806 research outputs found
Less but better: parameter-efficient fine-tuning of large language models for personality detection
No full textPersonality detection automatically identifies an individual's personality from various data sources, such as social media texts. However, as the parameter scale of language models continues to grow, the computational cost becomes increasingly difficult to manage. Fine-tuning also grows more complex, making it harder to justify the effort and reliably predict outcomes. We introduce a novel parameter-efficient fine-tuning framework, PersLLM, to address these challenges. In PersLLM, a large language model (LLM) extracts high-dimensional representations from raw data and stores them in a dynamic memory layer. PersLLM then updates the downstream layers with a replaceable output network, enabling flexible adaptation to various personality detection scenarios. By storing the features in the memory layer, we eliminate the need for repeated complex computations by the LLM. Meanwhile, the lightweight output network serves as a proxy for evaluating the overall effectiveness of the framework, improving the predictability of results. Experimental results on key benchmark datasets like Kaggle and Pandora show that PersLLM significantly reduces computational cost while maintaining competitive performance and strong adaptability
Deep learning-based real-time damage assessment of lithium-ion batteries under dynamic impact
No full textLithium-ion batteries are widely used in diverse applications due to their high energy density and long service life. However, minor mechanical impacts during operation often induce hidden internal damage, creating safety and performance risks. If such damage is not addressed promptly, it can lead to fire, explosion, or other severe consequences. This study presents a real-time lithium-ion battery (LIB) damage detection and assessment method based on a deep learning framework that integrates acoustic emission (AE) techniques, convolutional neural networks (CNN), and long short-term memory networks (LSTM). A drop-weight impact test rig is designed to simulate low-velocity mechanical impacts and to collect AE signals across multiple impact energy levels. Signal pre-processing, including Savitzky-Golay smoothing, normalization, and Gaussian noise-based data augmentation, enhances model robustness. The proposed CNN-BiLSTM model achieves an average accuracy of 95 % in classifying four damage levels. Furthermore, electrochemical performance characterization, including internal resistance and capacity decay after cycling, validates the reliability of the AE-based classification results. This approach provides a feasible, non-destructive, and intelligent solution for monitoring the health status of lithium-ion batteries under dynamic mechanical impacts, contributing to improved safety and extended service life.
GAMED: knowledge adaptive multi-experts decoupling for multimodal fake news detection
No full textMultimodal fake news detection often involves modelling heterogeneous data sources, such as vision and language. Existing detection methods typically rely on fusion effectiveness and cross-modal consistency to model the content, complicating understanding how each modality affects prediction accuracy. Additionally, these methods are primarily based on static feature modelling, making it difficult to adapt to the dynamic changes and relationships between different data modalities. This paper develops a significantly novel approach, GAMED, for multimodal modelling, which focuses on generating distinctive and discriminative features through modal decoupling to enhance cross-modal synergies, thereby optimizing overall performance in the detection process. GAMED leverages multiple parallel expert networks to refine features and pre-embed semantic knowledge to improve the experts' ability in information selection and viewpoint sharing. Subsequently, the feature distribution of each modality is adaptively adjusted based on the respective experts' opinions. GAMED also introduces a novel classification technique to dynamically manage contributions from different modalities, while improving the explainability of decisions. Experimental results on the Fakeddit and Yang datasets demonstrate that GAMED performs better than recently developed state-of-the-art models. The source code can be accessed at https://github.com/slz0925/GAMED
LL4G: self-supervised dynamic optimization for graph-based personality detection
No full textGraph-based personality detection constructs graph structures from textual data, particularly social media posts. Current methods often struggle with sparse or noisy data and rely on static graphs, limiting their ability to capture dynamic changes between nodes and relationships. This paper introduces LL4G, a self-supervised framework leveraging large language models (LLMs) to optimize graph neural networks (GNNs). LLMs extract rich semantic features to generate node representations and to infer explicit and implicit relationships. The graph structure adaptively adds nodes and edges based on input data, continuously optimizing itself. The GNN then uses these optimized representations for joint training on node reconstruction, edge prediction, and contrastive learning tasks. This integration of semantic and structural information generates robust personality profiles. Experimental results on Kaggle and Pandora datasets show LL4G outperforms state-of-the-art models
EmoPerso: enhancing personality detection with self-supervised emotion-aware modelling
No full textPersonality detection from text is commonly performed by analysing users' social media posts. However, existing methods heavily rely on large-scale annotated datasets, making it challenging to obtain high-quality personality labels. Moreover, most studies treat emotion and personality as independent variables, overlooking their interactions. In this paper, we propose a novel self-supervised framework, EmoPerso, which improves personality detection through emotion-aware modelling. EmoPerso first leverages generative mechanisms for synthetic data augmentation and rich representation learning. It then extracts pseudo-labeled emotion features and jointly optimizes them with personality prediction via multi-task learning. A cross-attention module is employed to capture fine-grained interactions between personality traits and the inferred emotional representations. To further refine relational reasoning, EmoPerso adopts a self-taught strategy to enhance the model's reasoning capabilities iteratively. Extensive experiments on two benchmark datasets demonstrate that EmoPerso surpasses state-of-the-art models. The source code is available at https://github.com/slz0925/EmoPerso.</p
Translingual adaptations: Asian works in late nineteenth- and early twentieth- century French literature
No full textThis dissertation rethinks an important genre of world literature overlooked by previous scholars, namely, creative adaptations of Eastern works by Western authors based on received translations. Specifically, my dissertation focuses on how three major nineteenth- and twentieth-century French authors—Théophile Gautier, Stéphane Mallarmé, and Victor Segalen—adapted Sanskrit, Chinese, and Tibetan works into mainstream French literature. In a broader sense, my work reads these adaptations as significant contributions to world literature. My approach consists of closely comparing the adaptations with the previous translations on which Gautier, Mallarmé, and Segalen based their retellings, and against the source works in Chinese, Sanskrit, and Tibetan. In so doing, I tease out examples of truncation, addition, paraphrase, annotation, and unintentional misreading that befell these Asian works as they traveled across time, space, and cultural spheres. The first chapter traces how the vernacular romance “Heying Lou,” by Chinese writer Li Yu, morphed into Théophile Gautier’s novella “Le Pavillon sur l’eau.” The original story, first rendered into English by John Francis Davis under the name “The Shadow in the Water,” was then further translated into French by Jean-Pierre Abel-Rémusat under the title “L’Ombre dans l’eau,” upon which Gautier later based his “Le Pavillon sur l’eau.” By teasing out each author’s input in this multilayered transmission, I reveal how Gautier at once converged with and diverged from Li Yu by intentionally misinterpreting China and unwittingly recovering some important narrative traits of the “Heying Lou” lost in previous translations. The second chapter deals with Mallarmé’s retelling of the Nalopākhyānam episode of the Sanskrit epic Mahābhārata. Mallarmé based his Nala et Damayantî on Mary Summer’s adaptation of the same name, which in turn vulgarized the French Sanskritist Émile Burnouf’s translation of the Nalopākhyānam titled Nala épisode du Mahâbhârata. Indeed, when adapting Burnouf’s literalist translation, Summer embraced an assimilative approach while injecting much Orientalist cliché. For instance, she eroticized Damayantī’s body and wishfully smoothed out what she thought to be disjointed cuts between scenes in the Sanskrit original. Her systematic recourse to abridgement not only revamped the internal structures of the story, but also weakened the character Damayantī’s image as a fully empowered Indian woman who possessed wits, volition, and rationality. Mallarmé’s poetic license, in contrast, enabled him to go beyond received norms of nineteenth-century popular narratives pertaining to the Orient. First, he singled out elements in the Nalopākhyānam that directly resonated with his own poetic agenda. Second, he relinquished Summer’s pseudorealism, mitigated many Orientalist trappings, and switched to a more symbolic treatment of plot details. Finally, Mallarmé adopted a prose style reminiscent of classical Sanskrit, owing notably to his pursuit of condensation and syntactic ellipsis, his poetics of suggestion, and his tendency to multiply nominal appositives at the expense of finite verbs. In short, although Mallarmé’s stylistic idiosyncrasies are not easy to digest, by “Sanskritizing” his phraseology, so to speak, he effectively transmuted the Nalopākhyānam into a fresh narrative consisting of evocative, highly aestheticized, and rapidly shifting images. The third chapter examines how Segalen interpolated snippets of Chinese classics into his prose poem titled Stèles, while the fourth studies Segalen’s recasting of a small portion of the Tibetan classic Padma bka’ thang in his long poem Thibet. These last two chapters counteract Segalen’s image as a progressive modernist writer invested in East-West intercultural dialogue. By delving deeply into the way Segalen reworked his primary sources both in Stèles and in Thibet, I show that (1) Segalen’s Orientalism was not always his own, but often replicated and amplified that of previous translators; (2) Segalen was not a post-Mallarméan modernist, and his literary stance was not at all revolutionary; and (3) Segalen remained a cultural imperialist at heart who was keen to forge a pedantic, self-centered, and metaphoric Orient that had no room for concrete realities. More specifically, his authorial subjectivity either negated or superimposed itself on his Chinese sources in Stèles, while in Thibet, his ego went so far as to impersonate a valiant male French explorer embarking on the journey of conquering a female Tibet. Although the abstruse intertextuality underpinning Segalen’s Orientalism has thus far spared him from criticism directed at less recondite writers such as Pierre Loti, his appropriation of Chinese ideographs and Tibetan prosody belies a continuation of the Romantic poets’ fancy for the East. Although these foreign elements may indicate Segalen’s inclusiveness for some critics, in reality they function more as marketing gimmicks for readers barely able to verify their authenticity.Ph.D.Includes bibliographical referencesby Yunfei Ba
Cromapanax lobatus Grierson 1991
No full textCromapanax lobatus Grierson (1991: 19). Type: Bhutan. Gaylegphug District, 26 54’N, 90 31 E, steep valley with thinned subtropical forest, 340 m, 21 March 1982, Grierson & Long 3897 fruiting branch only (lectotype here designated: E [E00265244, fruiting branch only]). = Macropanax undulatus (Wall. ex G. Don 1834: 394) Seemann (1864: 294). Type: India, Khasia, Wallich Cat. No. 491 6 A (lectotype designated by Shang, 1983: K [K000810350]; isolectotypes: BR [BR0000005630462, BR0000005631117], G, W).Published as part of Deng, Yunfei & Tong, Yi, 2022, The identity of Cromapanax and lectotypification of C. lobatus (Araliaceae), pp. 195-199 in Phytotaxa 567 (2) on page 199, DOI: 10.11646/phytotaxa.567.2.8, http://zenodo.org/record/714180
GAMED: Knowledge Adaptive Multi-Experts Decoupling for Multimodal Fake News Detection
No full textMultimodal fake news detection often involves modelling heterogeneous data sources, such as vision and language. Existing detection methods typically rely on fusion effectiveness and cross-modal consistency to model the content, complicating understanding how each modality affects prediction accuracy. Additionally, these methods are primarily based on static feature modelling, making it difficult to adapt to the dynamic changes and relationships between different data modalities. This paper develops a significantly novel approach, GAMED, for multimodal modelling, which focuses on generating distinctive and discriminative features through modal decoupling to enhance cross-modal synergies, thereby optimizing overall performance in the detection process. GAMED leverages multiple parallel expert networks to refine features and pre-embed semantic knowledge to improve the experts’ ability in information selection and viewpoint sharing. Subsequently, the feature distribution of each modality is adaptively adjusted based on the respective experts’ opinions. GAMED also introduces a novel classification technique to dynamically manage contributions from different modalities, while improving the explainability of decisions. Experimental results on the Fakeddit and Yang datasets demonstrate that GAMED performs better than recently developed state-of-the-art models
Acoustical Observation with Multiple Wave Gliders for Internet of Underwater Things
No full textOne of the challenges of Internet of Underwater Things (IoUT) is the design of nodes for collecting information from underwater, with features of low cost, long-term, long range, voyage routing, and real-time communications. Wave glider, has shown great potential acting as IoUT nodes through its persistent, long range travelling, and flexibility underwater. In this paper, we propose an architecture of Internet of Underwater Things, involving multiple wave gliders as nodes for acoustical observation. We present target localisation method via acoustical observation of nodes with multiple wave gliders deployed underwater, by which precision of bearing estimation of each node is required to achieve high precision of localisation. With the data collected, we apply a compensation method of bearing estimate when the hydrophone array is rotating during the observation. The feasibility of acoustical observation of wave gliders has been validated through both simulation and sea trial data, which is of great potential to be nodes for constructing IoUT
Reliability analysis and anomaly detection considering long-range dependence effects
No full textThesis (Ph.D.)-- Wichita State University, College of Engineering, Dept. of of Industrial, Systems and Manufacturing EngineeringReliability analysis and anomaly detection are crucial to the Prognostics and Health
Management (PHM) of modern complex systems. Recently, with the advancement of
measurement technology, a Long-Range Dependence/Long-Term Memory (LRD/LTM) effect has
recently been detected in reliability and quality monitoring fields. The LRD effect is a type of non-
Markovian property and refers to the high dependence between two measurements across a longtime
interval or a long-distance range. In mathematics, the LRD effect indicates that the
autocorrelation of the metrics is non-summable. In reliability and anomaly detection fields, most
studies have been conducted ignoring the LRD effect, which could incur some serious issues, such
as biased lifetime prediction or inaccurate anomaly detection. To overcome these challenges, we
propose novel reliability analysis and anomaly detection approaches to integrate the LRD effect.
Specifically, in Chapter 2 we propose a reliability analysis considering the LRD effect and random
errors simultaneously under normal operating conditions. In Chapter 3, we develop a reliability
analysis integrating the LRD effect under accelerated conditions. In Chapter 4, we propose a
quality control analysis on surface anomaly detection considering LRD. In Chapter 5, we develop
an LRD-integrated anomaly detection using 3D composite structure information. Results show
that the proposed LRD-integrated approaches, by considering the LRD effect, significantly
outperforms the conventional models that ignore the LRD effect
- …
