Federated learning (FL) enables massive distributed Information and Communication Technology (ICT) devices to learn a global consensus model without any participants revealing their own data to the central server. However, the practicality, communication expense and non-independent and identical distribution (Non-IID) data challenges in FL still need to be concerned. In this work, we propose the Semi-Federated Learning (Semi-FL) which differs from the FL in two aspects, local clients clustering and in-cluster training. A sequential training manner is designed for our in-cluster training in this paper which enables the neighboring clients to share their learning models. The proposed Semi-FL can be easily applied to future mobile communication networks and require less up-link transmission bandwidth. Numerical experiments validate the feasibility, learning performance and the robustness to Non-IID data of the proposed Semi-FL. The Semi-FL extends the existing potentials of FL.

The renewed interest in convolutional neural networks [12, 15] in computer vision and signal processing has lead to a major leap in generalization performance on common task benchmarks, supported by the recent advances in graphical processing hardware and the collection of huge labelled datasets for training and evaluation. Convolutional neural networks pose major a challenge to statistical learning theory. First and foremost a convolutional network learns from data, jointly, both a feature representation through its hidden layers and a prediction function through its ultimate layer. A convolutional neural network implements a function unfolding as a composition of basic functions (respectively nonlinearity, convolution, and pooling), which appear to model well visual information in images. Yet the relevant function spaces to analyze their statistical performance remain unclear. The analysis of convolutional neural networks (CNNs) has been an active research topic. Different viewpoints have been developed. A straightforward viewpoint is to dismiss completely the grid-or latticestructure of images and analyze a multi-layer perceptron (MLP) instead acting on vectorized images, which has the downside the set aside the most interesting property CNNs which is to model well images that is data with a 2D lattice structure.

Automated medical image classification with convolutional neural networks (CNNs) has great potential to impact healthcare, particularly in resource-constrained healthcare systems where fewer trained radiologists are available. However, little is known about how well a trained CNN can perform on images with the increased noise levels, different acquisition protocols, or additional artifacts that may arise when using low-cost scanners, which can be underrepresented in datasets collected from well-funded hospitals. In this work, we investigate how a model trained to triage head computed tomography (CT) scans performs on images acquired with reduced x-ray tube current, fewer projections per gantry rotation, and limited angle scans. These changes can reduce the cost of the scanner and demands on electrical power but come at the expense of increased image noise and artifacts. We first develop a model to triage head CTs and report an area under the receiver operating characteristic curve (AUROC) of 0.77. We then show that the trained model is robust to reduced tube current and fewer projections, with the AUROC dropping only 0.65% for images acquired with a 16x reduction in tube current and 0.22% for images acquired with 8x fewer projections. Finally, for significantly degraded images acquired by a limited angle scan, we show that a model trained specifically to classify such images can overcome the technological limitations to reconstruction and maintain an AUROC within 0.09% of the original model.

In this paper, we proposed two modified neural network architectures based on SFANet and SegNet respectively for accurate and efficient crowd counting. Inspired by SFANet, the first model is attached with two novel multi-scale-aware modules, called ASSP and CAN. This model is called M-SFANet. The encoder of M-SFANet is enhanced with ASSP containing parallel atrous convolution with different sampling rates and hence able to extract multi-scale features of the target object and incorporate larger context. To further deal with scale variation throughout an input image, we leverage contextual module called CAN which adaptively encodes the scales of the contextual information. The combination yields an effective model for counting in both dense and sparse crowd scenes. Based on the SFANet's decoder structure, M-SFANet's decoder has dual paths, for density map generation and attention map generation. The second model is called M-SegNet. For M-SegNet, we simply change bilinear upsampling used in SFANet to max unpooling originally from SegNet and propose the faster model while providing competitive counting performance. Designed for high-speed surveillance applications, M-SegNet has no additional multi-scale-aware module in order to not increase the complexity. Both models are encoder-decoder based architectures and end-to-end trainable. We also conduct extensive experiments on four crowd counting datasets and one vehicle counting dataset to show that these modifications yield algorithms that could outperform some state-of-the-art crowd counting methods.

It explains how they are used in computer vision and NLP and the fundamental differences between the two types of artificial neural networks.

The world has been revolving much around the terms "Machine Learning" and "Deep Learning" recently. With or without our knowledge every day we are using these technologies. There are tons of other applications too. No wonder why "Deep Learning" and "Machine Learning along with Data Science" are the most sought after talent in the technology world now a days. But the problem is that, when you think about learning these technologies, a misconception that lots of maths, statistics, complex algorithms and formulas needs to be studied prior to that.

Machine Learning became one of the hottest domain of Computer Science. Each larger company is either applying Machine Learning or thinking about doing so soon to solve their problems and understand their data sets. That means it's time to learn about Machine Learning, especially if you're looking for new Computer Science challenges. A great way to do that is to read a couple of books. If you're just getting started with Machine Learning definitely read this book: Introduction to Machine Learning with Python is a gentle introduction into machine learning.

Nature Machine Intelligence published a joint paper from researchers at Intel Labs and Cornell University demonstrating the ability of Intel's neuromorphic test chip, Loihi, to learn and recognize 10 hazardous chemicals, even in the presence of significant noise and occlusion. The work demonstrates how neuromorphic computing could be used to detect smells that are precursors to explosives, narcotics and more. Loihi learned each new odor from a single example without disrupting the previously learned smells, requiring up to 3000x fewer training samples per class compared to a deep learning solution and demonstrating superior recognition accuracy. The research shows how the self-learning, low-power, and "brain-like" properties of neuromorphic chips – combined with algorithms derived from neuroscience – could be the answer to creating "electronic nose" systems that recognize odors under real-world conditions more effectively than conventional solutions. "We are developing neural algorithms on Loihi that mimic what happens in your brain when you smell something," said Nabil Imam, senior research scientist in Intel's Neuromorphic Computing Lab.

Today Supermicro announced the industry's broadest portfolio of validated NGC-Ready systems optimized to accelerate AI and deep learning applications. Supermicro is highlighting many of these systems today at the Supermicro GPU Live Forum in conjunction with NVIDIA GTC Digital. Supermicro NGC-Ready systems allow customers to train AI models using NVIDIA V100 Tensor Core GPUs and to perform inference using NVIDIA T4 Tensor Core GPUs. NGC hosts GPU-optimized software containers for deep learning, machine learning and HPC applications, pre-trained models, and SDKs that can run anywhere the Supermicro NGC-Ready systems are deployed whether in data centers, cloud, edge micro-datacenters, or in distributed remote locations as environment-resilient and secured NVIDIA-Ready for Edge servers powered by the NVIDIA EGX intelligent edge platform. With over 26 years of experience delivering state-of-the-art computing solutions, Supermicro systems are the most power-efficient, the highest performing, and the best value," said Charles Liang, CEO and president of Supermicro. "With support for fast networking and storage, as well as NVIDIA GPUs, our Supermicro NGC-Ready systems are the most scalable and reliable servers to support AI. Customers can run their AI infrastructure with the highest ROI." Supermicro currently leads the industry with the broadest portfolio of NGC-Ready Servers optimized for data center and cloud deployments and is continuing to expand its portfolio. In addition, the company offers five validated NGC-Ready for Edge servers (EGX) optimized for edge inferencing applications. NVIDIA's container registry, NGC, enables superior performance for deep learning frameworks and pre-trained AI models with state-of-the-art accuracy," said Ian Buck, vice president and general manager of Accelerated Computing at NVIDIA.

Convolutional Neural Networks which are the backbones of most of the Computer Vision Applications like Self-Driving Cars, Facial Recognition Systems etc are a special kind of Neural Network architectures in which the basic matrix-multiplication operation is replaced by a convolution operation. They specialize in processing data which has a grid-like topology. Examples include time-series data and image-data which can be thought of as a 2-D grid of pixels. The Convolutional Neural Networks was first introduced by Fukushima by the name Neocognitron in 1980. It was inspired by the hierarchical model of the nervous system as proposed by Hubel and Weisel.