Automatic few-shot font generation (AFFG), aiming at generating new fonts with only a few glyph references, reduces the labor cost of manually designing fonts. However, traditional AFFG paradigm style-content disentanglement cannot capture diverse local details different So, many component-based approaches are proposed to tackle this problem. The issue is that they usually require special pre-defined components, e.g., strokes and radicals, which infeasible for languages. In paper, we present...

With the exponential growth of data created at network edge, decentralized and Gossip-based training deep learning (DL) models on edge computing (EC) gains tremendous research momentum, owing to its capability learn from resource-strenuous nodes with limited connectivity. Today's devices are extremely heterogeneous, e.g., hardware software stacks, result in high performance variation time inducing extra delay synchronize converge. The large body prior art accelerates DL, being or model...

Recently academia and industry has growing interest in the sixth generation network, which aims to support a rich range of applications with higher capacity greater coverage than existing 5G connections. One such promising that can benefit from 6G is Decentralised Federated Learning, privacy-preserving machine learning paradigm. Also, it relies heavily on peer-to-peer mobile connection among edge devices, instead powerful central server cloud. However, data device heterogeneity, highly...

Modern distributed engines are increasingly deployed to accelerate large-scaled deep learning (DL) training jobs. While the parallelism of workers/nodes promises scalability, computation and communication overheads underlying iterative solving algorithms, e.g., stochastic gradient decent, unfortunately become bottleneck for DL Existing approaches address such limitations by designing more efficient synchronization algorithms model compressing techniques, but do not adequately issues relating...

The core of many large-scale machine learning (ML) applications, such as neural networks (NN), support vector (SVM), and convolutional network (CNN), is the training algorithm that iteratively updates model parameters by processing massive datasets. From a plethora studies aiming at accelerating ML, being data parallelization parameter server, prevalent assumption all points are equivalently relevant to updating. In this article, we challenge proposing criterion measure point's effect on...

The rapid development of the Internet leads to growth network information, we call it information explosion. is full and difficult for users find this useful knowledge ocean. Web has become world's largest repository, there an urgent need efficient access valuable vast amounts web information. purpose paper study process algorithm analysis text mining system based on artificial intelligence. This presents document feature acquisition genetic algorithm. Selecting suitable features important...

The current super-resolution methods cannot fully exploit the global and local information of original low-resolution image, resulting in loss some information. In order to solve problem, we propose a multiscale residual dense network (MRDN) for image super-resolution. This is constructed based on network. It can integrate avoid losing too much deep level network, while extracting more under different receptive fields. addition, reduce redundancy parameters MRDN, further develop lightweight...

Chaotic systems are sensitive to certain signals and immune noise at the same time, properties of which make it potential application in signal detection equipment fault diagnosis. In this paper, some Virtual Instruments (VIs) designed implement Duffing oscillator is used for weak detection. The basic design idea VIs presented diagrams shown paper. At last, results simulated real given. It that use VI field advantageous.

Automatic few-shot font generation (AFFG), aiming at generating new fonts with only a few glyph references, reduces the labor cost of manually designing fonts. However, traditional AFFG paradigm style-content disentanglement cannot capture diverse local details different So, many component-based approaches are proposed to tackle this problem. The issue is that they usually require special pre-defined components, e.g., strokes and radicals, which infeasible for languages. In paper, we present...

With the advent of digital economy era, an increasing number public infrastructure projects are moving towards smart, digital, and technological solutions. The city manhole covers is continuously increasing, but problems such as loss, damage, displacement these frequently occur. Traditional manual inspections unable to comprehensively monitor cover information address anomalies in a timely manner, which hampers urban management poses threats pedestrian safety.This project combines Internet...

In view of the development status evaluation system scientific, this paper proposes an algorithm based on weighted Dijkstra classification path model. It takes existing scientific and technological achievements as model input, obtains division achievements. According to a variety criteria, determines comprehensive index data core nodes according newly generated set achievements, it reference data, performance rating each achievement. The whole process is carried out in standardized,...

The application and reward of electric power scientific technological achievements has been regarded as an important indicator the year-end assessment enterprises departments, which plays role in development personal career. Therefore, pay special attention to application, spend a lot human material resources focus on work every year end year, with low efficiency effectiveness. Under this demand, paper puts forward efficient way extract key content achievements. Through series operations de...

In recent years, small objects detection has received extensive attention from scholars for its important value in application. Some effective methods have been proposed. However, the data collected real scenes are often foggy images, so models trained with these difficult to extract discriminative object features such images. addition, existing algorithms ignore texture information and high-level semantic of tiny objects, which limits improvement performance. Aiming at above problems, this...

Snowflakes in the image usually reduce visibility of background and affect quality. Nowadays, most single snow removal tasks have made some progress through design more complex deep learning models, but them use specific datasets, generalization ability is not enough. We deal with problem from point view datasets propose a effective method to synthesize images. The full Bayesian generation model established, which layer parameterized as generator input latent variables. To resolve this...