arXiv subjects

Dongliang Chen

Publications and source records attributed to Dongliang Chen.

6 recordsLinked to original sources

arXiv2026

APEX: Learning Adaptive Priorities for Multi-Objective Alignment in Vision-Language Generation

Multi-objective alignment for text-to-image generation is commonly implemented via static linear scalarization, but fixed weights often fail under heterogeneous rewards, leading to optimization imbalance where models overfit high-variance, high-responsiveness objectives (e.g., OCR) while under-optimizing perceptual goals. We identify two mechanistic causes: variance hijacking, where reward dispersion induces implicit reweighting that dominates the normalized training signal, and gradient conflicts, where competing objectives produce opposing update directions and trigger seesaw-like oscillations. We propose APEX (Adaptive Priority-based Efficient X-objective Alignment), which stabilizes heterogeneous rewards with Dual-Stage Adaptive Normalization and dynamically schedules objectives via P^3 Adaptive Priorities that combine learning potential, conflict penalty, and progress need. On Stable Diffusion 3.5, APEX achieves improved Pareto trade-offs across four heterogeneous objectives, with balanced gains of +1.31 PickScore, +0.35 DeQA, and +0.53 Aesthetics while maintaining competitive OCR accuracy, mitigating the instability of multi-objective alignment.

cs.CV↗

arXiv2025

CHIPS: Efficient CLIP Adaptation via Curvature-aware Hybrid Influence-based Data Selection

Adapting CLIP to vertical domains is typically approached by novel fine-tuning strategies or by continual pre-training (CPT) on large domain-specific datasets. Yet, data itself remains an underexplored factor in this process. We revisit this task from a data-centric perspective: Can effective data selection substitute for large-scale datasets in CPT? We introduce CHIPS (Curvature-aware Hybrid Influence in Projection Subspace), which assigns each image-text pair a utility score that integrates three complementary factors aligned with three goals: faithfulness via a curvature-aware and Newton-style alignment computed in CLIP's end-point subspace; scalability via an InfoNCE-aware curvature estimator with Johnson-Lindenstrauss (JL) sketching; and retention via a selection-aware relevance weight combined with learnability to balance target adaptation against general-domain preservation. We justify this design theoretically by proving a lower-bound guarantee on the proxy's correlation with full-parameter alignment and by characterizing the bias-variance trade-offs introduced by curvature mixing and JL sketching. We evaluate CHIPS empirically across various settings: 1) CHIPS attains state-of-the-art performance among selection baselines on 17 medical benchmarks, matches full-dataset CPT with 30% of the data, and outperforms half-dataset CPT using only 10%; 2) on 31 general-domain benchmarks, CHIPS yields the least performance drop under all retention ratios.

cs.LG↗

arXiv2024

Out-of-plane orientated self-trapped excitons enabled polarized light guiding in 2D perovskites

Active optical waveguides combine light source and waveguides together in an individual component, which are essential for the integrated photonic chips. Although 1D luminescent materials based optical waveguides were extensively investigated, 2D waveguides allow photons to flow within a plane and serve as an ideal component for the ultracompact photonic circuits. Nevertheless, light guiding in 2D planar structures normally relies on the precise control of molecular orientation, which is complicated and low yield. Here, we report a strategy to guide polarized light in 2D microflakes by making use of the out-of-plane (OP) orientation of self-trapped excitons in as-synthesized 2D perovskite microplates. A space confined crystallization method is developed to synthesize 2D perovskite microflakes with dominated broad self-trapped excitons emission at room temperature, which are highly OP orientated with a percentage of the OP component over 85%. Taking advantages of the negligible absorption coefficient and improved coupling efficiency of OP orientated self-trapped exciton emission to the planar waveguide mode of the as-synthesized perovskite microflakes, we have achieved a broadband polarized light guiding with a full width at half maximum over 120 nm. Our findings provide a promising platform for the development of ultracompact photonic circuits.

physics.optics↗

arXiv2024

Microscopic Insights into Fatigue Mechanism in Wurtzite Ferroelectric Al$_{0.65}$Sc$_{0.35}$N: Oxygen Infiltration Enabled Grain Amorphization Spanning Boundary to Bulk

For the first time, the fatigue behavior involving external oxygen in highly Sc-doped AlN ferroelectric film was observed using transmission electron microscope techniques. Despite increasing the Sc composition in AlScN film contributes to reducing the device operation voltage, the inherent affinity of Sc for oxygen introduces instability in device performance. In this study, oxygen incorporation at top electrode edges and grain boundaries accompanied with an increase in current leakage and the disappearance of ferroelectric properties, was observed in nanoscale after long-term field cycling. This observation indicates the emergence of non-ferroelectric and even amorphous states. This presented work revealed solid experimental evidence of an oxygen-involved fatigue mechanism, providing valuable insights into the physical nature of the ferroelectric properties of AlScN films.

cond-mat.mtrl-sci↗

arXiv2021

MeetDurian: A Gameful Mobile App to Prevent COVID-19 Infection

The COVID-19 problem has not gone away with the passing of the seasons. Even though most countries have achieved remarkable results in fighting against epidemic diseases and preventing and controlling viruses, the general public is still far from understanding the new crown virus and lacks imagination on its transmission law. In this paper, we propose MeetDurian: a cross-platform mobile application that exploits a location-based game to improve users' hygiene habits and reduce virus dispersal. We present its main features, its architecture, and its core technologies. Finally, we report a set of experiments that prove the acceptability and usability of MeetDurian. An illustrative demo of the mobile app features is shown in the following video: https://youtu.be/Vqg7nFDQuOU.

cs.HC↗

arXiv2008

Electronic structure of LaFe1-xCoxAsO from first principle calculations

Based on the first-principles calculations, we have investigated the geometry, binding properties, density of states and band structures of the novel superconductor LaFe1-xCoxAsO and its parent compounds with the ZrCuSiAs structure. We demonstrate that La-O bond and TM-As (TM=Fe or Co) bond are both strongly covalent, while the LaO and TMAs layers have an almost ionic interaction through the Bader charge analysis. Partial substitution of iron with cobalt modify the Fermi level from a steep edge to a flat slope, which explains why in this system Co doping suppresses the spin density wave (SDW) transition.

cond-mat.supr-con↗