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
Learning Heterogeneous Mixture of Scene Experts for Large-Scale Neural Radiance Fields
No full textRecent Neural Radiance Field (NeRF) methods on large-scale scenes have demonstrated promising results and underlined the importance of scene decomposition for scalable NeRFs. Although these methods achieved reasonable scalability, there are several critical problems remaining unexplored in the existing large-scale NeRF modeling methods, i.e., learnable decomposition, modeling scene heterogeneity, and modeling efficiency. In this paper, we introduce Switch-NeRF++, a Heterogeneous Mixture of Hash Experts (HMoHE) network that addresses these challenges within a unified framework. Our framework is a highly scalable NeRF that learns heterogeneous decomposition and heterogeneous Neural Radiance Fields efficiently for large-scale scenes in an end-to-end manner. In our framework, a gating network learns to decompose scenes into partitions and allocates 3D points to specialized NeRF experts. This gating network is co-optimized with the experts by our proposed Sparsely Gated Mixture of Experts (MoE) NeRF framework. Our network architecture incorporates a hash-based gating network and distinct heterogeneous hash experts. The hash-based gating efficiently learns the decomposition of the large-scale scene. The distinct heterogeneous hash experts consist of hash grids of different resolution ranges. This enables effective learning of the heterogeneous representation of different decomposed scene parts within large-scale complex scenes. These design choices make our framework an end-to-end and highly scalable NeRF solution for real-world large-scale scene modeling to achieve both quality and efficiency. We evaluate our accuracy and scalability on existing large-scale NeRF datasets. Additionally, we also introduce a new dataset with very large-scale scenes (> 6.5km2) from UrbanBIS. Extensive experiments demonstrate that our approach can be easily scaled to various large-scale scenes and achieve state-of-the-art scene rendering accuracy. Furthermore, our method exhibits significant efficiency gains, with an 8x acceleration in training and a 16x acceleration in rendering compared to the best-performing competitor Switch-NeRF. The codes and trained models will be released in https://github.com/MiZhenxing/Switch-NeRF.</p
UV-assisted coaxial printing: a multiphysics model on two-phase flow with in-situ photopolymerization
No full textCoaxial printing enables precise fabrication of core-shell structures for applications in functional composites, tissue engineering and flexible electronics. However, existing models fail to capture the multiphysics complexity of UV-assisted coaxial printing, where fluid dynamics, interfacial interactions and photopolymerization kinetics are tightly coupled. This study develops a novel multiphysics framework integrating two-phase flow dynamics, photopolymerization kinetics and structural evolution to model UV-assisted coaxial printing of hollow microfluidic interconnects. The model combines autocatalytic photopolymerization kinetics, phase-filed based two-phase flow simulations and cure-dependent material properties. Experimental validation confirms the accuracy of models in predicting flow regimes and printed outcomes. Transition from dripping to jetting was found when both inner and outer flow rates increased. In-situ photopolymerization induced additional channel shrinkage to final printed outcomes. Printed outcomes also demonstrate that outer flow rate inversely controls channel diameter, while inner flow rate proportionally influences it. This work enables predictive optimization of coaxial printing for targeted applications by multiphysics modeling.</p
Microstructural regulation of interfacial transition zone by biochar: A novel strategy for enhancing fire resistance
No full textDesign and development of biochar-cement composites (BCC) is an innovative strategy to reduce the carbon footprint of cement-based materials. Incorporating biochar properly can help regulate the microstructure of the interfacial transition zone (ITZ) in cement composites. Yet the thermal response of porous biochar within the ITZ and the fire resistance of BCC remain unclear. This study investigated the regulatory mechanisms and reinforcing effects of biochar incorporation on the fire resistance of cementitious composites. The results revealed that biochar as fine aggregates could mitigate thermal-mismatch stress between the aggregates and the cement matrix when exposed to high temperatures, probably attributed to the low thermal expansion coefficient (1 × 10−6 K−1), large plastic deformation capability, and strong elasticity of biochar. Additionally, biochar strengthened the ITZ at 500 °C by facilitating the formation of a high-modulus CaCO3 reinforcement layer, resulting in micromechanical properties exceeding 45 GPa. The superior thermal insulation capacity of biochar prevented local cracks at the interlayer, resulting in a ∼31% reduction in porosity and ∼58% reduction in decomposition of hydration products. Furthermore, full-scale fire testing provided direct support to the microscale observation that replacing fine aggregates with 30 wt% biochar could comply with the fire safety standards while maintaining acceptable structural functionality. However, excessive biochar content (40 wt%) resulted in a significant increase in porosity and cracks due to high-temperature degradation. This study elucidated the microstructural mechanisms by which biochar can enhance the fire resistance of cement-based materials through ITZ regulation, providing a novel strategy for developing high-performance, low-carbon construction materials.</p
Harnessing strengthening-metastability synergy for extreme work hardening in additively manufactured titanium alloys
No full textRapid bottom-up fabrication via additive manufacturing (AM) unlocks unprecedented design freedom for geometrically complex and lightweight titanium (Ti) alloys, a critical material for next-generation aerospace systems and 3C (computer, communication and consumer electronics) products. However, conventional AM Ti alloys face a persistent dilemma: achieving yield strengths above 1 GPa catastrophically degrades work hardening (typically < 2 GPa) and uniform ductility (< 5%). Here, we harness a strengthening-metastability synergy strategy via AM to demonstrate the powerful CoCrNi additive-strengthened Ti alloy with an outstanding combination of loss-free yield strength and drastically enhanced work hardening. Unlike traditional metastable alloys with incomplete phase transformation (β → β/α'), our design triggers a complete two-step martensitic transformation (β → β/α' → α'/α' twin) during deformation, without residual matrix and forming hierarchically mutual twin structures. This unique transformation pathway sustains a successive work hardening, achieving a record work hardening rate of 5.7 GPa and uniform elongation of 9.3% (triple that of base alloy), while maintaining 1030 MPa yield strength. The dual emphasis on synergy strategy and mechanistic innovation via the non-equilibrium AM process directly addresses the structural sector’s urgent need for high-performance yet sustainable metallic solutions.</p
LLM-driven Medical Report Generation via Communication-efficient Heterogeneous Federated Learning
No full textLarge Language Models (LLMs) have demonstrated significant potential in Medical Report Generation (MRG), yet their development requires large amounts of medical image-report pairs, which are commonly scattered across multiple centers. Centralizing these data is exceptionally challenging due to privacy regulations, thereby impeding model development and broader adoption of LLM-driven MRG models. To address this challenge, we present FedMRG, the first framework that leverages Federated Learning (FL) to enable privacy-preserving, multi-center development of LLM-driven MRG models, specifically designed to overcome the critical challenge of communication-efficient LLM training under multi-modal data heterogeneity. To start with, our framework tackles the fundamental challenge of communication overhead in federated LLM tuning by employing low-rank factorization to efficiently decompose parameter updates, significantly reducing gradient transmission costs and making LLM-driven MRG feasible in bandwidth-constrained FL settings. Furthermore, we observed the dual heterogeneity in MRG under the FL scenario: varying image characteristics across medical centers, as well as diverse reporting styles and terminology preferences. To address the data heterogeneity, we further enhance FedMRG with (1) client-aware contrastive learning in the MRG encoder, coupled with diagnosis-driven prompts, which capture both globally generalizable and locally distinctive features while maintaining diagnostic accuracy; and (2) a dual-adapter mutual boosting mechanism in the MRG decoder that harmonizes generic and specialized adapters to address variations in reporting styles and terminology. Through extensive evaluation of our established FL-MRG benchmark, we demonstrate the generalizability and adaptability of FedMRG, underscoring its potential in harnessing multi-center data and generating clinically accurate reports while maintaining communication efficiency.</p
