We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

The ImageNet Large Scale Visual Recognition Challenge is a benchmark in object category classification and detection on hundreds of object categories and millions of images. The challenge has been run annually from 2010 to present, attracting participation from more than fifty institutions. This paper describes the creation of this benchmark dataset and the advances in object recognition that have been possible as a result. We discuss the challenges of collecting large-scale ground truth annotation, highlight key breakthroughs in categorical object recognition, provide a detailed analysis of the current state of the field of large-scale image classification and object detection, and compare the state-of-the-art computer vision accuracy with human accuracy. We conclude with lessons learned in the 5 years of the challenge, and propose future directions and improvements.

/pdf/imagenet-large-scale-visual-recognition-challenge-caky5nmmix.pdf

ImageNet Large Scale Visual Recognition Challenge

The central building block of convolutional neural networks (CNNs) is the convolution operator, which enables networks to construct informative features by fusing both spatial and channel-wise information within local receptive fields at each layer. A broad range of prior research has investigated the spatial component of this relationship, seeking to strengthen the representational power of a CNN by enhancing the quality of spatial encodings throughout its feature hierarchy. In this work, we focus instead on the channel relationship and propose a novel architectural unit, which we term the “Squeeze-and-Excitation” (SE) block, that adaptively recalibrates channel-wise feature responses by explicitly modelling interdependencies between channels. We show that these blocks can be stacked together to form SENet architectures that generalise extremely effectively across different datasets. We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs at slight additional computational cost. Squeeze-and-Excitation Networks formed the foundation of our ILSVRC 2017 classification submission which won first place and reduced the top-5 error to 2.251 percent, surpassing the winning entry of 2016 by a relative improvement of ${\sim }$ ∼ 25 percent. Models and code are available at https://github.com/hujie-frank/SENet .

/pdf/squeeze-and-excitation-networks-50icgwqgeu.pdf

Squeeze-and-Excitation Networks

We propose a deep learning method for single image super-resolution (SR). Our method directly learns an end-to-end mapping between the low/high-resolution images. The mapping is represented as a deep convolutional neural network (CNN) that takes the low-resolution image as the input and outputs the high-resolution one. We further show that traditional sparse-coding-based SR methods can also be viewed as a deep convolutional network. But unlike traditional methods that handle each component separately, our method jointly optimizes all layers. Our deep CNN has a lightweight structure, yet demonstrates state-of-the-art restoration quality, and achieves fast speed for practical on-line usage. We explore different network structures and parameter settings to achieve trade-offs between performance and speed. Moreover, we extend our network to cope with three color channels simultaneously, and show better overall reconstruction quality.

Image Super-Resolution Using Deep Convolutional Networks

We propose Convolutional Block Attention Module (CBAM), a simple yet effective attention module for feed-forward convolutional neural networks. Given an intermediate feature map, our module sequentially infers attention maps along two separate dimensions, channel and spatial, then the attention maps are multiplied to the input feature map for adaptive feature refinement. Because CBAM is a lightweight and general module, it can be integrated into any CNN architectures seamlessly with negligible overheads and is end-to-end trainable along with base CNNs. We validate our CBAM through extensive experiments on ImageNet-1K, MS~COCO detection, and VOC~2007 detection datasets. Our experiments show consistent improvements in classification and detection performances with various models, demonstrating the wide applicability of CBAM. The code and models will be publicly available.

CBAM: Convolutional Block Attention Module

Full duplex (FD) communication is considered as a promising technology in the development of 5G-advanced and 6G systems as it theoretically doubles the capability of channel. However, this improvement relying on a simultaneously bidirectional manner of communication induces intrinsic self-interference (SI) demanding to be fully cancelled, which is generally intractable. This challenge is minimized in the scenario of integrated access and backhaul (IAB) networks operated in millimeter wave (mmW) band, as its transceivers are stationary and the complexity of SI is greatly reduced by beamforming technique. As a matter of fact, FD can be a pioneering technique to unlock the full potential mmW-based IAB network by releasing its suffering of the half-duplex inefficiency. This article presents the design principle and validation of a practical SI cancellation (SIC) technique in the case of high transmission power class in mmW-based IAB networks. The proposed technique sequentially eliminates SI from the spatial, RF, and digital domains that reduces the SI down to the noise floor. To validate the technique, a prototype system is developed in accordance with 5G IAB specifications and field tests are conducted. Results suggest that the designed mmW SIC transceiver significantly reduces residual-interference to noise ratio (R-INR) to 2.1 dB or less. Additionally, a system-level simulation is conducted in line with 5G IAB evaluation methodology, which explores the potential performance gain of the proposed technique in presence of cross-link interference (CLI). Results indicate that the method could yield a cell throughput gain of about 84%, compared to the current time division duplex deployment. 

Realizing High Power Full Duplex in Millimeter Wave System: Design, Prototype and Results


With the advent of the information age, the iterative speed of software update is gradually accelerating which makes software development severely limited by software testing. Test case prioritization is an effective way to accelerate software testing progress. With the introduction of heuristic algorithm to this task, the processing efficiency of test cases has been greatly improved. However, to overcome the shortcomings of slow convergence speed and easy fall into local optimum, the improved whale optimization algorithm is proposed for test case prioritization. Firstly, a model called n-dimensional directed search space is established for the swarm intelligence algorithm. Secondly, the enhanced whale optimization algorithm is applied to test case prioritization while the backtracking behavior is conducted for individuals when hitting the wall. In addition, a separate storage space for Pareto second optimization is also designed to filter the optimal solutions of the multi-objective tasks. Finally, both single-objective and multi-objective optimization experiments are carried out for open source projects and real-world projects, respectively. The results show that the improved whale optimization algorithm using n-dimensional directed search space is more conducive to the decisions of test case prioritization with fast convergence speed.


/pdf/directed-search-based-on-improved-whale-optimization-lqz90y9i.pdf

Directed Search Based on Improved Whale Optimization Algorithm for Test Case Prioritization

Variational quantum algorithms, which consist of optimal parameterized quantum circuits, are promising for demonstrating quantum advantages in the noisy intermediate-scale quantum (NISQ) era. Apart from classical computational resources, diﬀerent kinds of quantum resources have their contributions in the process of computing, such as information scrambling and entanglement. Characterizing the relation between complexity of speciﬁc problems and quantum resources consumed by solving these problems is helpful for us to understand the structure of VQAs in the context of quantum information processing. In this work, we focus on the quantum approximate optimization algorithm (QAOA), which aims to solve combinatorial optimization problems. We study information scrambling and entanglement in QAOA circuits respectively, and discover that for a harder problem, more quantum resource is required for the QAOA circuit to obtain the solution. We note that in the future, our results can be used to benchmark complexity of quantum many-body problems by information scrambling or entanglement accumulation in the computing process.

Information scrambling and entanglement in quantum approximate optimization algorithm circuits

Microstructure and Mechanical Behavior of Functionally Graded Cemented Carbides with Conifecr High Entropy Alloy Binder

This paper emphasize the importance of the inter-sector self-interference (SI) cancellation to realize the full duplex (FD) for the practical three-sectored cell deployment scenario. We measure the inter-sector SI, the interference coming from the neighboring sector in the same site, using our FD enabled prototype testbed, and then we conduct the system-level simulation to validate the the impact of the inter-sector SI along with the intra-sector SI and cross link interference (CLI) in mmWave integrated access and backhaul (IAB) networks. From our extensive simulation results, the FD operation compared with the half duplex (HD) operation achieves throughput gain of 70% with inter-sector SI cancellation (SIC), while the FD operation achieves throughput gain of 20% without inter-sector SIC. Conclusively, the inter-sector SI should be effectively cancelled out to guarantee the meaningful FD gain over HD.

Chen Qian

Papers

Realizing High Power Full Duplex in Millimeter Wave System: Design, Prototype and Results

Directed Search Based on Improved Whale Optimization Algorithm for Test Case Prioritization

Information scrambling and entanglement in quantum approximate optimization algorithm circuits

Microstructure and Mechanical Behavior of Functionally Graded Cemented Carbides with Conifecr High Entropy Alloy Binder

Measurement and System-level Evaluation of Inter-Sector Self-Interference on Full Duplex in mmWave Band