Hong Kong University of Science and Technology
Hong Kong University of Science and Technology Institutional RepositoryNot a member yet
162821 research outputs found
Sort by
Cyclohexyl acetate functions like a volatile sex pheromone mimic in Caenorhabditis nematodes
No full textBackground: Nematodes communicate via diverse sex pheromones, including long-range volatile signals, short-range chemical cues, and contact-dependent molecules. While the ascaroside family of small molecules that mediate short-range attraction is well characterized, the identities and roles of volatile sex pheromones (VSPs) that act over longer ranges remain unknown. Results: Using GC–MS analysis of crude VSP extracts, we identified cyclohexyl acetate (CA) as a candidate mimic, sharing retention time and mass spectral features with natural VSPs. Behavioral assays demonstrated that CA acts as a concentration-dependent, male-specific attractant in Caenorhabditis. Pre-exposure to VSPs induced cross-adaptation to CA, suggesting shared sensory processing. Surprisingly, genetic and calcium imaging analyses revealed that CA perception is mediated primarily by AWCon (str-2-expressing) neurons and involves VSP chemoreceptor srd-1-independent pathways, which are distinct from the neural pathways involved in natural VSP perception. Conclusions: These data indicate that CA is unlikely to be a major VSP constituent; rather, it is a structural analog that elicits male-specific attraction via a parallel sensory circuit. The endogenous source of CA in C. remanei remains unresolved; our data do not establish whether females produce CA. Its structural and behavioral mimicry provides new insights into the complexity of chemosensory signaling and the potential for interspecies chemical eavesdropping in nematode ecology.</p
Continuous Diffusion Model for Self-supervised Denoising and Super-Resolution on Fluorescence Microscopy Images
No full textRecent studies have shown that Joint Denoising and Super-Resolution (JDSR) approaches can produce high-quality medical images. However, obtaining ground truth images for training is often infeasible in fluorescence microscopy (FM). Moreover, current JDSR methods can be impractical as they only rely on a fixed scale for resolution. Existing methods aim to learn a direct mapping between low-quality input and high-quality output in pixel space. In this paper, we introduce the Continuous Diffusion Model. This novel self-supervised method iteratively retrieves a noise-free image in continuous space, enabling us to generate clean images at any arbitrary resolutions. Our quantitative analysis confirms the effectiveness of our proposed method. We outperformed both supervised and self-supervised methods by an average of 30.5%/7.4% in terms of RMSE/SSIM. Our output preserves fine details and structures better than other state-of-the-art methods, as demonstrated by the qualitative analysis. The source code is available at https://github.com/colinsctsang/ContinuousDiffusionModel.</p
Real-time bi-level aircraft trajectory optimisation under the presence of unsteady wind
No full textIn the era towards green aviation, efficient flight operations and environmental sustainability are paramount concerns for airlines. In day-to-day operations, this can be achieved by flying aircraft in favourable wind conditions to reduce economic cost and environmental impact. This paper presents a bi-level aircraft trajectory optimisation algorithm designed for operations in unsteady wind conditions. The algorithm begins with a modified Dijkstra approach applied to a coarse grid, providing a globally optimised solution in the low-fidelity space. This is followed by high-fidelity optimisation using shape optimisation principles with free-form deformation. By reformulating the problem as a shape optimisation, the constrained optimisation is converted into an unconstrained one with fewer variables, thereby substantially reducing the computational cost. The bi-level approach accelerates convergence and minimises the risk of local minima. Applied to 2435 flights operated by our airline partner in 2022, the algorithm demonstrated average travel time reductions of 13.1%, 1.7%, and 1.2% for routes between Hong Kong and London, Sydney, and Singapore, respectively. The method is shown to be especially effective for east-west flights in the presence of jet streams. With GPU acceleration and CPU multiprocessing, the algorithm enables real-time trajectory optimisation, offering notable potential benefits in terms of sustainability, passenger comfort, and operational efficiency
Collaborative catalysis of single atoms and atomic clusters as dual sites for confined peroxymonosulfate activation to coordinate radical and singlet oxygen pathways
No full textSingle-atom catalysts have attracted tremendous research interest in advanced oxidation water treatment, while their efficiency and recyclability were restricted by speciation. Deliberate manipulation of single-atom (SAs) accumulation and distribution to tune the catalyst efficiency and route constitutes an area of great interest but a challenging pursuit. Here, we developed a series of novel spherical porous carbon materials from biomass waste with a three-dimensional distribution of Fe. Fe SAs and atomic clusters (ACs) are simultaneously formed in catalysts, and their distribution from the outer surface to the inner core varies with pyrrolic N content. The cooperation of SAs and ACs serves as a pair of redox sites linked with graphitic carbon to coordinate the electron circulation with peroxymonosulfate (PMS), generating radical and singlet oxygen for antibiotic degradation. The confined environment and the cooperation of dual sites perpendicular to the carbon plane were first reported, which enhanced the generation of reactive species. These advantages afforded efficient degradation (up to 229 % improved in kinetics and 133 % improved in degradation) with heightened recyclability (up to 150 % improved) compared to surface-doped SAs or ACs-dominated catalyst. Furthermore, this approach achieved efficient utilization of atomic metal sites compared to previously documented metal-based atomically dispersed catalysts. This study reveals the synergy of SAs and ACs in a confined environment for cyclic electron transfer and inspires a new version for designing atomically dispersed Fe within the 3D structure for advanced water decontamination.</p
Effects of anions on the separation of Li<sup>+</sup> and Na<sup>+</sup> using ZIF-8 membranes
No full textIn this work, we investigate the effects of anions (Cl−, NO3−, SO42−) on the separation of Li+ and Na+ using ZIF-8 membranes through molecular dynamics (MD) simulations. Numerical results show that anions play a crucial role in separation processes. Large anions, such as SO42−, cannot go through the ZIF-8 membrane and completely prohibit the transport of Li+ and Na+. In contrast, smaller anions (Cl− and NO3−) can permeate the membrane more easily, facilitating the migration of Li+ and Na+ through the membrane. Compared to Cl−, the transport of NO3− enhances the flux of Li+ and improves the Li+/Na+ separation ratio. Temperature effects are also examined. The Li+/Na+ separation ratio decreases as the temperature is increased, while the fluxes of Li+ and Na+ are enhanced due to higher diffusion coefficients of Li+ and Na+. A case study using NO3− to improve Li+/Na+ separation for the system containing SO42− is demonstrated. The findings in this work offer insights into the extraction of Li+ from seawater or salt lakes under complex environments.</p
On a class of complete permutation quadrinomials
No full textLet f(x)=ax3q+bx2q+1+cxq+2+dx3∈Fq2[x], where Fq2 is the finite field of order q2 and q=2m for some positive integer m. Tu et al. (Finite Fields Appl. 68: 1-20, 2020) proposed a sufficient condition under which f(x) is a complete permutation on Fq2. In this paper, we show that this sufficient condition is also necessary, and when f(x) is a complete permutation, then f(x) and f(x)+x are simultaneously linear equivalent to x2x‾ and x2x‾+γx for some γ∈Fq2⁎ satisfying ord(γq−1)=3. This result leads to a complete characterization of the complete permutation quadrinomials of the above form f(x).</p
SQ-DeAR: Sparsified and Quantized Gradient Compression for Distributed Training
No full textThe data-parallel distributed training technique is a de facto approach in training large-scale deep neural networks (DNNs) using synchronous stochastic gradient descent (S-SGD). However, S-SGD requires iteratively aggregating the distributed gradients through an AllReduce collective, which easily results in significant data communication across distributed GPUs and thus limits the scaling efficiency of the training system. In this paper, we propose an efficient and practical gradient sparsification and quantization algorithm, named SQ-DeAR, which not only significantly reduces the communication traffic through gradient sparsification and quantization, but also allows overlapping communications with both feed-forward and backpropagation computations through decoupling the communication collective operation. In addition, to improve the computation efficiency of gradient sparsification, we design a batched gradient sparsification to reduce the number of GPU launches. Performance evaluation on a 32-GPU cluster shows that SQ-DeAR outperforms state-of-the-art solutions by 1.17×-7.0×.</p
BuildingGPT: Query building semantic data using large language models and vector-graph retrieval-augmented generation
No full textSemantic web technologies offer a promising avenue for integrating heterogeneous building data across the entire Architecture, Engineering, Construction, and Operation (AECO) lifecycle. Yet querying such ontology-based data remains challenging, as it typically demands expertise in formal query languages like SPARQL Protocol and RDF Query Language (SPARQL). To address this, we present BuildingGPT, a novel framework that harnesses Large Language Models (LLMs) and a vector-graph retrieval-augmented generation (VG-RAG) pipeline. This approach empowers users to intuitively query building data grounded in semantic ontologies (e.g., the Brick Schema) using natural language. Evaluations on synthetic question types derived from 45 Brick Schema-based real building models—covering four complexities and six question types (18,555 samples)—demonstrate that VG-RAG outperforms baselines by 20–40 %, surpassing 90 % accuracy on four question types. By seamlessly integrating vector-based and graph-based retrieval to produce SPARQL queries, BuildingGPT simplifies data access for non-experts and enhances both reliability and responsiveness for complex, domain-specific inquiries. As a scalable, training-free solution compatible with diverse building ontologies, BuildingGPT significantly improves the accessibility and interoperability of semantic building data, ultimately fostering more informed decision-making and efficient asset management across the AECO industry. For a video demonstration, visit: https://www.youtube.com/watch?v=TtXVGVKuW8A</p
Sparse representation of irregularly sampled geotechnical data for probabilistic gridded spatial interpolation using dictionary learning
No full textSpatial interpolation constitutes a foundational component of spatial analysis, aiming to bridge discrete observations and continuous geoscientific modeling. However, irregular and clustered sampling layouts are common in real-world scenarios and often hinder the performance of feature-based spatial interpolation methods, such as dictionary learning. An unreliable and inconsistent representation of irregularly sampled data may lead to unstable interpolation results, reduced accuracy, and unreliable uncertainty quantification. To address these challenges, this study proposes a space transformation approach to derive a sparse representation of irregularly sampled geodata for gridded spatial interpolation. Spatially varying measurements with an irregular data distribution are first mapped onto a regular grid via affine transformation, and Bayesian compressive sensing is then adopted to generate conditional spatial predictions with quantified statistical uncertainty from the transformed dataset. Subsequently, the predicted random field samples are converted back into the original irregular data space, preserving the correspondence between points across different data spaces. The performance of the proposed method is demonstrated using both simulated data and real European soil moisture data. The results indicate that the proposed method facilitates concise representations of irregularly sampled geodata across different spatial resolutions and enables probabilistic spatial predictions from sparse data with consistent statistical uncertainty.</p
Traceable Algorithm Unrolling Network: An Interpretable Deep Sparse Representation Model for Mechanical Fault Diagnosis
No full textIn mechanical fault diagnosis (MFD), intelligent fault diagnosis (IFD) methods perform excellently regarding diagnosis accuracy. However, those methods are generally constructed with an excessive number of unprincipled parameters, resulting in uninterpretable architecture, ambiguity in the diagnosis process, and unclear decision-making basis. Thus, a traceable algorithm unrolling (TAU) network for interpretable MFD is proposed to overcome the above limitations. First, a mechanism-driven feature extractor (FE) is constructed by unrolling the iterative algorithm of sparse coding, aiming at encoding interpretable features from vibration signals. Second, a theory-based feature clustering (FC) algorithm is executed through the dynamic routing mechanism of the capsule network (CN), where the inner product serves as a measure for the association between input and output features. Finally, a post hoc interpretability strategy based on the coupling matrix is introduced to investigate how the TAU generates diagnostic results from the learned features and to verify whether these features are associated with faults, thereby enhancing the credibility of the diagnosis results. In addition, the simulation and experiment are designed to verify the decision-making mechanism and diagnostic performance of TAU. The results demonstrate that TAU makes diagnostic decisions based on the high-dimensional feature mapping associated with fault characteristic frequencies. Meanwhile, TAU outperforms the compared methods in fault diagnosis performance.</p