, the selforganizing non-parametric FRB system and the multi-layer structure of DLNNs [11].…”

“…In practice, these prototypes are the local maxima of the multi-modal typicality/data density automatically extracted from the observed data as described in [2]. Prototypes are instrumental to the DRB approach, they set this approach apart from the mainstream approaches including DLNNs [2], [11], [12]. These prototypes are identified based on the disclosed ensemble properties and mutual distribution of the data using parameter-free operators through a fully autonomous, online, non-parametric, noniterative and "one-pass" learning process [2].…”

“…Most existing machine learning methods require a huge amount of training data, while the DRB system can learn even from a single example, that is, to start "from scratch". The non-iterative online autonomous learning algorithm further enables the DRB system to learn lifelong with newly observed samples continuously even after training/deployment, namely, to be "evolving" [9], [12].…”

“…The system can easily distinguish new samples with unfamiliar patterns based on the lower degrees of membership/satisfaction. This further enables the system to learn continuously without or with very little human intervention or supervision by exploiting prototype-based semi-supervised and active learning algorithms [12]. Then, the DRB system can learn new prototypes as well as new classes actively in an autonomous manner "on the fly", and users can be (optionally) involved in double-checking the newly gained information afterwards.…”

Actively Semi-Supervised Deep Rule-based Classifier Applied to Adverse Driving Scenarios

“…This paper presents a multi-layer neuro-fuzzy model called Deep Rule-Based (DRB) system [9], [20]. The DRB approach has as a learning engine composed of a massively parallel set of 0-order fuzzy rules, which are able to selfadapt and provide transparent and human understandable IF ... THEN representation [21].…”