arXiv subjects

Gang Zhang

Publications and source records attributed to Gang Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2025

KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling

DETR is a novel end-to-end transformer architecture object detector, which significantly outperforms classic detectors when scaling up. In this paper, we focus on the compression of DETR with knowledge distillation. While knowledge distillation has been well-studied in classic detectors, there is a lack of researches on how to make it work effectively on DETR. We first provide experimental and theoretical analysis to point out that the main challenge in DETR distillation is the lack of consistent distillation points. Distillation points refer to the corresponding inputs of the predictions for student to mimic, which have different formulations in CNN detector and DETR, and reliable distillation requires sufficient distillation points which are consistent between teacher and student. Based on this observation, we propose the first general knowledge distillation paradigm for DETR (KD-DETR) with consistent distillation points sampling, for both homogeneous and heterogeneous distillation. Specifically, we decouple detection and distillation tasks by introducing a set of specialized object queries to construct distillation points for DETR. We further propose a general-to-specific distillation points sampling strategy to explore the extensibility of KD-DETR. Extensive experiments validate the effectiveness and generalization of KD-DETR. For both single-scale DAB-DETR and multis-scale Deformable DETR and DINO, KD-DETR boost the performance of student model with improvements of $2.6\%-5.2\%$. We further extend KD-DETR to heterogeneous distillation, and achieves $2.1\%$ improvement by distilling the knowledge from DINO to Faster R-CNN with ResNet-50, which is comparable with homogeneous distillation methods.The code is available at https://github.com/wennyuhey/KD-DETR.

cs.CV↗

arXiv2025

Theorem-Validated Reverse Chain-of-Thought Problem Generation for Geometric Reasoning

Large Multimodal Models (LMMs) face limitations in geometric reasoning due to insufficient Chain of Thought (CoT) image-text training data. While existing approaches leverage template-based or LLM-assisted methods for geometric CoT data creation, they often face challenges in achieving both diversity and precision. To bridge this gap, we introduce a two-stage Theorem-Validated Reverse Chain-of-Thought Reasoning Synthesis (TR-CoT) framework. The first stage, TR-Engine, synthesizes theorem-grounded geometric diagrams with structured descriptions and properties. The second stage, TR-Reasoner, employs reverse reasoning to iteratively refine question-answer pairs by cross-validating geometric properties and description fragments. Our approach expands theorem-type coverage, corrects long-standing misunderstandings, and enhances geometric reasoning. Fine-grained CoT improves theorem understanding and increases logical consistency by 24.5%. Our best models surpass the baselines in MathVista and GeoQA by 10.1% and 4.7%, outperforming advanced closed-source models like GPT-4o.

cs.AI↗

arXiv2025

An Information Theory-inspired Strategy for Automatic Network Pruning

Despite superior performance on many computer vision tasks, deep convolution neural networks are well known to be compressed on devices that have resource constraints. Most existing network pruning methods require laborious human efforts and prohibitive computation resources, especially when the constraints are changed. This practically limits the application of model compression when the model needs to be deployed on a wide range of devices. Besides, existing methods are still challenged by the missing theoretical guidance. In this paper we propose an information theory-inspired strategy for automatic model compression. The principle behind our method is the information bottleneck theory, i.e., the hidden representation should compress information with each other. We thus introduce the normalized Hilbert-Schmidt Independence Criterion (nHSIC) on network activations as a stable and generalized indicator of layer importance. When a certain resource constraint is given, we integrate the HSIC indicator with the constraint to transform the architecture search problem into a linear programming problem with quadratic constraints. Such a problem is easily solved by a convex optimization method with a few seconds. We also provide a rigorous proof to reveal that optimizing the normalized HSIC simultaneously minimizes the mutual information between different layers. Without any search process, our method achieves better compression tradeoffs comparing to the state-of-the-art compression algorithms. For instance, with ResNet-50, we achieve a 45.3%-FLOPs reduction, with a 75.75 top-1 accuracy on ImageNet. Codes are avaliable at https://github.com/MAC-AutoML/ITPruner/tree/master.

cs.CV↗

arXiv2025

On Data Synthesis and Post-training for Visual Abstract Reasoning

This paper is a pioneering work attempting to address abstract visual reasoning (AVR) problems for large vision-language models (VLMs). We make a common LLaVA-NeXT 7B model capable of perceiving and reasoning about specific AVR problems, surpassing both open-sourced (e.g., Qwen-2-VL-72B) and closed-sourced powerful VLMs (e.g., GPT-4o) with significant margin. This is a great breakthrough since almost all previous VLMs fail or show nearly random performance on representative AVR benchmarks. Our key success is our innovative data synthesis and post-training process, aiming to fully relieve the task difficulty and elicit the model to learn, step by step. Our 7B model is also shown to be behave well on AVR without sacrificing common multimodal comprehension abilities. We hope our paper could serve as an early effort in this area and would inspire further research in abstract visual reasoning.

cs.CV↗

arXiv2025

Spin-Polarized Antiferromagnetic Spintronics

Spin-polarized antiferromagnets (AFMs), including altermagnets, noncollinear AFMs, and two-dimensional layer-polarized AFMs, have emerged as transformative materials for next-generation spintronic and optoelectronic technologies. These systems uniquely combine spin-polarized electronic states with vanishing net magnetization, enabling ultrafast spin dynamics, high-density integration, and robustness against stray magnetic fields. Their unconventional symmetry-breaking mechanisms-governed by crystal symmetry, chiral spin textures, or interlayer potential control-give rise to emergent phenomena previously exclusive to ferromagnets: nonrelativistic spin-momentum locking, spontaneous anomalous transport phenomena, gate-tunable magneto-optical responses, and nonrelativistic spin-polarized current. This review systematically examines the fundamental principles linking symmetry, band topology, and transport properties across these material classes, synthesizing recent breakthroughs in both theory and experiment. We further identify critical challenges in achieving room-temperature functionality, scalable Neel vector control, and coherent spin-current manipulation, while outlining pathways to harness these materials for ultra-low-power memory, spin-logic architectures, and quantum information technologies.

cond-mat.mtrl-sci↗

arXiv2025

On the maximum partial-dual genus of a planar graph

Let $G$ be an embedded graph and $A$ an edge subset of $G$. The partial dual of $G$ with respect to $A$, denoted by $G^A$, can be viewed as the geometric dual $G^*$ of $G$ over $A$. If $A=E(G)$, then $G^A=G^*$. Denote by $γ(G^A)$ the genus of the embedded graph $G^A$. The maximum partial-dual genus of $G$ is defined as $$^\partialγ_{M}(G):=\max_{A \subseteq E(G)}γ(G^A).$$ For any planar graph $G$, it had been proved that $^\partialγ_{M}(G)$ does not rely on the embeddings of $G$. In this paper, we further prove that if $G$ is a connected planar graph of order $n\geq 2$, then $^{\partial}γ_{M}(G)\geq \frac{n-n_2-2n_1}{2}+1$, where $n_i$ is the number of vertices of degree $i$ in $G$. As a consequence, if $G$ is a connected planar graph of order $n$ with minimum degree at least 3, then $^{\partial}γ_{M}(G) \geq \frac{n}{2}+1$. Denote by $G^c$ the complement of a graph $G$ and by $χ(G^c)$ the chromatic number of $G^c$. Moreover, we prove that if $G \ncong K_4$ is a $λ$-edge-connected planar graph of order $n$, then $^{\partial}γ_{M}(G) \geq f(n,λ,χ(G^c))$, where $f(n,λ,χ(G^c))$ is a function of $n$, $λ$ and $χ(G^c)$. The first lower bound is tight for any $n$, and the second lower bound is tight for some 3-edge-connected graphs.

math.CO↗

arXiv2025

Ferroelectric Chirality-Driven Direction-Tunable and Spin-Invertible Corner States in 2D MOF-Based Magnetic Second-Order Topological Insulators

Despite the rapid progress in predicting 2D magnetic second-order topological insulators (SOTIs), effective strategies for manipulating their spin-polarized corner states remain largely unexplored. The interplay between ferroelectricity, chirality, magnetism, and topology presents an untapped opportunity for controlling these corner states. Here, we propose a novel approach for tuning spin-polarized corner states in 2D magnetic SOTIs by inducing ferroelectric chirality in 2D metal-organic frameworks (MOFs) with intrinsic structural flexibility. Through symmetry analysis, we strategically replace pyrazine (pyz) ligands with 2-pyrazinolate (2-pyzol) ligands in the 2D MOF Cr(pyz)2, leading to the emergence of a new 2D magnetic SOTI, Cr(2-pyzol)2, which facilitates ferroelectric chirality controlled spin-polarized corner states in both spin channels. Through first-principle calculations, we demonstrate that Cr(2-pyzol)2 belongs to ferroelectric chiral systems, and its corner states can be directionally tuned in real space and spin-inverted in spin space upon ferroelectric chirality switching. Our work represents the first attempt to simultaneously manipulate corner states in both real space and spin space, offering a new strategy for integrating ferroelectric chirality into 2D MOF-based magnetic SOTIs.

cond-mat.mtrl-sci↗

arXiv2025

Hyperparameter Optimization and Force Error Correction of Neuroevolution Potential for Predicting Thermal Conductivity of Wurtzite GaN

As a representative of wide-bandgap semiconductors, wurtzite gallium nitride (GaN) has been widely utilized in high-power devices due to high breakdown voltage and low specific on resistance. Accurate prediction of wurtzite GaN thermal conductivity is a prerequisite for designing effective thermal management systems of electronic applications. Machine learning driven molecular dynamics simulation offers a promising approach to predicting the thermal conductivity of large-scale systems without requiring predefined parameters. However, these methods often underestimate the thermal conductivity of materials with inherently high thermal conductivity due to the large predicted force error compared with first-principle calculation, posing a critical challenge for their broader application. In this study, we successfully developed a neuroevolution potential for wurtzite GaN and accurately predicted its thermal conductivity, 259 W/m-K at room temperatue, achieving excellent agreement with reported experimental measurements. The hyperparameters of neuroevolution potential (NEP) were optimized based on systematic analysis of reproduced energy and force, structural feature, computational efficiency. Furthermore, a force prediction error correction method was implemented, effectively reducing the error caused by the additional force noise in the Langevin thermostat by extrapolating to the zero-force error limit. This study provides valuable insights and hold significant implication for advancing efficient thermal management technologies in wide bandgap semiconductor devices.

cond-mat.mtrl-sci↗

arXiv2025

Softening of Vibrational Modes and Anharmonicity Induced Thermal Conductivity Reduction in a-Si:H at High Temperatures

Hydrogenated amorphous silicon (a-Si:H) has garnered considerable attention in the semiconductor industry, particularly for its use in solar cells and passivation layers for high performance silicon solar cells, owing to its exceptional photoelectric properties and scalable manufacturing processes. A comprehensive understanding of thermal transport mechanism in a-Si:H is essential for optimizing thermal management and ensuring the reliable operation of these devices. In this study, we developed a neuroevolution machine learning potential based on first-principles calculations of energy, forces, and virial, which enables accurate modeling of interatomic interactions in both a-Si:H and a-Si systems. Using the homogeneous nonequilibrium molecular dynamics (HNEMD) method, we systematically investigated the thermal conductivity of a-Si:H and a-Si across a temperature range of 300-1000 K and hydrogen concentrations ranging from 6 to 12 at%. Our simulation results found that thermal conductivity of a-Si:H with 12 at% hydrogen was significantly reduced by 12% compared to that of a-Si at 300 K. We analyzed the spectral thermal conductivity, vibrational density of states and lifetimes of vibrational modes, and revealed the softening of vibrational modes and anharmonicity effects contribute to the reduction of thermal conductivity as temperature and hydrogen concentration increase. Furthermore, the influence of hydrogen concentration and temperature on diffuson and propagon contribution to thermal conductivity of a-Si:H was revealed. This study provides valuable insights for developing thermal management strategies in silicon-based semiconducting devices and advances the understanding of thermal transport in amorphous systems.

cond-mat.mtrl-sci↗

arXiv2025

Interpretable Face Anti-Spoofing: Enhancing Generalization with Multimodal Large Language Models

Face Anti-Spoofing (FAS) is essential for ensuring the security and reliability of facial recognition systems. Most existing FAS methods are formulated as binary classification tasks, providing confidence scores without interpretation. They exhibit limited generalization in out-of-domain scenarios, such as new environments or unseen spoofing types. In this work, we introduce a multimodal large language model (MLLM) framework for FAS, termed Interpretable Face Anti-Spoofing (I-FAS), which transforms the FAS task into an interpretable visual question answering (VQA) paradigm. Specifically, we propose a Spoof-aware Captioning and Filtering (SCF) strategy to generate high-quality captions for FAS images, enriching the model's supervision with natural language interpretations. To mitigate the impact of noisy captions during training, we develop a Lopsided Language Model (L-LM) loss function that separates loss calculations for judgment and interpretation, prioritizing the optimization of the former. Furthermore, to enhance the model's perception of global visual features, we design a Globally Aware Connector (GAC) to align multi-level visual representations with the language model. Extensive experiments on standard and newly devised One to Eleven cross-domain benchmarks, comprising 12 public datasets, demonstrate that our method significantly outperforms state-of-the-art methods.

cs.CV↗

arXiv2024

ALoRE: Efficient Visual Adaptation via Aggregating Low Rank Experts

Parameter-efficient transfer learning (PETL) has become a promising paradigm for adapting large-scale vision foundation models to downstream tasks. Typical methods primarily leverage the intrinsic low rank property to make decomposition, learning task-specific weights while compressing parameter size. However, such approaches predominantly manipulate within the original feature space utilizing a single-branch structure, which might be suboptimal for decoupling the learned representations and patterns. In this paper, we propose ALoRE, a novel PETL method that reuses the hypercomplex parameterized space constructed by Kronecker product to Aggregate Low Rank Experts using a multi-branch paradigm, disentangling the learned cognitive patterns during training. Thanks to the artful design, ALoRE maintains negligible extra parameters and can be effortlessly merged into the frozen backbone via re-parameterization in a sequential manner, avoiding additional inference latency. We conduct extensive experiments on 24 image classification tasks using various backbone variants. Experimental results demonstrate that ALoRE outperforms the full fine-tuning strategy and other state-of-the-art PETL methods in terms of performance and parameter efficiency. For instance, ALoRE obtains 3.06% and 9.97% Top-1 accuracy improvement on average compared to full fine-tuning on the FGVC datasets and VTAB-1k benchmark by only updating 0.15M parameters.

cs.CV↗

arXiv2024

Continual SFT Matches Multimodal RLHF with Negative Supervision

Multimodal RLHF usually happens after supervised finetuning (SFT) stage to continually improve vision-language models' (VLMs) comprehension. Conventional wisdom holds its superiority over continual SFT during this preference alignment stage. In this paper, we observe that the inherent value of multimodal RLHF lies in its negative supervision, the logit of the rejected responses. We thus propose a novel negative supervised finetuning (nSFT) approach that fully excavates these information resided. Our nSFT disentangles this negative supervision in RLHF paradigm, and continually aligns VLMs with a simple SFT loss. This is more memory efficient than multimodal RLHF where 2 (e.g., DPO) or 4 (e.g., PPO) large VLMs are strictly required. The effectiveness of nSFT is rigorously proved by comparing it with various multimodal RLHF approaches, across different dataset sources, base VLMs and evaluation metrics. Besides, fruitful of ablations are provided to support our hypothesis. We hope this paper will stimulate further research to properly align large vision language models.

cs.LG↗

arXiv2024

Isolation partitions in graphs

Let $G$ be a graph and $k \geq 3$ an integer. A subset $D \subseteq V(G)$ is a $k$-clique (resp., cycle) isolating set of $G$ if $G-N[D]$ contains no $k$-clique (resp., cycle). In this paper, we prove that every connected graph with maximum degree at most $k$, except $k$-clique, can be partitioned into $k+1$ disjoint $k$-clique isolating sets, and that every connected claw-free subcubic graph, except 3-cycle, can be partitioned into four disjoint cycle isolating sets. As a consequence of the first result, every $k$-regular graph can be partitioned into $k+1$ disjoint $k$-clique isolating sets.

math.CO↗

arXiv2024

Stable dynamic helix state in the nonintegrable XXZ Heisenberg model

We investigate the influence of external fields on the stability of spin helix states in an XXZ Heisenberg model. Exact diagonalization on a finite system shows that random transverse fields in the x and y directions drive the transition from integrability to nonintegrability. In such a system, the helix state can be regarded as a quantum scar. Simultaneously, the presence of uniform z field enables the helix state to better maintain its dynamical nature, allowing for a clearer understanding of its evolutionary behavior. However, the entanglement entropy reveals that irrespective of the presence of a uniform z field, as long as the system remains chaotic, the scar extent of the helix state shows no significant variation.

quant-ph↗

arXiv2024

Improving Multi-modal Large Language Model through Boosting Vision Capabilities

We focus on improving the visual understanding capability for boosting the vision-language models. We propose \textbf{Arcana}, a multiModal language model, which introduces two crucial techniques. First, we present Multimodal LoRA (MM-LoRA), a module designed to enhance the decoder. Unlike traditional language-driven decoders, MM-LoRA consists of two parallel LoRAs -- one for vision and one for language -- each with its own parameters. This disentangled parameters design allows for more specialized learning in each modality and better integration of multimodal information. Second, we introduce the Query Ladder adapter (QLadder) to improve the visual encoder. QLadder employs a learnable ``\textit{ladder}'' structure to deeply aggregates the intermediate representations from the frozen pretrained visual encoder (e.g., CLIP image encoder). This enables the model to learn new and informative visual features, as well as remaining the powerful capabilities of the pretrained visual encoder. These techniques collectively enhance Arcana's visual perception power, enabling it to leverage improved visual information for more accurate and contextually relevant outputs across various multimodal scenarios. Extensive experiments and ablation studies demonstrate the effectiveness and generalization capability of our Arcana. The code and re-annotated data are available at \url{https://arcana-project-page.github.io}.

cs.CV↗

arXiv2024

SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection

LiDAR-based 3D object detection plays an essential role in autonomous driving. Existing high-performing 3D object detectors usually build dense feature maps in the backbone network and prediction head. However, the computational costs introduced by the dense feature maps grow quadratically as the perception range increases, making these models hard to scale up to long-range detection. Some recent works have attempted to construct fully sparse detectors to solve this issue; nevertheless, the resulting models either rely on a complex multi-stage pipeline or exhibit inferior performance. In this work, we propose SAFDNet, a straightforward yet highly effective architecture, tailored for fully sparse 3D object detection. In SAFDNet, an adaptive feature diffusion strategy is designed to address the center feature missing problem. We conducted extensive experiments on Waymo Open, nuScenes, and Argoverse2 datasets. SAFDNet performed slightly better than the previous SOTA on the first two datasets but much better on the last dataset, which features long-range detection, verifying the efficacy of SAFDNet in scenarios where long-range detection is required. Notably, on Argoverse2, SAFDNet surpassed the previous best hybrid detector HEDNet by 2.6% mAP while being 2.1x faster, and yielded 2.1% mAP gains over the previous best sparse detector FSDv2 while being 1.3x faster. The code will be available at https://github.com/zhanggang001/HEDNet.

cs.CV↗

arXiv2024

FullAnno: A Data Engine for Enhancing Image Comprehension of MLLMs

Multimodal Large Language Models (MLLMs) have shown promise in a broad range of vision-language tasks with their strong reasoning and generalization capabilities. However, they heavily depend on high-quality data in the Supervised Fine-Tuning (SFT) phase. The existing approaches aim to curate high-quality data via GPT-4V, but they are not scalable due to the commercial nature of GPT-4V and the simplicity of the prompts used to instruct the model. To this end, we devised the FullAnno system, which is a data engine that can generate large-scale, high-quality, and fine-grained image annotations consisting of the category and position of objects, region descriptions, text information, as well as image dense captions. This engine is characterized by its cascade annotation process, which involves multiple expert models and employs rich prompts to instruct LLMs in generating dense image captions. We re-annotated the COCO and Visual Genome datasets using our FullAnno system, tripling the number of object annotations and increasing the length of the original image captions by a factor of 15. Experiments show that the regenerated annotation can significantly enhance the capabilities of LLaVA-v1.5 on several benchmarks. The re-annotated data are available at: https://arcana-project-page.github.io

cs.CV↗

arXiv2024

Add-SD: Rational Generation without Manual Reference

Diffusion models have exhibited remarkable prowess in visual generalization. Building on this success, we introduce an instruction-based object addition pipeline, named Add-SD, which automatically inserts objects into realistic scenes with rational sizes and positions. Different from layout-conditioned methods, Add-SD is solely conditioned on simple text prompts rather than any other human-costly references like bounding boxes. Our work contributes in three aspects: proposing a dataset containing numerous instructed image pairs; fine-tuning a diffusion model for rational generation; and generating synthetic data to boost downstream tasks. The first aspect involves creating a RemovalDataset consisting of original-edited image pairs with textual instructions, where an object has been removed from the original image while maintaining strong pixel consistency in the background. These data pairs are then used for fine-tuning the Stable Diffusion (SD) model. Subsequently, the pretrained Add-SD model allows for the insertion of expected objects into an image with good rationale. Additionally, we generate synthetic instances for downstream task datasets at scale, particularly for tail classes, to alleviate the long-tailed problem. Downstream tasks benefit from the enriched dataset with enhanced diversity and rationale. Experiments on LVIS val demonstrate that Add-SD yields an improvement of 4.3 mAP on rare classes over the baseline. Code and models are available at https://github.com/ylingfeng/Add-SD.

cs.CV↗