Most deep neural networks (DNNs) consist fundamentally of convolutional and/or fully connected layers, wherein the linear transform can be cast as the product between a filter matrix and a data matrix obtained by arranging feature tensors into columns. Recently proposed deformable butterfly (DeBut) decomposes the filter matrix into generalized, butterfly-like factors, thus achieving network compression orthogonal to the traditional ways of pruning or low-rank decomposition. This work reveals an intimate link between DeBut and a systematic hierarchy of depthwise and pointwise convolutions, which explains the empirically good performance of DeBut layers. By developing an automated DeBut chain generator, we show for the first time the viability of homogenizing a DNN into all DeBut layers, thus achieving extreme sparsity and compression. Various examples and hardware benchmarks verify the advantages of All-DeBut networks. In particular, we show it is possible to compress a PointNet to <5% parameters with <5% accuracy drop, a record not achievable by other compression schemes.

A novel deep neural network (DNN) architecture is proposed wherein the filtering and linear transform are realized solely with product quantization (PQ). This results in a natural implementation via content addressable memory (CAM), which transcends regular DNN layer operations and requires only simple table lookup. Two schemes are developed for the end-to-end PQ prototype training, namely, through angle- and distance-based similarities, which differ in their multiplicative and additive natures with different complexity-accuracy tradeoffs. Even more, the distance-based scheme constitutes a truly multiplier-free DNN solution. Experiments confirm the feasibility of such Product-QuantizEd Content Addressable Memory Network (PECAN), which has strong implication on hardware-efficient deployments especially for in-memory computing.

Recent results have revealed an interesting observation in a trained convolutional neural network (CNN), namely, the rank of a feature map channel matrix remains surprisingly constant despite the input images. This has led to an effective rank-based channel pruning algorithm [23], yet the constant rank phenomenon remains mysterious and unexplained. This work aims at demystifying and interpreting such rank behavior from a frequency-domain perspective, which as a bonus suggests an extremely efficient Fast Fourier Transform (FFT)-based metric for measuring channel importance without explicitly computing its rank. We achieve remarkable CNN channel pruning based on this analytically sound and computationally efficient metric, and adopt it for repetitive pruning to demonstrate robustness via our scheme named Energy-Zoned Channels for Robust Output Pruning (EZCrop), which shows consistently better results than other state-of-the-art channel pruning methods. The codes and Appendix are publicly available at: https://github.com/ruilin0212/EZCrop.

We introduce a new kind of linear transform named Deformable Butterfly (DeBut) that generalizes the conventional butterfly matrices and can be adapted to various input-output dimensions. It inherits the fine-to-coarse-grained learnable hierarchy of traditional butterflies and when deployed to neural networks, the prominent structures and sparsity in a DeBut layer constitutes a new way for network compression. We apply DeBut as a drop-in replacement of standard fully connected and convolutional layers, and demonstrate its superiority in homogenizing a neural network and rendering it favorable properties such as light weight and low inference complexity, without compromising accuracy. The natural complexity-accuracy tradeoff arising from the myriad deformations of a DeBut layer also opens up new rooms for analytical and practical research. The codes and Appendix are publicly available at: https://github.com/ruilin0212/DeBut.

Elasticities in depth, width, kernel size and resolution have been explored in compressing deep neural networks (DNNs). Recognizing that the kernels in a convolutional neural network (CNN) are 4-way tensors, we further exploit a new elasticity dimension along the input-output channels. Specifically, a novel nuclear-norm rank minimization factorization (NRMF) approach is proposed to dynamically and globally search for the reduced tensor ranks during training. Correlation between tensor ranks across multiple layers is revealed, and a graceful tradeoff between model size and accuracy is obtained. Experiments then show the superiority of NRMF over the previous non-elastic variational Bayesian matrix factorization (VBMF) scheme.

Instance segmentation of remote sensing images (RSIs) is an active yet challenging task because of the huge scale variation and arbitrary complex shapes of objects. To address these issues, we propose an adaptive fusion and mask refinement (AFMR) instance segmentation network for RSIs in this paper. More precisely, AFMR consists of an adaptive fusion module to learn multi-scale complementary spatial features in an unsupervised manner, and a content-aware module for segmentation mask refinement. These two modules enable a better feature learning of convolutional neural network and boost the performance by 1.5% compared with the baseline method. In iSAID, a large-scale dataset for RSIs instance segmentation, our AFMR framework achieves the state-of-the-art accuracy, which verifies the superiority of the proposed method.

This paper investigates a novel relaying strategy for cooperative networks over Nakagami-m fading channel, namely selection incremental relaying (SIR), which combines selection relaying (SR) with incremental relaying (IR) based on decode-and-forward (DF). The closed-form expression for the outage probabilities of both single relay and opportunistic relay strategies are derived. Simulation results confirm the presented mathematical analysis and show that the proposed SIR strategy outperforms DF, SR and IR strategy over all signal-to-noise ratio (SNR) values.

Incremental selection amplify-and-forward relaying (ISAF) has the best performance of outage probability among relaying strategies and has been identified as the best protocol. This paper proposes to jointly optimize power allocation (PA) and best-relay position based on cooperative diversity for ISAF in order to minimize outage probability. The closed-form solution to the adaptive allocation algorithm based on high signal-to-noise ratio (SNR) approximation of outage probability is presented, assuming maximum ratio combining (MRC) at the destination. Simulation results have shown that the system performance of outage probability with the proposal outperforms the fixed PA algorithms, and much less power is needed to achieve the same outage probability at the same effective rate. The power saving can be as much as 9 dB compared to the equal power allocation scheme.