The model parameters of convolutional neural networks (CNNs) are determined by backpropagation (BP). In this work, we propose an interpretable feedforward (FF) design without any BP as a reference. The FF design adopts a data-centric approach. It derives network parameters of the current layer based on data statistics from the output of the previous layer in a one-pass manner. To construct convolutional layers, we develop a new signal transform, called the Saab (Subspace approximation with adjusted bias) transform. It is a variant of the principal component analysis (PCA) with an added bias vector to annihilate activation's nonlinearity. Multiple Saab transforms in cascade yield multiple convolutional layers. As to fully-connected (FC) layers, we construct them using a cascade of multi-stage linear least squared regressors (LSRs). The classification and robustness (against adversarial attacks) performances of BP-and FF-designed CNNs applied to the MNIST and the CIFAR-10 datasets are compared. Finally, we comment on the relationship between BP and FF designs.
The PointHop method was recently proposed by Zhang et al. for 3D point cloud classification with unsupervised feature extraction. It has an extremely low training complexity while achieving stateof-the-art classification performance. In this work, we improve the PointHop method furthermore in two aspects: 1) reducing its model complexity in terms of the model parameter number and 2) ordering discriminant features automatically based on the cross-entropy criterion. The resulting method is called PointHop++. The first improvement is essential for wearable and mobile computing while the second improvement bridges statistics-based and optimization-based machine learning methodologies. With experiments conducted on the ModelNet40 benchmark dataset, we show that the PointHop++ method performs on par with deep neural network (DNN) solutions and surpasses other unsupervised feature extraction methods.
Abstract-Widespread use of white light-emitting diodes and ubiquitous smart devices offer the opportunity to establish visible light communications (VLC) which has become a hot research topic based on the growing number of publications over the last decade. Camera-based VLC, namely optical camera communications (OCC), provide many unique features when compared with a single photodiode-based system, such as the ability to separate incident light in spatial and color domains. OCC technology represents a promising approach to utilize the benefits of VLC in beyond-5G scenarios and is one of the key technologies of the Internet of Things. Establishing a long communication channel in OCC, as well as non-flickering illumination in using low-frame-rate camera detectors, requires special modulation schemes. This article provides an overview of the principles of three categories of modulation schemes for OCC systems using a low-frame-rate camera detector. In addition, a series of undersampled modulation schemes are proposed and discussed to achieve flicker-free OCC with higher spectral efficiency. In addition, framing structures are designed to solve problems occurring in OCC systems using particular modulation schemes. To evaluate the performance of these modulation schemes, measured bit error rate values are shown. Finally, challenges in the implementation of OCC systems are also outlined.Index Terms -optical camera communications, non-flickering illumination, undersampled phase shift ON-OFF keying, undersampled quadrature-amplitude-modulation, visible light communications, optical wireless communications
-Multi-antenna techniques capable of exploiting the elevation dimension are anticipated to be an important air-interface enhancement targeted to handle the expected growth in mobile traffic.In order to enable the development and evaluation of such multi-antenna techniques, the 3 rd generation partnership project (3GPP) has recently developed a 3-dimensional (3D) channel model.The existing 2-dimensional (2D) channel models do not capture the elevation channel characteristics lending them insufficient for such studies. This article describes the main components of the newly developed 3D channel model and the motivations behind introducing them. One key aspect is the ability to model channels for users located on different floors of a building (at different heights). This is achieved by capturing a user height dependency in modelling some channel characteristics including pathloss, line-of-sight (LOS) probability, etc. In general this 3D channel model follows the framework of WINNERII/WINNER+ while also extending the applicability and the accuracy of the model by introducing some height and distance dependent elevation related parameters.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.