arXiv subjects

Jiahe Fan

Publications and source records attributed to Jiahe Fan.

10 recordsLinked to original sources

arXiv2026

Training-Free Knowledge Transfer Across Model Scales through Activation-Guided Pruning

Heterogeneous model fusion seeks to combine models that differ in tasks, initializations, architectures, or scales. We study an underexplored cross-scale setting: improving a small recipient language model with a stronger donor despite substantial architectural mismatch. We ask whether useful capabilities can be transferred without explicit neuron-wise semantic alignment. Building on the observation that truncating a large model to a smaller architecture and injecting it with a tiny mixing weight can already improve the recipient, we propose Activation-Prune-Merge (APM), an activation-guided framework for cross-scale fusion. APM constructs task-conditioned activation maps on the donor, selects salient layers, hidden dimensions, attention heads, and MLP neurons to prune it to the recipient architecture, and injects the resulting donor slice into the original recipient using a micro interpolation coefficient. This formulation treats the donor as a source of concentrated functional components rather than requiring precise structural transplantation. Across 16 benchmarks spanning reasoning, mathematics, code generation, instruction following, and classification, APM improves the overall average accuracy from 55.5% to 60.6% over the original 3B recipient. RTE accuracy increases from 64.3% to 82.3%, QNLI from 52.3% to 65.7%, and BoolQ from 70.8% to 79.2%. Analyses of injection ratios and sequential multi-stage fusion further suggest that activation-guided extraction improves the quality of the transferable donor slice while preserving the small-ratio fusion regime. These results provide evidence that cross-scale heterogeneous fusion can succeed without explicit semantic alignment when the donor contribution is sufficiently concentrated and carefully selected.

cs.LG↗

arXiv2026

Understanding Knowledge Transfer Mechanism in Heterogeneous MLLM Fusion: A Simple Linear Approach

Training-free fusion of heterogeneous multimodal large language models (MLLMs) provides a direct route for cross-scale capability transfer, yet improvements in aggregate performance do not reveal what a smaller model actually inherits. Existing studies are largely designed and evaluated on limited task sets or aggregate metrics; as evaluation expands to broader task collections, whether different capabilities can transfer across scales remains poorly understood. To investigate this question, we introduce Cross-Scale Directional Parameter Injection (CDPI), a simple linear probe to analyze cross-scale knowledge transfer during heterogeneous fusion. A local theoretical analysis indicates that knowledge transfer selectivity is determined at first order by capability-dependent responses to a shared injection direction, while second-order curvature effects constrain the effective transfer regime. Across four Qwen3-VL model pairs and twelve multimodal benchmarks, our experiments reveal a consistent pattern of selectivity: gains concentrate on reasoning, particularly high-level reasoning, whereas perception performance remains close to that of the original target model. Component-wise ablations further show that high-level reasoning gains arise primarily from the language model, while ratio analysis finds that positive selective transfer occurs mainly in the small-ratio regime. These findings recast cross-scale heterogeneous MLLM fusion as selective language-side reasoning transfer within a narrow, low-interference regime, rather than broad capability inheritance.

cs.CV↗

arXiv2026

Rethinking Heterogeneous LLM Merging: A Weighted Model Averaging Perspective

Can large language models with substantially different parameter spaces be merged by direct weighted averaging, without training or semantic alignment? Existing heterogeneous fusion methods typically introduce distillation, adapters, learned latent spaces, routing, or feature alignment, leaving open whether a simpler recipe can work for genuinely different billion-parameter checkpoints. We revisit this counterintuitive question through training-free dimensional adaptation followed by ratio-controlled interpolation. In union-style merging, we expand the smaller model into the larger parameter space; in intersection-style merging, we truncate the larger model into the smaller parameter space. Across Qwen-family model pairs and benchmarks covering mathematical reasoning, code generation, language understanding, commonsense reasoning, knowledge, and instruction following, deterministic expansion largely preserves the source model function, and small-ratio interpolation can improve over strong source checkpoints by transferring complementary capabilities. However, near-balanced interpolation often collapses, and task-level results reveal a seesaw effect in which gains on some capabilities coexist with regressions on others. These results show that simple parameter averaging, when paired with lightweight dimensional adaptation and carefully controlled ratios, is a surprisingly strong baseline for heterogeneous LLM merging, suggesting that the limits of direct weighted fusion may also bound what more complex heterogeneous merging methods can achieve at scale.

cs.AI↗

arXiv2026

Unsupervised Collaborative Domain Adaptation for Driving Scene Parsing

Reliable driving scene parsing is a fundamental capability for autonomous vehicles operating in open and dynamic driving environments. However, adapting perception models to new deployment domains remains challenging because pixel-level annotations are expensive to obtain, while source-domain data are often inaccessible due to privacy, security, or ownership constraints. Existing source-free unsupervised domain adaptation methods typically rely on a single pre-trained source model, which makes the adapted perception system vulnerable to source-specific biases and limits its robustness under diverse road layouts, illumination conditions, weather patterns, and traffic conditions. This article presents an unsupervised collaborative domain adaptation (UCDA) framework for driving scene parsing in a source-free setting, which transfers complementary knowledge from multiple pre-trained source models to a unified target model without accessing any original source samples. To compare predictions from independently trained models, UCDA constructs a class-level prototype memory bank and estimates cross-model prediction reliability through prototype similarity, reducing the effect of inconsistent confidence scales across source models. Based on the resulting complementary supervision, UCDA adopts a two-stage transfer strategy: multiple source models are first refined on unlabeled target-domain driving data through collaborative optimization with positive and negative consistency constraints, and their validated expertise is then distilled into a single deployable target model. Comprehensive evaluations on public driving-scene datasets and real-world data collected from an autonomous vehicle platform demonstrate that UCDA effectively consolidates complementary multi-source knowledge, improving target-domain scene parsing reliability and generalization across diverse driving environments.

cs.CV↗

arXiv2026

Structured prototype regularization for synthetic-to-real driving scene parsing

Driving scene parsing is critical for autonomous vehicles to operate reliably in complex real-world traffic environments. To reduce the reliance on costly pixel-level annotations, synthetic datasets with automatically generated labels have become a popular alternative. However, models trained on synthetic data often perform poorly when applied to real-world scenes due to the synthetic-to-real domain gap. Despite the success of unsupervised domain adaptation in narrowing this gap, most existing methods mainly focus on global feature alignment while overlooking the semantic structure of the feature space. As a result, semantic relations among classes are insufficiently modeled, limiting the model's ability to generalize. To address these challenges, this study introduces a novel unsupervised domain adaptation framework that explicitly regularizes semantic feature structures to significantly enhance driving scene parsing performance in real-world scenarios. Specifically, the proposed method enforces inter-class separation and intra-class compactness by leveraging class-specific prototypes, thereby enhancing the discriminability and structural coherence of feature clusters. An entropy-based noise filtering strategy improves the reliability of pseudo labels, while a pixel-level attention mechanism further refines feature alignment. Extensive experiments on representative benchmarks demonstrate that the proposed method consistently outperforms recent state-of-the-art methods. These results underscore the importance of preserving semantic structure for robust synthetic-to-real adaptation in driving scene parsing tasks.

cs.CV↗

arXiv2022

Computer Vision for Road Imaging and Pothole Detection: A State-of-the-Art Review of Systems and Algorithms

Computer vision algorithms have been prevalently utilized for 3-D road imaging and pothole detection for over two decades. Nonetheless, there is a lack of systematic survey articles on state-of-the-art (SoTA) computer vision techniques, especially deep learning models, developed to tackle these problems. This article first introduces the sensing systems employed for 2-D and 3-D road data acquisition, including camera(s), laser scanners, and Microsoft Kinect. Afterward, it thoroughly and comprehensively reviews the SoTA computer vision algorithms, including (1) classical 2-D image processing, (2) 3-D point cloud modeling and segmentation, and (3) machine/deep learning, developed for road pothole detection. This article also discusses the existing challenges and future development trends of computer vision-based road pothole detection approaches: classical 2-D image processing-based and 3-D point cloud modeling and segmentation-based approaches have already become history; and Convolutional neural networks (CNNs) have demonstrated compelling road pothole detection results and are promising to break the bottleneck with the future advances in self/un-supervised learning for multi-modal semantic segmentation. We believe that this survey can serve as practical guidance for developing the next-generation road condition assessment systems.

cs.CV↗

arXiv2021

Multi-Scale Feature Fusion: Learning Better Semantic Segmentation for Road Pothole Detection

This paper presents a novel pothole detection approach based on single-modal semantic segmentation. It first extracts visual features from input images using a convolutional neural network. A channel attention module then reweighs the channel features to enhance the consistency of different feature maps. Subsequently, we employ an atrous spatial pyramid pooling module (comprising of atrous convolutions in series, with progressive rates of dilation) to integrate the spatial context information. This helps better distinguish between potholes and undamaged road areas. Finally, the feature maps in the adjacent layers are fused using our proposed multi-scale feature fusion module. This further reduces the semantic gap between different feature channel layers. Extensive experiments were carried out on the Pothole-600 dataset to demonstrate the effectiveness of our proposed method. The quantitative comparisons suggest that our method achieves the state-of-the-art (SoTA) performance on both RGB images and transformed disparity images, outperforming three SoTA single-modal semantic segmentation networks.

cs.CV↗

arXiv2018

Mobile Robot Localisation and Navigation Using LEGO NXT and Ultrasonic Sensor

Mobile robots are becoming increasingly important both for individuals and industries. Mobile robotic technology is not only utilised by experts in this field but is also very popular among amateurs. However, implementing a mobile robot to perform tasks autonomously can be expensive because of the need for various types of sensors and the high price of robot platforms. Hence, in this paper we present a mobile robot localisation and navigation system which uses a LEGO ultrasonic sensor in an indoor map based on the LEGO MINDSTORM NXT. This provides an affordable and ready-to-use option for most robot fans. In this paper, an effective method is proposed to extract useful information from the distorted readings collected by the ultrasonic sensor. Then, the particle filter is used to localise the robot. After robot's position is estimated, a sampling-based path planning method is proposed for the robot navigation. This method reduces the robot accumulative motion error by minimising robot turning times and covering distances. The robot localisation and navigation algorithms are implemented in MATLAB. Simulation results show an average accuracy between 1 and 3 cm for three different indoor map locations. Furthermore, experiments performed in a real setup show the effectiveness of the proposed methods.

cs.RO↗

arXiv2015

Anomalous valley polarization in monolayer MoSe2

Modern electronic devices heavily rely on the accurate control of charge and spin of electrons. The emergence of controllable valley degree of freedom brings new possibilities and presents a promising prospect towards valleytronics. Recently, valley excitation selected by chiral optical pumping has been observed in monolayer MoS2. In this work, we report polarized photoluminescence (PL) measurements for monolayer MoSe2, another member of the family of transition-metal-dichalcogenides (MX2), and observe drastic difference from the outcomes of MoS2. In particular, we identify a valley polarization (VP) up to 70% for B exciton, while that for A exciton is less than 3%. Besides, we also find a small but finite negative VP for A- trion. These results reveal several new intra- and inter-valley scattering processes which significantly affect valley polarization, hence provide new insights into exciton physics in monolayer MX2 and possible valleytronic applications.

cond-mat.mtrl-sci↗

arXiv2013

Low-Frequency Raman Modes and Electronic Excitations In Atomically Thin MoS2 Crystals

Atomically thin MoS$_{2}$ crystals have been recognized as a quasi-2D semiconductor with remarkable physics properties. This letter reports our Raman scattering measurements on multilayer and monolayer MoS$_{2}$, especially in the low-frequency range ($<$50 cm$^{-1}$). We find two low-frequency Raman modes with contrasting thickness dependence. With increasing the number of MoS$_{2}$ layers, one shows a significant increase in frequency while the other decreases following a 1/N (N denotes layer-number) trend. With the aid of first-principle calculations we assign the former as the shear mode $E_{2g}^{2}$ and the latter as the compression vibrational mode. The opposite evolution of the two modes with thickness demonstrates novel vibrational modes in atomically thin crystal as well as a new and more precise way to characterize thickness of atomically thin MoS$_{2}$ films. In addition, we observe a broad feature around 38 cm$^{-1}$ (~5 meV) which is visible only under near-resonance excitation and pinned at the fixed energy independent of thickness. We interpret the feature as an electronic Raman scattering associated with the spin-orbit coupling induced splitting in conduction band at K points in their Brillouin zone.

cond-mat.mes-hall↗