arXiv subjects

Chenyan Wu

Publications and source records attributed to Chenyan Wu.

18 recordsLinked to original sources

arXiv2026

DPN-LE: Dual Personality Neuron Localization and Editing for Large Language Models

With the widespread adoption of large language models (LLMs), understanding their personality representation mechanisms has become critical. As a novel paradigm in Personality Editing, most existing methods employ neuron-editing to locate and modify LLM neurons, requiring changes to numerous neurons and leading to significant performance degradation. This raises a fundamental question: Are all modified neurons directly related to personality representation? In this work, we investigate and quantify this specificity through assessments of general capability impact and representation-level patterns. We find that: 1) Current methods can change personalities but reduce overall performance. 2) Neurons are multifunctional, connecting personality traits and general knowledge. 3) Opposing personality traits demonstrate distinctly mutually exclusive representation patterns. Motivated by these findings, we propose DPN-LE (Dual Personality Neuron Localization and Editing), which identifies personality-specific neurons by contrasting MLP activations between high-trait and low-trait samples. DPN-LE constructs layer-wise steering vectors and applies dual-criterion filtering based on Cohen's $d$ effect size and activation magnitude to isolate mutually exclusive neuron subsets. Sparse linear intervention on these neurons enables precise personality control at inference time. Using only 1,000 contrastive sample pairs per trait, DPN-LE intervenes on $\sim$0.5\% of neurons while achieving competitive personality control and substantially better capability preservation across reasoning tasks. Experiments on LLaMA-3-8B-Instruct and Qwen2.5-7B-Instruct demonstrate the effectiveness and generalizability of our approach.

cs.CL↗

arXiv2026

Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics

Methods for controlling large language models (LLMs), including local weight fine-tuning, LoRA-based adaptation, and activation-based interventions, are often studied in isolation, obscuring their connections and making comparison difficult. In this work, we present a unified view that frames these interventions as dynamic weight updates induced by a control signal, placing them within a single conceptual framework. Building on this view, we propose a unified preference-utility analysis that separates control effects into preference, defined as the tendency toward a target concept, and utility, defined as coherent and task-valid generation, and measures both on a shared log-odds scale using polarity-paired contrastive examples. Across methods, we observe a consistent trade-off between preference and utility: stronger control increases preference while predictably reducing utility. We further explain this behavior through an activation manifold perspective, in which control shifts representations along target-concept directions to enhance preference, while utility declines primarily when interventions push representations off the model's valid-generation manifold. Finally, we introduce a new steering approach SPLIT guided by this analysis that improves preference while better preserving utility. Code is available at https://github.com/zjunlp/EasyEdit/blob/main/examples/SPLIT.md.

cs.CL↗

arXiv2026

Associated Representations of finite pattern groups

In this paper, we consider the construction of irreducible representations of finite pattern groups in terms of Panov's associative polarization, which is a finite-field analogue of Kirillov's orbital method. Using this construction, first, we are able to classify the irreducible representations of the unipotent radical of the standard parabolic subgroups of $\mathrm{GL}_n$ with 4 parts; second, we can parameterize irreducible characters of degree $q$ in terms of coadjoint orbits of cardinality $q^2$, for any finite pattern groups $G$ over $\mathbb{F}_q,$ where $\mathbb{F}_q$ is a finite field with $q$ elements.

math.RT↗

arXiv2025

Intertwining Algebras and Affine Hecke Algebras for Finite Central Extensions of Classical $p$-adic Groups with Application to Metaplectic Groups

For a finite central extension $\tilde{G}$ of a classical $p$-adic reductive group, we consider the endomorphism algebra of some induced projective generator \`a la Bernstein of the category of smooth representations of $\tilde{G}$. In the case where the Levi subgroups decompose, we compute this algebra to get a result similar to the one previously obtained by the first author for classical $p$-adic groups, showing that this intertwining algebra is a twisted semi-direct product of an affine Hecke algebra with parameters by a twisted finite group algebra. We discuss also the case with non-decomposed Levi subgroups. We then give an application to the category of genuine representations of a $p$-adic metaplectic group. Using results of C. M\oe glin relative to the Howe correspondence, we show that the Bernstein components of these groups are equivalent to tensor products of categories of unipotent representations of classical groups. This generalizes a previous result of the first author. It implies an equivalence of categories between the category of genuine representations of the $p$-adic metaplectic group and the direct sums of those of smooth representations of the corresponding split odd special orthogonal group and its pure inner form.

math.RT↗

arXiv2025

DuoSpaceNet: Leveraging Both Bird's-Eye-View and Perspective View Representations for 3D Object Detection

Multi-view camera-only 3D object detection largely follows two primary paradigms: exploiting bird's-eye-view (BEV) representations or focusing on perspective-view (PV) features, each with distinct advantages. Although several recent approaches explore combining BEV and PV, many rely on partial fusion or maintain separate detection heads. In this paper, we propose DuoSpaceNet, a novel framework that fully unifies BEV and PV feature spaces within a single detection pipeline for comprehensive 3D perception. Our design includes a decoder to integrate BEV and PV features into unified detection queries, as well as a feature enhancement strategy that enriches different feature representations. In addition, DuoSpaceNet can be extended to handle multi-frame inputs, enabling more robust temporal analysis. Extensive experiments on nuScenes dataset show that DuoSpaceNet surpasses both BEV-based baselines (e.g., BEVFormer) and PV-based baselines (e.g., Sparse4D) in 3D object detection and BEV map segmentation, verifying the effectiveness of our proposed design.

cs.CV↗

arXiv2023

Unlocking the Emotional World of Visual Media: An Overview of the Science, Research, and Impact of Understanding Emotion

The emergence of artificial emotional intelligence technology is revolutionizing the fields of computers and robotics, allowing for a new level of communication and understanding of human behavior that was once thought impossible. While recent advancements in deep learning have transformed the field of computer vision, automated understanding of evoked or expressed emotions in visual media remains in its infancy. This foundering stems from the absence of a universally accepted definition of "emotion", coupled with the inherently subjective nature of emotions and their intricate nuances. In this article, we provide a comprehensive, multidisciplinary overview of the field of emotion analysis in visual media, drawing on insights from psychology, engineering, and the arts. We begin by exploring the psychological foundations of emotion and the computational principles that underpin the understanding of emotions from images and videos. We then review the latest research and systems within the field, accentuating the most promising approaches. We also discuss the current technological challenges and limitations of emotion analysis, underscoring the necessity for continued investigation and innovation. We contend that this represents a "Holy Grail" research problem in computing and delineate pivotal directions for future inquiry. Finally, we examine the ethical ramifications of emotion-understanding technologies and contemplate their potential societal impacts. Overall, this article endeavors to equip readers with a deeper understanding of the domain of emotion analysis in visual media and to inspire further research and development in this captivating and rapidly evolving field.

cs.CV↗

arXiv2023

MUG: Multi-human Graph Network for 3D Mesh Reconstruction from 2D Pose

Reconstructing multi-human body mesh from a single monocular image is an important but challenging computer vision problem. In addition to the individual body mesh models, we need to estimate relative 3D positions among subjects to generate a coherent representation. In this work, through a single graph neural network, named MUG (Multi-hUman Graph network), we construct coherent multi-human meshes using only multi-human 2D pose as input. Compared with existing methods, which adopt a detection-style pipeline (i.e., extracting image features and then locating human instances and recovering body meshes from that) and suffer from the significant domain gap between lab-collected training datasets and in-the-wild testing datasets, our method benefits from the 2D pose which has a relatively consistent geometric property across datasets. Our method works like the following: First, to model the multi-human environment, it processes multi-human 2D poses and builds a novel heterogeneous graph, where nodes from different people and within one person are connected to capture inter-human interactions and draw the body geometry (i.e., skeleton and mesh structure). Second, it employs a dual-branch graph neural network structure -- one for predicting inter-human depth relation and the other one for predicting root-joint-relative mesh coordinates. Finally, the entire multi-human 3D meshes are constructed by combining the output from both branches. Extensive experiments demonstrate that MUG outperforms previous multi-human mesh estimation methods on standard 3D human benchmarks -- Panoptic, MuPoTS-3D and 3DPW.

cs.CV↗

arXiv2023

Theta correspondence and simple factors in global Arthur parameters

By using results on poles of $L$-functions and theta correspondence, we give a bound on $b$ for $(χ,b)$-factors of the global Arthur parameter of a cuspidal automorphic representation $π$ of a classical group or a metaplectic group where $χ$ is a conjugate self-dual automorphic character and $b$ is an integer which is the dimension of an irreducible representation of $\mathrm{SL}_{2}(\mathbf{CC})$. We derive a more precise relation when $π$ lies in a generic global $A$-packet.

math.NT↗

arXiv2023

Bodily expressed emotion understanding through integrating Laban movement analysis

Body movements carry important information about a person's emotions or mental state and are essential in daily communication. Enhancing the ability of machines to understand emotions expressed through body language can improve the communication of assistive robots with children and elderly users, provide psychiatric professionals with quantitative diagnostic and prognostic assistance, and aid law enforcement in identifying deception. This study develops a high-quality human motor element dataset based on the Laban Movement Analysis movement coding system and utilizes that to jointly learn about motor elements and emotions. Our long-term ambition is to integrate knowledge from computing, psychology, and performing arts to enable automated understanding and analysis of emotion and mental state through body language. This work serves as a launchpad for further research into recognizing emotions through analysis of human movement.

cs.AI↗

arXiv2023

Learning to Adapt to Online Streams with Distribution Shifts

Test-time adaptation (TTA) is a technique used to reduce distribution gaps between the training and testing sets by leveraging unlabeled test data during inference. In this work, we expand TTA to a more practical scenario, where the test data comes in the form of online streams that experience distribution shifts over time. Existing approaches face two challenges: reliance on a large test data batch from the same domain and the absence of explicitly modeling the continual distribution evolution process. To address both challenges, we propose a meta-learning approach that teaches the network to adapt to distribution-shifting online streams during meta-training. As a result, the trained model can perform continual adaptation to distribution shifts in testing, regardless of the batch size restriction, as it has learned during training. We conducted extensive experiments on benchmarking datasets for TTA, incorporating a broad range of online distribution-shifting settings. Our results showed consistent improvements over state-of-the-art methods, indicating the effectiveness of our approach. In addition, we achieved superior performance in the video segmentation task, highlighting the potential of our method for real-world applications.

cs.LG↗

arXiv2022

Periods and $(χ,b)$-factors of Cuspidal Automorphic Forms of Metaplectic Groups

We give constraints on the existence of $(χ,b)$-factors in the global $A$-parameter of a genuine cuspidal automorphic representation $σ$ of a metaplectic group in terms of the invariant, lowest occurrence index, of theta lifts to odd orthogonal groups. We also give a refined result that relates the invariant, first occurrence index, to non-vanishing of period integrals of residues of Eisenstein series associated to the cuspidal datum $χ\otimesσ$. This complements our previous results for symplectic groups.

math.NT↗

arXiv2022

Poles of Eisenstein series and theta lifts for unitary groups

We derive a precise relation of poles of Eisenstein series associated to the cuspidal datum $χ\otimesσ$ and the lowest occurrence of theta lifts of a cuspidal automorphic representation $σ$ of a unitary group, where $χ$ is a conjugate self-dual character. A key ingredient of the proof is the computation of period integrals of truncated Eisenstein series.

math.NT↗

arXiv2022

Virtual Abelian varieties of $\mathrm{GL}_2$-type

This paper studies a class of Abelian varieties that are of $\GL_2$-type and with isogenous classes defined over a number field $k$. We treat the cases when their endomorphism algebras are either (1) a totally real field $K$ or (2) a totally indefinite quaternion algebra over a totally real field $K$. Among the isogenous class of such an Abelian variety, we identify one whose Galois conjugates can be described in terms of actions of Atkin-Lehner operators and the class group of $K$. Thus we deduce that such Abelian varieties are parametrised by finite quotients of certain PEL Shimura varieties. These new families of moduli spaces are further analysed when they are of dimension $2$. We provide explicit numerical bounds for when they are surfaces of general type. In addition, for two particular examples, we show that they are both rational surfaces by computing the coordinates of inequivalent elliptic points and studying the intersections of Hirzebruch cycles with exceptional divisors.

math.AG↗

arXiv2020

MEBOW: Monocular Estimation of Body Orientation In the Wild

Body orientation estimation provides crucial visual cues in many applications, including robotics and autonomous driving. It is particularly desirable when 3-D pose estimation is difficult to infer due to poor image resolution, occlusion or indistinguishable body parts. We present COCO-MEBOW (Monocular Estimation of Body Orientation in the Wild), a new large-scale dataset for orientation estimation from a single in-the-wild image. The body-orientation labels for around 130K human bodies within 55K images from the COCO dataset have been collected using an efficient and high-precision annotation pipeline. We also validated the benefits of the dataset. First, we show that our dataset can substantially improve the performance and the robustness of a human body orientation estimation model, the development of which was previously limited by the scale and diversity of the available training data. Additionally, we present a novel triple-source solution for 3-D human pose estimation, where 3-D pose labels, 2-D pose labels, and our body-orientation labels are all used in joint training. Our model significantly outperforms state-of-the-art dual-source solutions for monocular 3-D human pose estimation, where training only uses 3-D pose labels and 2-D pose labels. This substantiates an important advantage of MEBOW for 3-D human pose estimation, which is particularly appealing because the per-instance labeling cost for body orientations is far less than that for 3-D poses. The work demonstrates high potential of MEBOW in addressing real-world challenges involving understanding human behaviors. Further information of this work is available at https://chenyanwu.github.io/MEBOW/.

cs.CV↗

arXiv2017

Periods and $(χ,b)$-factors of Cuspidal Automorphic Forms of Symplectic Groups

In this paper, we introduce a new family of period integrals attached to irreducible cuspidal automorphic representations $σ$ of symplectic groups $\mathrm{Sp}_{2n}(\mathbb{A})$, which detects the right-most pole of the $L$-function $L(s,σ\timesχ)$ for some character $χ$ of $F^\times\backslash\mathbb{A}^\times$ of order at most $2$, and hence the occurrence of a simple global Arthur parameter $(χ,b)$ in the global Arthur parameter $ψ$ attached to $σ$. We also give a characterisation of first occurrences of theta correspondence by (regularised) period integrals of residues of certain Eisenstein series.

math.NT↗

arXiv2016

On a Critical Case of Rallis Inner Product Formula

Let $π$ be a genuine cuspidal representation of the metaplectic group of rank $n$. We consider the theta lifts to the orthogonal group associated to a quadratic space of dimension $2n+1$. We show a case of regularised Rallis inner product formula that relates the pairing of theta lifts to the central value of the Langlands $L$-function of $π$ twisted by a character. The bulk of this article focuses on proving a case of regularised Siegel-Weil formula, on which the Rallis inner product formula is based and whose proof is missing in the literature.

math.NT↗

arXiv2014

On $(χ,b)$-factors of Cuspidal Automorphic Representations of Unitary Groups I

Following the idea of [GJS09] for orthogonal groups, we introduce a new family of period integrals for cuspidal automorphic representations $σ$ of unitary groups and investigate their relation with the occurrence of a simple global Arthur parameter $(χ,b)$ in the global Arthur parameter $ψ_σ$ associated to $σ$, by the endoscopic classification of Arthur ([Art13], [Mok13], [KMSW14]). The argument uses the theory of theta correspondence. This can be viewed as a part of the $(χ,b)$-theory outlined in [Jia14] and can be regarded as a refinement of the theory of theta correspondences and poles of certain $L$-functions, which was outlined in [Ral91].

math.NT↗

arXiv2014

Irreducibility of Theta Lifting for Unitary Groups

This article shows that for unitary dual reductive pairs the first occurrence of theta lift of an irreducible cuspidal automorphic representation is irreducible. It also proves a refined tower property for theta lifts and the involutive property for twisted theta lifts.

math.NT↗