Accelerating materials discovery using machine learning

In this paper, we propose GlyphGAN: style-consistent font generation based on generative adversarial networks (GANs). GANs are a framework for learning a generative model using a system of two neural networks competing with each other. One network generates synthetic images from random input vectors, and the other discriminates between synthetic and real images. The motivation of this study is to create new fonts using the GAN framework while maintaining style consistency over all characters. In GlyphGAN, the input vector for the generator network consists of two vectors: character class vector and style vector. The former is a one-hot vector and is associated with the character class of each sample image during training. The latter is a uniform random vector without supervised information. In this way, GlyphGAN can generate an infinite variety of fonts with the character and style independently controlled. Experimental results showed that fonts generated by GlyphGAN have style consistency and diversity different from the training images without losing their legibility.

GlyphGAN: Style-consistent font generation based on generative adversarial networks

Generating personal handwriting fonts with large amounts of characters is a boring and time-consuming task. For example, the official standard GB18030-2000 for commercial font products consists of 27,533 Chinese characters. Consistently and correctly writing out such huge amounts of characters is usually an impossible mission for ordinary people. To solve this problem, we propose a system, EasyFont, to automatically synthesize personal handwriting for all (e.g., Chinese) characters in the font library by learning style from a small number (as few as 1%) of carefully-selected samples written by an ordinary person. Major technical contributions of our system are twofold. First, we design an effective stroke extraction algorithm that constructs best-suited reference data from a trained font skeleton manifold and then establishes correspondence between target and reference characters via a non-rigid point set registration approach. Second, we develop a set of novel techniques to learn and recover users’ overall handwriting styles and detailed handwriting behaviors. Experiments including Turing tests with 97 participants demonstrate that the proposed system generates high-quality synthesis results, which are indistinguishable from original handwritings. Using our system, for the first time, the practical handwriting font library in a user’s personal style with arbitrarily large numbers of Chinese characters can be generated automatically. It can also be observed from our experiments that recently-popularized deep learning based end-to-end methods are not able to properly handle this task, which implies the necessity of expert knowledge and handcrafted rules for many applications.

EasyFont: A Style Learning-Based System to Easily Build Your Large-Scale Handwriting Fonts

Aesthetic evaluation of Chinese calligraphy is one of the most challenging tasks in Artificial Intelligence. This paper attempts to solve this problem by proposing a number of aesthetic feature representations and feeding them into Artificial Neural Networks. Specifically, 22 global shape features are presented to describe a given handwritten Chinese character from different aspects according to classical calligraphic rules, and a new 10-dimensional feature vector is introduced to represent the component layout information using sparse coding. Moreover, a Chinese Handwriting Aesthetic Evaluation Database (CHAED) is also built by collecting 1000 Chinese handwriting images with diverse aesthetic qualities and inviting 33 subjects to evaluate the aesthetic quality for each calligraphic image. Finally, back propagation neural networks are constructed with the concatenation of the proposed features as input and then trained on our CHAED database for the aesthetic evaluation of Chinese calligraphy. Experimental results demonstrate that the proposed AI system provides a comparable performance with human evaluation. Through our experiments, we also compare the importance of each individual feature and reveal the relationship between our aesthetic features and the aesthetic perceptions of human beings.

/pdf/aesthetic-visual-quality-evaluation-of-chinese-handwritings-1tvljc9xrj.pdf

Aesthetic visual quality evaluation of Chinese handwritings

The Chinese stroke evaluation and generation systems required in an autonomous calligraphy robot play a crucial role in producing high-quality writing results with good diversity. These systems often suffer from inefficiency and non-optima despite of intensive research effort investment by the robotic community. This paper proposes a new learning system to allow a robot to automatically learn to write Chinese calligraphy effectively. In the proposed system, the writing quality evaluation subsystem assesses written strokes using a convolutional auto-encoder network (CAE), which enables the generation of aesthetic strokes with various writing styles. The trained CAE network effectively excludes poorly written strokes through stroke reconstruction, but guarantees the inheritance of information from well-written ones. With the support of the evaluation subsystem, the writing trajectory model generation subsystem is realized by multivariate normal distributions optimized by differential evolution (DE), a type of heuristic optimization search algorithm. The proposed approach was validated and evaluated using a dataset of nine stroke categories; high-quality written strokes have been resulted with good diversity which shows the robustness and efficacy of the proposed approach and its potential in autonomous action-state space exploration for other real-world applications.

/pdf/a-data-driven-robotic-chinese-calligraphy-system-using-34ups34hlj.pdf

A data-driven robotic Chinese calligraphy system using convolutional auto-encoder and differential evolution

K-nearest neighbor (KNN) is a classic classifier, which is simple and effective. Adaboost is a combination of several weak classifiers as a strong classifier to improve the classification effect. These two classifiers have been widely used in the field of machine learning. In this paper, based on information fuzzy granulation, KNN and Adaboost, we propose two algorithms, a fuzzy granule K-nearest neighbor (FGKNN) and a boosted fuzzy granule K-nearest neighbor (BFGKNN), for classification. By introducing granular computing, we normalize the process of solving problem as a structured and hierarchical process. Structured information processing is focused, so the performance including accuracy and robust can be enhanced to data classification. First, a fuzzy set is introduced, and an atom attribute fuzzy granulation is performed on samples in the classified system to form fuzzy granules. Then, a fuzzy granule vector is created by multiple attribute fuzzy granules. We design the operators and define the measure of fuzzy granule vectors in the fuzzy granule space. And we also prove the monotonic principle of the distance of fuzzy granule vectors. Furthermore, we also give the definition of the concept of K-nearest neighbor fuzzy granule vector and present FGKNN algorithm and BFGKNN algorithm. Finally, we compare the performance among KNN, Back Propagation Neural Network (BPNN), Support Vector Machine (SVM), Logistic Regression (LR), FGKNN and BFGKNN on UCI data sets. Theoretical analysis and experimental results show that FGKNN and BFGKNN have better performance than that of the methods mentioned above if the appropriate parameters are given.

Boosted K-nearest neighbor classifiers based on fuzzy granules

Computationally evaluating and synthesizing Chinese calligraphy

Granular computing has advantage of knowledge discovery for complex data. In the paper, we present Fuzzy Granular Hyperplane Classifiers (FGHCs) for data classification from a new angle of Granular Computing. First, we introduce a fuzzy granular hyperplane concept by defining fuzzy granule, fuzzy granular vector, metrics and operators. Next, for binary classification problem, we present solving optimal fuzzy granular hyperplane through evolution strategy; the learning algorithm of parameters and the prediction algorithm of instances are also proposed. Finally, a multi-classification prediction model is designed by combining a set of Fuzzy Granular Hyperplane Classifiers based on vote strategy. In order to evaluate performance, we employed 10-fold cross validation to verify on UCI dataset and Alzheimer's Disease Voice dataset. Theoretical analysis and experiments demonstrated that FGHCs have good performance.

/pdf/fuzzy-granular-hyperplane-classifiers-4zwtf22qrf.pdf

Fuzzy Granular Hyperplane Classifiers

In this paper, we design a recognition system of the handwritten numerals and English characters based on BP neural network. In the system, we first make some preprocess to the image. Secondly, we extract the structural and statistical features of the image. Thirdly, we train a model on the data sets via BP neural network. Finally, we can predict the test image using the trained model. The experiments show that the handwritten numbers recognition rate can reach more than 94 % and the handwritten English characters is 44 % respectively. Besides that, the recognition rate of Handwritten number and English together is 78 %. It has been proved that our method is effective and robust.

Handwritten Numbers and English Characters Recognition System

The traditional 2D convolution approaches are inappropriate for capturing information about local objects and their scale changes. To better depict 3D structures, the fundamental concept behind it is to mix depth information with RGB data to understand 3D space via depth estimations for creating depth maps of 2D images. However, it places a great deal of reliance on the precision of depth estimate, which results in subpar performance. In this paper, a monocular 3D object detection algorithm is proposed to resolve the issues mentioned earlier. The algorithm depends on the estimation of depth, which is essential for accurately detecting objects using a single camera. To begin with, a Normalization-based Attention Module has been deployed to a backbone network with two branches to extract RGB and depth maps separately. To merge the RGB and depth map features, a depth-guided dynamic local convolutional module is subsequently implemented. Depth map learns exemplar kernels from RGB images using a weight learning submodule and dilation convolution. Then it acts locally on the corresponding RGB images to solve the scale-sensitive problem of objects. By utilizing this approach, it becomes possible to extract feature representations from 2D images and map them onto 3D space, minimizing the disparity between the 2D and 3D representations. Experiments indicate that under the moderate setting of the class car - which is the most significant level of the KITTI dataset - our method outperforms other cutting-edge algorithms.

Yuping Song

Papers

Boosted K-nearest neighbor classifiers based on fuzzy granules

Computationally evaluating and synthesizing Chinese calligraphy

Fuzzy Granular Hyperplane Classifiers

Handwritten Numbers and English Characters Recognition System

An Algorithm on Monocular 3D Object Detection Based on Depth Estimation