Search

Your search keyword '"Liu, Kaiyuan"' showing total 594 results
Start Over Author "Liu, Kaiyuan"
594 results on '"Liu, Kaiyuan"'

1. Moyun: A Diffusion-Based Model for Style-Specific Chinese Calligraphy Generation

Author

Liu, Kaiyuan, Mei, Jiahao, Zhang, Hengyu, Zhang, Yihuai, Wu, Xingjiao, Dong, Daoguo, and He, Liang

Subjects
Computer Science - Computer Vision and Pattern Recognition, Computer Science - Artificial Intelligence
Abstract

Although Chinese calligraphy generation has achieved style transfer, generating calligraphy by specifying the calligrapher, font, and character style remains challenging. To address this, we propose a new Chinese calligraphy generation model 'Moyun' , which replaces the Unet in the Diffusion model with Vision Mamba and introduces the TripleLabel control mechanism to achieve controllable calligraphy generation. The model was tested on our large-scale dataset 'Mobao' of over 1.9 million images, and the results demonstrate that 'Moyun' can effectively control the generation process and produce calligraphy in the specified style. Even for calligraphy the calligrapher has not written, 'Moyun' can generate calligraphy that matches the style of the calligrapher.

Published
2024

2. Take its Essence, Discard its Dross! Debiasing for Toxic Language Detection via Counterfactual Causal Effect

Author

Lu, Junyu, Xu, Bo, Zhang, Xiaokun, Liu, Kaiyuan, Zhang, Dongyu, Yang, Liang, and Lin, Hongfei

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence
Abstract

Current methods of toxic language detection (TLD) typically rely on specific tokens to conduct decisions, which makes them suffer from lexical bias, leading to inferior performance and generalization. Lexical bias has both "useful" and "misleading" impacts on understanding toxicity. Unfortunately, instead of distinguishing between these impacts, current debiasing methods typically eliminate them indiscriminately, resulting in a degradation in the detection accuracy of the model. To this end, we propose a Counterfactual Causal Debiasing Framework (CCDF) to mitigate lexical bias in TLD. It preserves the "useful impact" of lexical bias and eliminates the "misleading impact". Specifically, we first represent the total effect of the original sentence and biased tokens on decisions from a causal view. We then conduct counterfactual inference to exclude the direct causal effect of lexical bias from the total effect. Empirical evaluations demonstrate that the debiased TLD model incorporating CCDF achieves state-of-the-art performance in both accuracy and fairness compared to competitive baselines applied on several vanilla models. The generalization capability of our model outperforms current debiased models for out-of-distribution data.

Published
2024

3. CosmicMan: A Text-to-Image Foundation Model for Humans

Author

Li, Shikai, Fu, Jianglin, Liu, Kaiyuan, Wang, Wentao, Lin, Kwan-Yee, and Wu, Wayne

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

We present CosmicMan, a text-to-image foundation model specialized for generating high-fidelity human images. Unlike current general-purpose foundation models that are stuck in the dilemma of inferior quality and text-image misalignment for humans, CosmicMan enables generating photo-realistic human images with meticulous appearance, reasonable structure, and precise text-image alignment with detailed dense descriptions. At the heart of CosmicMan's success are the new reflections and perspectives on data and models: (1) We found that data quality and a scalable data production flow are essential for the final results from trained models. Hence, we propose a new data production paradigm, Annotate Anyone, which serves as a perpetual data flywheel to produce high-quality data with accurate yet cost-effective annotations over time. Based on this, we constructed a large-scale dataset, CosmicMan-HQ 1.0, with 6 Million high-quality real-world human images in a mean resolution of 1488x1255, and attached with precise text annotations deriving from 115 Million attributes in diverse granularities. (2) We argue that a text-to-image foundation model specialized for humans must be pragmatic -- easy to integrate into down-streaming tasks while effective in producing high-quality human images. Hence, we propose to model the relationship between dense text descriptions and image pixels in a decomposed manner, and present Decomposed-Attention-Refocusing (Daring) training framework. It seamlessly decomposes the cross-attention features in existing text-to-image diffusion model, and enforces attention refocusing without adding extra modules. Through Daring, we show that explicitly discretizing continuous text space into several basic groups that align with human body structure is the key to tackling the misalignment problem in a breeze., Comment: Accepted by CVPR 2024. The supplementary material is included. Project Page: https://cosmicman-cvpr2024.github.io

Published
2024

8. Market Analysis of the Game Industry Based on Survey Questionnaire

Author

Liu, Kaiyuan, Sun, Tianyi, Appolloni, Andrea, Series Editor, Caracciolo, Francesco, Series Editor, Ding, Zhuoqi, Series Editor, Gogas, Periklis, Series Editor, Huang, Gordon, Series Editor, Nartea, Gilbert, Series Editor, Ngo, Thanh, Series Editor, Striełkowski, Wadim, Series Editor, Dou, Peng, editor, and Zhang, Keying, editor

Published
2024
Full Text
View/download PDF

9. Time Scale of the Scour Process Around Monopile: A Case Study at Southwest Offshore Wind Farm, South Korea

Author

Zhu, Wenlong, Gu, Chen, Xu, Haibin, Liu, Kaiyuan, Zhou, Yuanxing, Ceccarelli, Marco, Series Editor, Agrawal, Sunil K., Advisory Editor, Corves, Burkhard, Advisory Editor, Glazunov, Victor, Advisory Editor, Hernández, Alfonso, Advisory Editor, Huang, Tian, Advisory Editor, Jauregui Correa, Juan Carlos, Advisory Editor, Takeda, Yukio, Advisory Editor, and Li, Shaofan, editor

Published
2024
Full Text
View/download PDF

15. Denoised Internal Models: a Brain-Inspired Autoencoder against Adversarial Attacks

Author

Liu, Kaiyuan, Li, Xingyu, Lai, Yurui, Zhang, Ge, Su, Hang, Wang, Jiachen, Guo, Chunxu, Guan, Jisong, and Zhou, Yi

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Despite its great success, deep learning severely suffers from robustness; that is, deep neural networks are very vulnerable to adversarial attacks, even the simplest ones. Inspired by recent advances in brain science, we propose the Denoised Internal Models (DIM), a novel generative autoencoder-based model to tackle this challenge. Simulating the pipeline in the human brain for visual signal processing, DIM adopts a two-stage approach. In the first stage, DIM uses a denoiser to reduce the noise and the dimensions of inputs, reflecting the information pre-processing in the thalamus. Inspired from the sparse coding of memory-related traces in the primary visual cortex, the second stage produces a set of internal models, one for each category. We evaluate DIM over 42 adversarial attacks, showing that DIM effectively defenses against all the attacks and outperforms the SOTA on the overall robustness., Comment: 16 pages, 3 figures

Published
2021
Full Text
View/download PDF

30. Research on the application of time shifting aeromagnetic method in detecting coal mine burning areas

Author

YU Yongning, LI Xiongwei, SHI Lei, LIU Kaiyuan, GUO Jianlei, and MA Guoqing

Subjects
coal spontaneous combustion, burning area, time-shifting aeromagnetic method, mesh generation, magnetic anomaly inversion, Mining engineering. Metallurgy, TN1-997
Abstract

The spontaneous combustion of coal seams leads to the formation of magnetic minerals in the overlying strata, exhibiting high magnetic anomaly features, providing a physical prerequisite for the magnetic method to detect the burning area. The aeromagnetic method has achieved good results in detecting coal mine burning areas, but it cannot effectively detect the development trend of coal mine burning areas. In order to solve the above problems, based on the aeromagnetic method method, a time-shifting aeromagnetic method is proposed. It involves conducting two aeromagnetic detections within a certain time interval. Based on the difference between the two aeromagnetic inversion results, the features of the coal mine burning area over time are determined. It achieves the goal of effectively detecting the distribution range and development trend of the coal mine burning area. In order to balance the terrain fitting effect and inversion calculation efficiency in undulating areas, a composite mesh generation method of regular and irregular grids is adopted. The tetrahedral irregular grid generation is used in undulating areas on the surface, and hexahedral regular grid generation is used in areas below the surface. The results show that the regular and irregular composite mesh generation method not only meets the requirements for inversion precision under undulating terrain conditions, but also improves the inversion calculation efficiency by nearly 6 times compared to the tetrahedral irregular mesh generation method. A numerical model is established based on actual geological conditions. The actual testing is conducted using unmanned aerial vehicles and aviation optical pump magnetometers. The numerical simulation and actual measurement results indicate that the time-shifting aeromagnetic method can accurately detect the distribution range of burning areas and the development trend of burning areas over time. It provides a basis for carrying out fire prevention and extinguishing work in coal mines.

Published
2023
Full Text
View/download PDF

31. ADORA2A-driven proline synthesis triggers epigenetic reprogramming in neuroendocrine prostate and lung cancers

Author

Jing, Na, Zhang, Kai, Chen, Xinyu, Liu, Kaiyuan, Wang, Jinming, Xiao, Lingling, Zhang, Wentian, Ma, Pengfei, Xu, Penghui, Cheng, Chaping, Wang, Deng, Zhao, Huifang, He, Yuman, Ji, Zhongzhong, Xin, Zhixiang, Sun, Yujiao, Zhang, Yingchao, Bao, Wei, Gong, Yiming, Fan, Liancheng, Ji, Yiyi, Zhuang, Guanglei, Wang, Qi, Dong, Baijun, Zhang, Pengcheng, Xue, Wei, Gao, Wei- Qiang, and Zhu, Helen He

Subjects
Prostate cancer -- Diagnosis -- Care and treatment -- Genetic aspects, Molecular mechanics -- Analysis, Lung cancer -- Diagnosis -- Care and treatment, Cancer cells -- Analysis, Androgen suppression therapy -- Patient outcomes, Proline -- Analysis, Health care industry
Abstract

Cell lineage plasticity is one of the major causes for the failure of targeted therapies in various cancers. However, the driver and actionable drug targets in promoting cancer cell lineage plasticity are scarcely identified. Here, we found that a G protein-coupled receptor, ADORA2A, is specifically upregulated during neuroendocrine differentiation, a common form of lineage plasticity in prostate cancer and lung cancer following targeted therapies. Activation of the ADORA2A signaling rewires the proline metabolism via an ERK/MYC/PYCR cascade. Increased proline synthesis promotes deacetylases SIRT6/7-mediated deacetylation of histone H3 at lysine 27 (H3K27), and thereby biases a global transcriptional output toward a neuroendocrine lineage profile. Ablation of Adora2a in genetically engineered mouse models inhibits the development and progression of neuroendocrine prostate and lung cancers, and, intriguingly, prevents the adenocarcinoma-to-neuroendocrine phenotypic transition. Importantly, pharmacological blockade of ADORA2A profoundly represses neuroendocrine prostate and lung cancer growth in vivo. Therefore, we believe that ADORA2A can be used as a promising therapeutic target to govern the epigenetic reprogramming in neuroendocrine malignancies., Introduction Lineage plasticity is often exploited by cancer cells to acquire therapeutic resistance (1). Lineage transition from adenocarcinoma (AD) to aggressive neuroendocrine (NE) derivatives is a common type of cancer [...]

Published
2023
Full Text
View/download PDF

41. Evaluation Methodology and Establishment of Indicator System for Green and Livable Sustainable Development

Author

Liu Kaiyuan, Jiao Xinmeng, and Xie Kang

Subjects
Environmental sciences, GE1-350
Abstract

At the current stage of China’s low-carbon concept and the international community’s sustainable development trend, the construction industry as one of China’s largest carbon emission industry, emission reduction is the implementation of the “dual-carbon” strategy is an inevitable requirement, but also to promote China’s economy and society as a whole to the objective needs of the green transformation. “Green and livable” has also become a higher pursuit of the industry. While our country at this stage has insufficient understanding of the evaluation system of green livability and sustainable development, this paper analyzes and evaluates the building from multiple perspectives, such as expert questionnaire, indicator establishment, weight calculation, evaluation method, and so on. The significance of this paper is to provide a new evaluation program for the inadequacy of the national green livable building.

Published
2024
Full Text
View/download PDF

47. Gremlin1 is a therapeutically targetable FGFR1 ligand that regulates lineage plasticity and castration resistance in prostate cancer

Author

Cheng, Chaping, Wang, Jinming, Xu, Penghui, Zhang, Kai, Xin, Zhixiang, Zhao, Huifang, Ji, Zhongzhong, Zhang, Man, Wang, Deng, He, Yuman, Jing, Na, Fan, Liancheng, Liu, Kaiyuan, Li, Fei, Liu, Chengcheng, Gong, Yiming, Cui, Suli, Sun, Zhe, Sun, Di, Yao, Xinlai, Li, Hongjun, Zhang, Jian, Zhang, Pengcheng, Dong, Baijun, Xue, Wei, Qian, Xueming, Gao, Wei-Qiang, and Zhu, Helen He

Published
2022
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results