Due to the rapid development of printed circuit board (PCB) design technology, inspection PCB surface defects has become an increasingly critical issue. The classification facilitates root causes detects' identification. As may be intensive, actual should not considered as a binary or multi-category problem. This type problem is called multi-label Recently, one deep learning frameworks, convolutional neural network (CNN) major breakthrough in many areas image processing, especially...

Question generation aims to generate questions according the given context and answer, it has made significant progress in both Chinese English languages. However, research on Tibetan question is still early stages, with key challenges including omission of crucial keywords that render unanswerable. Existing large-scale models do not provide robust support for low-resource languages, such as GPT or BERT. To solve problem, this paper proposes based sentences knowledge graph. The generator...

Convolutional neural network is the most important algorithm in field of deep learning. The traditional convolution usually uses Sigmoid or Relu as activation function, but two sides are saturated, and has a dead zone, which very easy to cause gradient disappearance explosion. In this paper, Mish function introduced into LENET-5 convolutional network, overcomes shortcomings function. At same time, fixed memory step descent method used replace optimization part, improves global convergence algorithm.

The pre-trained language model is trained on large-scale unlabeled text and can achieve state-of-the-art results in many different downstream tasks. However, the current mainly concentrated Chinese English fields. For low resource such as Tibetan, there lack of a monolingual model. To promote development Tibetan natural processing tasks, this paper collects training data from websites constructs vocabulary that cover 99.95% words corpus by using Sentencepiece. Then, we train named TiBERT...

In recent years, the large-scale open Chinese and English question answering (QA) corpora have provided important support for application of deep learning in QA systems. However, low-resource languages, such as Tibetan, it is difficult to construct satisfactory systems, owing lack Tibetan corpora. To solve this problem, paper proposes a corpus generation model, called QuGAN. This model combines Quasi-Recurrent Neural Networks Reinforcement Learning. The used generator Generative Adversarial...

The task of relation extraction is classifying the relations between two entities in a sentence. Distant supervision can automatically align texts based on Knowledge Base without labeled training data. For low-resource language extraction, such as Tibetan, main problem lack In this paper, we propose an improved distant supervised model Piecewise Convolutional Neural Network (PCNN) to expand Tibetan corpus. We add self-attention mechanism and soft-label method decrease wrong labels, use...

This paper proposes a Tibetan automatic word segmentation approach, which is taking the advantage of case-auxiliary words and continuous feature. Meanwhile, we also conduct further investigation on chunk based words. Finally, an experiment performed to verify algorithm proposed in this paper, results prove method effective.

Currently, research on question answering (QA) with deep learning methods is a hotspot in natural language processing. In addition, most of the mainly focused English or Chinese since there are large-scale open corpora, such as WikiQA DoubanQA. However, how to use QA low resource languages, like Tibetan becomes challenge. this paper, we propose hybrid network model for QA, which combines convolutional neural and long short memory (LSTM) extract effective features from small-scale corpora....

English is one of the key subjects basic education in many countries; more and students tend to learn online. This paper takes middle school texts as research object proposes a method test knowledge graph construction. Through acquiring data, preprocessing corpus, designing feature vectors, this realizes extract points from tests based on SVM model construct an graph. It important standardization, automation systematization online learning.

Pre-trained language models are trained on large-scale unsupervised data, and they can fine-tune the model only small-scale labeled datasets, achieve good results. Multilingual pre-trained be multiple languages, understand languages at same time. At present, search mainly focuses rich resources, while there is relatively little research low-resource such as minority public multilingual not work well for languages. Therefore, this paper constructs a named MiLMo that performs better tasks,...

In the process of learning English, students need to do a lot exercises improve English performance. The knowledge points are important students, yet how extract from automatically is difficult, which foundation graph construction for learning. this paper, we use SVM realize extraction junior high school exercises. Firstly, paper obtains amounts question data through analyzing electronic documents, and uses NLP tools segment, POS tagging named entity recognition. Secondly, based on model,...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Generating and editing dynamic 3D head avatars are crucial tasks in virtual reality film production. However, existing methods often suffer from facial distortions, inaccurate movements, limited fine-grained capabilities. To address these challenges, we present DynamicAvatars, a model that generates photorealistic, moving video clips parameters associated with positions expressions. Our approach enables precise through novel prompt-based model, which integrates user-provided prompts guiding...

In addressing challenges within the field of Natural Language Processing (NLP), supervised fine-tuning is an efficient technique that allows pre-trained Large Models to adapt specific tasks. This especially crucial for low-resource languages, such as Tibetan, where demand high-quality datasets particularly pronounced. paper introduces Tibetan Instruction-Following Dataset (TIFD), comprising 11,535 JSON objects, each with four attributes: a unique identifier, instructions, input, and output....

Cross-language text similarity calculation is a critical and fundamental problem in natural language processing. It widely used cross-language research, such as information retrieval. In this paper, we the LDA (Latent Dirichlet Allocation) model to calculate similarities of Tibetan Chinese texts at topic level. Through modelling forecasting, are mapped feature space topics. This method reduced dimensions vector improved speed efficiency computation.

This paper is an exploration to find a way get the person attributes in profiles. Considering those exists large volume of unstructured data, and it very difficult gain short time. So, we use method combing pattern SVM extract attributes. Firstly, collect many raw profiles websites by our configurable crawler. Secondly, statistic methods do pre-processing works include lexical analysis name recognition. Thirdly, build patterns, which can model Also generalize patterns features. Finally,...

This paper researches on some key technologies of Tibetan automatic word segmentation. We propose a segmentation approach, which is taking the advantage case-auxiliary words and continuous feature. Meanwhile, resolution method overlapping ambiguity in proposed, based forward-backward scanning identification improved maximum probability algorithm. Finally, an experiment conducted, results prove algorithm effective.

Tibetan-Chinese named entity extraction can effectively improve the performance of cross language question answering system, information retrieval, machine translation and other researches. In condition no practical Tibetan recognition system model, this paper proposes a method to extract entities based on comparable corpus naturally annotated resources from webs. The main work is in following: (1) construction. (2) Combining sentence length, word matching boundary term features, using...

Tibetan-Chinese named entity extraction is the foundation of cross language information processing, and provides a basis for machine translation retrieval research. In this paper, we use multi-language links Wikipedia to obtain comparable corpus, combine sentence length, word matching boundary words together get parallel sentence. Then extract from corpus in three ways: (1) Extracting Natural labeling information. (2) Acquiring Tibetan entries Chinese entries. (3) Using sequence intersection...