Two-Dimensional Broad Learning System for Data Analytic

Recently, a novel learning algorithm called the broad learning system (BLS) has been established by using the theory of pseudoinverse and compressed sensing technology. As an alternative to deep learning, BLS has gained extensive attention in data analytics. Unlike the existing deep structures based learning tools (such as deep neural networks), it is able to rapidly achieve incremental learning, and does not require tedious retraining to remodel the system. However, given image data modeling tasks, 1-D BLS requires a conventional vectorization operation, the dimension of input will be quite large for high resolution image. This operation may result in huge computational complexity, thereby increasing the time of model construction. To resolve this issue, a 2-D version of BLS, namely 2-D-BLS, is first proposed to fast construct randomized learner models by using matrix inputs. Specifically, the proposed method employs left and right projecting vectors to replace the usual high dimensional input weight. Moreover, the theoretical analyses on the superiority of 2-D-BLS against BLS are presented in terms of the algorithm complexity and the difference of random parameter space. Empirical results on MNIST dataset, NORB 3-D object image recognition dataset, ORL dataset, YaleB dataset, and handwritten digit dataset demonstrate that the proposed 2-D-BLS outperforms the existing extensions of BLS and state-of-the-art 2-D machine learning methods in terms of the learning efficiency, showing good potential of 2-D-BLS for image data analytics.

Impact Statement—as a novel learning tool, BLS has achieved great success in many fields. Image recognition is one of the most widely used fields of artificial intelligence. But, BLS is a 1-D model, the vectorization operation is necessary to convert matrix input data into a row or a column vector. This operation will lead to “explosion of dimensionality” and increase the time of model construction when it is coming to image data analytic with 2-D input. This article first proposes a 2-D extension of BLS for high resolution image data analytic. The proposed method can not only avoid the “explosion of dimensionality,” but also construct the whole structure with matrix-inputs rapidly and dynamically. The improvement of this method is expected to provide a novel and great choice for the field of image recognition, such as face recognition, industrial inspection, behavior recognition, etc., in the future.