arXiv subjects

Haoyu Gao

Publications and source records attributed to Haoyu Gao.

30 records · Page 2Linked to original sources

arXiv2025

Design and Experimental Test of Datatic Approximate Optimal Filter in Nonlinear Dynamic Systems

Filtering is crucial in engineering fields, providing vital state estimation for control systems. However, the nonlinear nature of complex systems and the presence of non-Gaussian noises pose significant challenges to the performance of conventional filtering methods in terms of estimation accuracy and computational efficiency. In this work, we present a data-driven closed-loop filter, termed datatic approximate optimal filter (DAOF), specifically designed for nonlinear systems under non-Gaussian conditions. We first formulate a Markovian filtering problem (MFP), which inherently shares a connection with reinforcement learning (RL) as it aims to compute the optimal state estimate by minimizing the accumulated error. To solve MFP, we propose DAOF, which primarily incorporates a trained RL policy and features two distinct structural designs: DAOF-v1 and DAOF-v2. Designed for systems with explicit models, DAOF-v1 combines prediction and update phases, with the RL policy generating the update value. Meanwhile, DAOF-v2 bypasses system modeling by directly outputting the state estimate. Then, we utilize an actor-critic algorithm to learn the parameterized policy for DAOF. Experimental results on a 2-degree-of-freedom (2-DOF) vehicle system, equipped with explicit system models, demonstrate the superior accuracy and computational efficiency of DAOF-v1 compared to existing nonlinear filters. Moreover, DAOF-v2 showcases its unique ability to perform filtering without requiring explicit system modeling, as validated by a 14-DOF vehicle system.

eess.SY↗

arXiv2025

Adapting Installation Instructions in Rapidly Evolving Software Ecosystems

README files play an important role in providing installation-related instructions to software users and are widely used in open source software systems on platforms such as GitHub. However, these files often suffer from various documentation issues, leading to challenges in comprehension and potential errors in content. Despite their significance, there is a lack of systematic understanding regarding the documentation efforts invested in README files, especially in the context of installation-related instructions, which are crucial for users to start with a software project. To fill the research gap, we conducted a qualitative study, investigating 400 GitHub repositories with 1,163 README commits that focused on updates in installation-related sections. Our research revealed six major categories of changes in the README commits, namely pre-installation instructions, installation instructions, post-installation instructions, help information updates, document presentation, and external resource management. We further provide detailed insights into modification behaviours and offer examples of these updates. Based on our findings, we propose a README template tailored to cover the installation-related sections for documentation maintainers to reference when updating documents. We further validate this template by conducting an online survey, identifying that documentation readers find the augmented documents based on our template are generally of better quality. We further provide recommendations to practitioners for maintaining their README files, as well as motivations for future research directions... (too long for arxiv)

cs.SE↗

arXiv2024

Antelope: Potent and Concealed Jailbreak Attack Strategy

Due to the remarkable generative potential of diffusion-based models, numerous researches have investigated jailbreak attacks targeting these frameworks. A particularly concerning threat within image models is the generation of Not-Safe-for-Work (NSFW) content. Despite the implementation of security filters, numerous efforts continue to explore ways to circumvent these safeguards. Current attack methodologies primarily encompass adversarial prompt engineering or concept obfuscation, yet they frequently suffer from slow search efficiency, conspicuous attack characteristics and poor alignment with targets. To overcome these challenges, we propose Antelope, a more robust and covert jailbreak attack strategy designed to expose security vulnerabilities inherent in generative models. Specifically, Antelope leverages the confusion of sensitive concepts with similar ones, facilitates searches in the semantically adjacent space of these related concepts and aligns them with the target imagery, thereby generating sensitive images that are consistent with the target and capable of evading detection. Besides, we successfully exploit the transferability of model-based attacks to penetrate online black-box services. Experimental evaluations demonstrate that Antelope outperforms existing baselines across multiple defensive mechanisms, underscoring its efficacy and versatility.

cs.CR↗

arXiv2024

An Explicit Discrete-Time Dynamic Vehicle Model with Assured Numerical Stability

Numerical stability is of great significance for discrete-time dynamic vehicle model. Among the unstable factors, low-speed singularity stands out as one of the most challenging issues, which arises from that the denominator of tire side angle term only contains the vehicle longitudinal speed. Consequently, for the common low-speed and stop-start driving scenarios, the calculated tire slip angle will approach infinity, which will further lead to the numerical explosion of other vehicle states. In response to this critical challenge, we propose a discrete-time dynamic vehicle model that effectively mitigates the low-speed singularity issue, ensuring numerical stability and maintaining the explicit form-highly favored by model-based control algorithms. To validate the numerical stability of our model, we conduct a rigorous theoretical analysis, establishing sufficient conditions for stability, and conduct extensive empirical validation tests across a wide spectrum of speeds. Subsequent to the validation process, we conduct comprehensive simulations comparing our proposed model with both kinematic models and existing dynamic models discretized through the forward Euler method. The results demonstrate that our proposed model shows better comprehensive performance in terms of both the accuracy and numerical stability. Finally, the real vehicle experiments are carried out to support that our proposed model can closely aligns to the real vehicle trajectories showcasing its practicality and ease of use. Notably, our work stands as the pioneering endeavor in introducing an explicit discrete-time dynamic vehicle model suitable for common urban driving scenarios including low-speed and stop-start.

eess.SY↗

arXiv2024

Documenting Ethical Considerations in Open Source AI Models

Background: The development of AI-enabled software heavily depends on AI model documentation, such as model cards, due to different domain expertise between software engineers and model developers. From an ethical standpoint, AI model documentation conveys critical information on ethical considerations along with mitigation strategies for downstream developers to ensure the delivery of ethically compliant software. However, knowledge on such documentation practice remains scarce. Aims: The objective of our study is to investigate how developers document ethical aspects of open source AI models in practice, aiming at providing recommendations for future documentation endeavours. Method: We selected three sources of documentation on GitHub and Hugging Face, and developed a keyword set to identify ethics-related documents systematically. After filtering an initial set of 2,347 documents, we identified 265 relevant ones and performed thematic analysis to derive the themes of ethical considerations. Results: Six themes emerge, with the three largest ones being model behavioural risks, model use cases, and model risk mitigation. Conclusions: Our findings reveal that open source AI model documentation focuses on articulating ethical problem statements and use case restrictions. We further provide suggestions to various stakeholders for improving documentation practice regarding ethical considerations.

cs.SE↗

arXiv2023

Self-Explanation Prompting Improves Dialogue Understanding in Large Language Models

Task-oriented dialogue (TOD) systems facilitate users in executing various activities via multi-turn dialogues, but Large Language Models (LLMs) often struggle to comprehend these intricate contexts. In this study, we propose a novel "Self-Explanation" prompting strategy to enhance the comprehension abilities of LLMs in multi-turn dialogues. This task-agnostic approach requires the model to analyze each dialogue utterance before task execution, thereby improving performance across various dialogue-centric tasks. Experimental results from six benchmark datasets confirm that our method consistently outperforms other zero-shot prompts and matches or exceeds the efficacy of few-shot prompts, demonstrating its potential as a powerful tool in enhancing LLMs' comprehension in complex dialogue tasks.

cs.CL↗

arXiv2023

Evaluating Transfer Learning for Simplifying GitHub READMEs

Software documentation captures detailed knowledge about a software product, e.g., code, technologies, and design. It plays an important role in the coordination of development teams and in conveying ideas to various stakeholders. However, software documentation can be hard to comprehend if it is written with jargon and complicated sentence structure. In this study, we explored the potential of text simplification techniques in the domain of software engineering to automatically simplify GitHub README files. We collected software-related pairs of GitHub README files consisting of 14,588 entries, aligned difficult sentences with their simplified counterparts, and trained a Transformer-based model to automatically simplify difficult versions. To mitigate the sparse and noisy nature of the software-related simplification dataset, we applied general text simplification knowledge to this field. Since many general-domain difficult-to-simple Wikipedia document pairs are already publicly available, we explored the potential of transfer learning by first training the model on the Wikipedia data and then fine-tuning it on the README data. Using automated BLEU scores and human evaluation, we compared the performance of different transfer learning schemes and the baseline models without transfer learning. The transfer learning model using the best checkpoint trained on a general topic corpus achieved the best performance of 34.68 BLEU score and statistically significantly higher human annotation scores compared to the rest of the schemes and baselines. We conclude that using transfer learning is a promising direction to circumvent the lack of data and drift style problem in software README files simplification and achieved a better trade-off between simplification and preservation of meaning.

cs.SE↗

arXiv2023

Speech-Text Dialog Pre-training for Spoken Dialog Understanding with Explicit Cross-Modal Alignment

Recently, speech-text pre-training methods have shown remarkable success in many speech and natural language processing tasks. However, most previous pre-trained models are usually tailored for one or two specific tasks, but fail to conquer a wide range of speech-text tasks. In addition, existing speech-text pre-training methods fail to explore the contextual information within a dialogue to enrich utterance representations. In this paper, we propose Speech-text dialog Pre-training for spoken dialog understanding with ExpliCiT cRoss-Modal Alignment (SPECTRA), which is the first-ever speech-text dialog pre-training model. Concretely, to consider the temporality of speech modality, we design a novel temporal position prediction task to capture the speech-text alignment. This pre-training task aims to predict the start and end time of each textual word in the corresponding speech waveform. In addition, to learn the characteristics of spoken dialogs, we generalize a response selection task from textual dialog pre-training to speech-text dialog pre-training scenarios. Experimental results on four different downstream speech-text tasks demonstrate the superiority of SPECTRA in learning speech-text alignment and multi-turn dialog context.

cs.CL↗

arXiv2023

Unsupervised Dialogue Topic Segmentation with Topic-aware Utterance Representation

Dialogue Topic Segmentation (DTS) plays an essential role in a variety of dialogue modeling tasks. Previous DTS methods either focus on semantic similarity or dialogue coherence to assess topic similarity for unsupervised dialogue segmentation. However, the topic similarity cannot be fully identified via semantic similarity or dialogue coherence. In addition, the unlabeled dialogue data, which contains useful clues of utterance relationships, remains underexploited. In this paper, we propose a novel unsupervised DTS framework, which learns topic-aware utterance representations from unlabeled dialogue data through neighboring utterance matching and pseudo-segmentation. Extensive experiments on two benchmark datasets (i.e., DialSeg711 and Doc2Dial) demonstrate that our method significantly outperforms the strong baseline methods. For reproducibility, we provide our code and data at:https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/dial-start.

cs.CL↗

arXiv2013

Analyzing user behavior of the micro-blogging website Sinaweibo during hot social events

The spread and resonance of users' opinions on SinaWeibo, the most popular micro-blogging website in China, are tremendously influential, having significantly affected the processes of many real-world hot social events. We select 21 hot events that were widely discussed on SinaWeibo in 2011, and do some statistical analyses. Our main findings are that (i) male users are more likely to be involved, (ii) messages that contain pictures and those posted by verified users are more likely to be reposted, while those with URLs are less likely, (iii) gender factor, for most events, presents no significant difference in reposting likelihood.

cs.SI↗

arXiv2013

What are Chinese Talking about in Hot Weibos?

SinaWeibo is a Twitter-like social network service emerging in China in recent years. People can post weibos (microblogs) and communicate with others on it. Based on a dataset of 650 million weibos from August 2009 to January 2012 crawled from APIs of SinaWeibo, we study the hot ones that have been reposted for at least 1000 times. We find that hot weibos can be roughly classified into eight categories, i.e. Entertainment & Fashion, Hot Social Events, Leisure & Mood, Life & Health, Seeking for Help, Sales Promotion, Fengshui & Fortune and Deleted Weibos. In particular, Leisure & Mood and Hot Social Events account for almost 65% of all the hot weibos. This reflects very well the fundamental dual-structure of the current society of China: On the one hand, economy has made a great progress and quite a part of people are now living a relatively prosperous and fairly easy life. On the other hand, there still exist quite a lot of serious social problems, such as government corruptions and environmental pollutions. It is also shown that users' posting and reposting behaviors are greatly affected by their identity factors (gender, verification status, and regional location). For instance, (1) Two thirds of the hot weibos are created by male users. (2) Although verified users account for only 0.1% in SinaWeibo, 46.5% of the hot weibos are contributed by them. Very interestingly, 39.2% are written by SPA users. A more or less pathetic fact is that only 14.4% of the hot weibos are created by grassroots (individual users that are neither SPA nor verified). (3) Users from different areas of China have distinct posting and reposting behaviors which usually reflect very their local cultures. Homophily is also examined for people's reposting behaviors.

cs.SI↗

arXiv2012

Fashion, Cooperation, and Social Interactions

Fashion plays such a crucial rule in the evolution of culture and society that it is regarded as a second nature to the human being. Also, its impact on economy is quite nontrivial. On what is fashionable, interestingly, there are two viewpoints that are both extremely widespread but almost opposite: conformists think that what is popular is fashionable, while rebels believe that being different is the essence. Fashion color is fashionable in the first sense, and Lady Gaga in the second. We investigate a model where the population consists of the afore-mentioned two groups of people that are located on social networks (a spatial cellular automata network and small-world networks). This model captures two fundamental kinds of social interactions (coordination and anti-coordination) simultaneously, and also has its own interest to game theory: it is a hybrid model of pure competition and pure cooperation. This is true because when a conformist meets a rebel, they play the zero sum matching pennies game, which is pure competition. When two conformists (rebels) meet, they play the (anti-) coordination game, which is pure cooperation. Simulation shows that simple social interactions greatly promote cooperation: in most cases people can reach an extraordinarily high level of cooperation, through a selfish, myopic, naive, and local interacting dynamic (the best response dynamic). We find that degree of synchronization also plays a critical role, but mostly on the negative side. Four indices, namely cooperation degree, average satisfaction degree, equilibrium ratio and complete ratio, are defined and applied to measure people's cooperation levels from various angles. Phase transition, as well as emergence of many interesting geographic patterns in the cellular automata network, is also observed.

cs.MA↗