Search

Your search keyword '"Liu, Chaowei"' showing total 34 results
Start Over Author "Liu, Chaowei" Search Limiters Full Text
34 results on '"Liu, Chaowei"'

1. VRPTEST: Evaluating Visual Referring Prompting in Large Multimodal Models

Author

Li, Zongjie, Wang, Chaozheng, Liu, Chaowei, Ma, Pingchuan, Wu, Daoyuan, Wang, Shuai, and Gao, Cuiyun

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

With recent advancements in Large Multimodal Models (LMMs) across various domains, a novel prompting method called visual referring prompting has emerged, showing significant potential in enhancing human-computer interaction within multimodal systems. This method offers a more natural and flexible approach to human interaction with these systems compared to traditional text descriptions or coordinates. However, the categorization of visual referring prompting remains undefined, and its impact on the performance of LMMs has yet to be formally examined. In this study, we conduct the first comprehensive analysis of LMMs using a variety of visual referring prompting strategies. We introduce a benchmark dataset called VRPTEST, comprising 3 different visual tasks and 2,275 images, spanning diverse combinations of prompt strategies. Using VRPTEST, we conduct a comprehensive evaluation of eight versions of prominent open-source and proprietary foundation models, including two early versions of GPT-4V. We develop an automated assessment framework based on software metamorphic testing techniques to evaluate the accuracy of LMMs without the need for human intervention or manual labeling. We find that the current proprietary models generally outperform the open-source ones, showing an average accuracy improvement of 22.70%; however, there is still potential for improvement. Moreover, our quantitative analysis shows that the choice of prompt strategy significantly affects the accuracy of LMMs, with variations ranging from -17.5% to +7.3%. Further case studies indicate that an appropriate visual referring prompting strategy can improve LMMs' understanding of context and location information, while an unsuitable one might lead to answer rejection. We also provide insights on minimizing the negative impact of visual referring prompting on LMMs., Comment: 13 pages

Published
2023

3. DocStormer: Revitalizing Multi-Degraded Colored Document Images to Pristine PDF

Author

Liu, Chaowei, Li, Jichun, Teng, Yihua, Wang, Chaoqun, Xu, Nuo, Wu, Jihao, and Tu, Dandan

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

For capturing colored document images, e.g. posters and magazines, it is common that multiple degradations such as shadows, wrinkles, etc., are simultaneously introduced due to external factors. Restoring multi-degraded colored document images is a great challenge, yet overlooked, as most existing algorithms focus on enhancing color-ignored document images via binarization. Thus, we propose DocStormer, a novel algorithm designed to restore multi-degraded colored documents to their potential pristine PDF. The contributions are: firstly, we propose a "Perceive-then-Restore" paradigm with a reinforced transformer block, which more effectively encodes and utilizes the distribution of degradations. Secondly, we are the first to utilize GAN and pristine PDF magazine images to narrow the distribution gap between the enhanced results and PDF images, in pursuit of less degradation and better visual quality. Thirdly, we propose a non-parametric strategy, PFILI, which enables a smaller training scale and larger testing resolutions with acceptable detail trade-off, while saving memory and inference time. Fourthly, we are the first to propose a novel Multi-Degraded Colored Document image Enhancing dataset, named MD-CDE, for both training and evaluation. Experimental results show that the DocStormer exhibits superior performance, capable of revitalizing multi-degraded colored documents into their potential pristine digital versions, which fills the current academic gap from the perspective of method, data, and task.

Published
2023

4. On Extracting Specialized Code Abilities from Large Language Models: A Feasibility Study

Author

Li, Zongjie, Wang, Chaozheng, Ma, Pingchuan, Liu, Chaowei, Wang, Shuai, Wu, Daoyuan, Gao, Cuiyun, and Liu, Yang

Subjects
Computer Science - Software Engineering
Abstract

Recent advances in large language models (LLMs) significantly boost their usage in software engineering. However, training a well-performing LLM demands a substantial workforce for data collection and annotation. Moreover, training datasets may be proprietary or partially open, and the process often requires a costly GPU cluster. The intellectual property value of commercial LLMs makes them attractive targets for imitation attacks, but creating an imitation model with comparable parameters still incurs high costs. This motivates us to explore a practical and novel direction: slicing commercial black-box LLMs using medium-sized backbone models. In this paper, we explore the feasibility of launching imitation attacks on LLMs to extract their specialized code abilities, such as"code synthesis" and "code translation." We systematically investigate the effectiveness of launching code ability extraction attacks under different code-related tasks with multiple query schemes, including zero-shot, in-context, and Chain-of-Thought. We also design response checks to refine the outputs, leading to an effective imitation training process. Our results show promising outcomes, demonstrating that with a reasonable number of queries, attackers can train a medium-sized backbone model to replicate specialized code behaviors similar to the target LLMs. We summarize our findings and insights to help researchers better understand the threats posed by imitation attacks, including revealing a practical attack surface for generating adversarial code examples against LLMs., Comment: 13 pages

Published
2023

13. Identification and validation of autophagy-related genes during osteogenic differentiation of bone marrow mesenchymal stem cells

Author

Li Yan, Yao Xiu, Lin Yanjun, Xing Yifeng, Liu Chaowei, Xu Jianghan, and Wu Dong

Subjects
autophagy, bioinformatics, bone marrow mesenchymal stem cells, bone regeneration, osteogenesis, Medicine
Abstract

Objective(s): Osteogenic differentiation of bone marrow mesenchymal stem cells (BMSCs) is an essential stage in bone formation. Autophagy plays a pivotal role in the self-renewal potential and pluripotency of stem cells. This study aimed to explore the function of autophagy-related genes during osteogenic differentiation of BMSCs.Materials and Methods: The differentially expressed autophagy-related genes (ARGs) were obtained from the GEO and HADb databases. The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using R software. The PPI and hub gene mining networks were constructed using the STRING database and Cytoscape. Finally, the RT-qPCR was conducted to validate the expression level of ARGs in BMSCs.Results: Thirty-seven differentially expressed ARGs were finally obtained, including 12 upregulated and 25 downregulated genes. GO and KEGG enrichment analysis showed that most of these genes were enriched in apoptosis and autophagy. The PPI network revealed strong interactions between differentially expressed ARGs. The expression level of differentially expressed ARGs tested by RT-qPCR showed 6 upregulated ARGs, including FOXO1, MAP1LC3C, CTSB, FOXO3, CALCOCO2, FKBP1A, and 4 downregulated ARGs, including MAPK8IP1, NRG1, VEGFA, and ITGA6 were consistent with the expression of high-throughput sequencing data.Conclusion: We identified 37 ARGs during osteogenic differentiation using bioinformatics analysis. FOXO1, MAP1LC3C, CTSB, FOXO3, CALCOCO2, FKBP1A, MAPK8IP1, NRG1, VEGFA, and ITGA6 may regulate osteogenic differentiation of hBMSCs by involving autophagy pathway. This study provides new insight into the osteogenic differentiation of hBMSCs and may be available in developing therapeutic strategies for maxillofacial bone defects.

Published
2022
Full Text
View/download PDF

17. The Accuracy Differences of Using Unmanned Aerial Vehicle Images Monitoring Maize Plant Height at Different Growth Stages

Author

YANG Jin, MING Bo, YANG Fei, XU Honggen, LI Lulu, GAO Shang, LIU Chaowei, WANG Keru, and LI Shaokun

Subjects
accuracy difference, uav, monitoring plant height, maize, multispectral, Agriculture (General), S1-972, Technology (General), T1-995
Abstract

The digital elevation model (DEM) of maize population in field was constructed by using optical imaging equipment mounted on unmanned aerial vehicle (UAV) to study the accuracy difference of maize population height monitoring at different growth stages. Three cultivars and eight sowing date treatments were set up to structure maize population with different plant heights. A multi-rotor UAV with high-definition digital camera and multispectral imaging sensor was used to take RGB images and multispectral images in the experiment area on July 25th and August 27th, 2018, which were the biggest and smallest differences in plant height. The DEM data of maize population and canopy height were obtained with image pose correction, image mosaic, point cloud generation, and space reconstruction, et al. The canopy height and plant height were normalized, and the correlation between different cultivars and sowing date was analyzed based on UAV and manual plant height measurement. The feasibility of using DEM data of maize canopy to monitor the difference of plant height was clarified. The results showed that the height difference of maize population could be reflected by the digital elevation information obtained from high-definition RGB camera and multispectral camera. The plant height monitoring accuracy of HD RGB camera was higher than that of multispectral camera. However, the monitoring accuracy of plant height was not enough under the ready-made image equipment and treatment method. So, it was difficult to reflect the smaller plant height difference of maize population. Growth stage had a great influence on the monitoring of maize plant height. When the canopy of early growth stage has not completely covered the surface or the leaf yellow and withered in the later stage of growth. The plant height of the population affected by the exposed surface was seriously underestimated. In this study, the effects of UAV imaging equipment on monitoring maize plant height were analyzed. The influence factors can be used as reference for the application of the method in field production.

Published
2021
Full Text
View/download PDF

18. Construction of Microbial Consortium to Enhance Cellulose Degradation in Corn Straw during Composting.

Author

Li, Jie, Li, Juan, Yang, Ruopeng, Yang, Ping, Fu, Hongbo, Yang, Yongchao, and Liu, Chaowei

Subjects
CORN straw, CONGO red (Staining dye), BACILLUS subtilis, FILTER paper, GERMINATION, WHEAT straw
Abstract

The improper treatment of crop straw not only leads to resource wastage but also adversely impacts the ecological environment. However, the application of microorganisms can accelerate the decomposition of crop straw and improve its utilization. In this study, cellulose-degrading microbial strains were isolated from naturally decayed corn straw and screened using Congo red staining, along with assessing variations in carboxymethyl cellulase (CMCase) activity, filter paper enzyme (FPase) activity and β-glucosidase (β-Gase) activity, as well as the degradation rate. The eight strains, namely Neurospora intermedia isolate 29 (A1), Streptomyces isolate FFJC33 (A2), Gibberella moniliformis isolate FKCB-009 (A3), Fusarium fujikuroi isolate EFS3(2) (A4), Fusarium Fujikuroi isolate FZ04 (A5), Lysine bacillus macroides strain LNHL43 (B1), Bacillus subtilis strain MPF30 (B2) and Paenibacilli lautus strain ALEB-P1 (C), were identified and selected for microbial strain consortium design based on their high activities of CMCase, FPase and β-Gase. The fungi, bacteria and actinomycete strains were combined without antagonistic effects on corn straw decomposition. The results showed the A2B2 combination had a significantly higher FPase at 55.44 U/mL and β-Gase at 25.73 U/mL than the other two strain combinations (p < 0.05). Additionally, the degradation rate of this combination was 40.33%, which was considerably higher than that of the other strains/consortia. The strain combination A4B2C also had superior enzyme activity, including CMCase with a value of 35.03 U/mL, FPase with a value of 63.59 U/mL and β-Gase with a value of 26.15 U/mL, which were significantly different to those of the other three strain combinations (p < 0.05). Furthermore, seven single microbial strains with high cellulase activities were selected to construct various microbial consortiums for in situ composting in order to evaluate their potential. Taken as a whole, the results of composting, including temperature, moisture content, pH, E4/E6 value and seed germination index, indicated that the microbial strain consortium consisting of Neurospora intermediate isolate 29, Fusarium fujikuroi isolate EFS3(2), Fusarium fujikuroi isolate FZ04, Lysinibacillus macrolides, Lysinibacillus sphaericus, Bacillus subtilis and Paenibacillus lautus was advantageous for corn straw decomposition and yielded high-quality compost. The screened flora was able to effectively degrade corn straw. This study provides a novel solution for the construction of a microbial consortium for the composting of corn straw. [ABSTRACT FROM AUTHOR]

Published
2024
Full Text
View/download PDF

19. On Extracting Specialized Code Abilities from Large Language Models: A Feasibility Study

Author

Li, Zongjie, Wang, Chaozheng, Ma, Pingchuan, Liu, Chaowei, Wang, Shuai, Wu, Daoyuan, Gao, Cuiyun, Liu, Yang, Li, Zongjie, Wang, Chaozheng, Ma, Pingchuan, Liu, Chaowei, Wang, Shuai, Wu, Daoyuan, Gao, Cuiyun, and Liu, Yang

Abstract

Recent advances in large language models (LLMs) significantly boost their usage in software engineering. However, training a well-performing LLM demands a substantial workforce for data collection and annotation. Moreover, training datasets may be proprietary or partially open, and the process often requires a costly GPU cluster. The intellectual property value of commercial LLMs makes them attractive targets for imitation attacks, but creating an imitation model with comparable parameters still incurs high costs. This motivates us to explore a practical and novel direction: slicing commercial black-box LLMs using medium-sized backbone models. In this paper, we explore the feasibility of launching imitation attacks on LLMs to extract their specialized code abilities, such as

Published
2024

22. Screening of Germplasm and Construction of Evaluation System for Autotoxicity Tolerance during Seed Germination in Cucumber.

Author

Li, Jie, Li, Jian, Yang, Ping, Fu, Hongbo, Yang, Yongchao, and Liu, Chaowei

Subjects
CUCUMBERS, GERMPLASM, CINNAMIC acid, PRINCIPAL components analysis, CLUSTER analysis (Statistics), CULTIVARS, GERMINATION
Abstract

Due to the widespread use of intensive cropping patterns, the problem of continuous cropping obstacle, which is dominated by autotoxicity, has been becoming more and more prominent. Although many methods have been proposed to overcome the continuous cropping obstacle of cucumber, no study has reported the screening and evaluation of cucumber germplasm resistant to autotoxicity. In this study, 28 physiological indices related to the cucumber bud stage under cinnamic acid (CA) treatment were determined. In total, 45 cucumber cultivars were classified into three groups using principal component analysis and cluster analysis, and a model for evaluating cucumber resistance to autotoxicity was developed. The evaluation model was validated using autotoxicity-tolerant and non-autotoxicity-tolerant cultivars. The results showed that the growth of non-autotoxicity-tolerant cultivars was significantly inhibited compared to autotoxicity-tolerant cultivars. This indicated that the evaluation model of cucumber autotoxicity tolerance is reliable. The results of this study provide a valuable reference for the application of cucumber autotoxicity-tolerant germplasm resources and the development of autotoxicity-tolerant genes. [ABSTRACT FROM AUTHOR]

Published
2024
Full Text
View/download PDF

25. On the Feasibility of Specialized Ability Extracting for Large Language Code Models

Author

Li, Zongjie, Wang, Chaozheng, Ma, Pingchuan, Liu, Chaowei, Wang, Shuai, Wu, Daoyuan, and Gao, Cuiyun

Subjects
Software Engineering (cs.SE), FOS: Computer and information sciences, Computer Science - Software Engineering
Abstract

Recent progress in large language code models (LLCMs) has led to a dramatic surge in the use of software development. Nevertheless, it is widely known that training a well-performed LLCM requires a plethora of workforce for collecting the data and high quality annotation. Additionally, the training dataset may be proprietary (or partially open source to the public), and the training process is often conducted on a large-scale cluster of GPUs with high costs. Inspired by the recent success of imitation attacks in extracting computer vision and natural language models, this work launches the first imitation attack on LLCMs: by querying a target LLCM with carefully-designed queries and collecting the outputs, the adversary can train an imitation model that manifests close behavior with the target LLCM. We systematically investigate the effectiveness of launching imitation attacks under different query schemes and different LLCM tasks. We also design novel methods to polish the LLCM outputs, resulting in an effective imitation training process. We summarize our findings and provide lessons harvested in this study that can help better depict the attack surface of LLCMs. Our research contributes to the growing body of knowledge on imitation attacks and defenses in deep neural models, particularly in the domain of code related tasks., Comment: 11 pages

Published
2023
Full Text
View/download PDF

29. Comparison of Two Lab Simulation Methods of Multiple Heavy Metal Contamination on FCC Catalysts.

Author

Yang, Yong, Zu, Zixuan, Ma, Xueli, Liu, Chaowei, Su, Yi, Li, Hongwei, and Ji, Dong

Subjects
CATALYSTS, CATALYTIC cracking, PROBLEM solving, HEAVY metals, METALS, METAL catalysts
Abstract

Qualitative and quantitative description are key to solving the problem of heavy metal contamination on fluid catalytic cracking (FCC) catalysts. The loading efficiencies for different metals were compared for the two lab simulation methods of Multi-Cyclic Deactivation (MCD) and Advanced Catalyst Evaluation (ACE), and the microcatalytic performance of metal-contaminated catalysts was evaluated using an ACE Model C device. The results show that the MCD and ACE methods both obtain extremely high data accuracy, indicating that they can be used to ensure the parallel reliability of experimental results. The typical operating parameters for hydrothermal aging and metals loading can be adjusted to suit different metal types and content targets for either of these two simulation methods. Compared with an equilibrium catalyst from an industrial unit, the MCD method has the advantages of basic hydrothermal aging treatment with less metal loading efficiency, while the ACE method has an accurate metal amount and high loading efficiency for metal contamination, with a metal balance recovery rate above 99.5% at similar activation to the equilibrium catalyst. When used with a reasonable and effective metal pretreatment scheme, these two laboratory simulation methods can be used to evaluate new commercial catalysts and in fundamental experiments for the improvement of FCC catalysts for removal of metal contamination. [ABSTRACT FROM AUTHOR]

Published
2023
Full Text
View/download PDF

31. The Hydrocarbon Potential of Carboniferous Reservoirs in the Jimsar Sag, Northwest China: Implications for a Giant Volcanic-Petroleum Reserves

Author

Gong, Deyu, Song, Yong, Peng, Miao, Liu, Chaowei, Wang, Ruiju, and Wu, Wei’an

Subjects
General Earth and Planetary Sciences
Abstract

As an unconventional petroleum reservoir, the volcanic reservoir is essential for finding large-scale oil and gas reserves in deep sedimentary basins. Based on basin modeling, organic geochemistry, and organic petrology, this study evaluates the exploration potential of the Carboniferous volcanic-petroleum system in the Jimsar Sag of the southeastern Junggar Basin. The Carboniferous source rocks in the study area were developed in the Lower Carboniferous Songkharsu Formation, a set of marine–terrigenous transitional source rocks, lithologically composed of carbonaceous mudstone, mudstone, and coal. The kerogen is characterized by type II2–III, indicating a gas-prone source rock. Carbonaceous mudstones, mudstone, and coal are classified as medium–good, medium, and poor source rocks, respectively. The hydrocarbon-generating potential is close to that of the Kelameili gas field. The Carboniferous-reservoir oil in well J15 is characterized by 13C-enriched stable carbon isotopes, relatively high ƩC31–35/C30 and Pr/Ph ratios, relatively low density, gammacerane index, regular C29-sterane content, and undetected β-carotene, indicating a Songkharsu origin. The Carboniferous-reservoired gas from well J3301 has stable carbon isotope composition enriched in 13C, similar to the gas from the Kelameili gas field, and should be from the Songkharsu source rock. This set of source rocks is widely distributed in the Jimsar Sag with considerable thickness, among which thickness >100 m reaches 580 km2, accounting for 38.7% of the whole sag. The Songkharsu source rocks entered the main oil-generating window at the end of the Early Jurassic. The area of Songkharsu source rocks in the Jimsar Sag with gas-generating intensity greater than 20 × 108 m3/km2 reaches 1,015 km2, whereas that with oil-generating intensity greater than 500 × 104 t/km2 reaches 1,146 km2. However, 13 Carboniferous volcanic lithological traps were found in the sag, covering an area of 230 km2. The Carboniferous volcanic-petroleum system in the Jimsar Sag has the resource potential to form large and medium oil and gas fields, which could become a critical replacement field for volcanic oil and gas exploration in the Junggar Basin after the Klameili gas field.

Published
2022
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results