Searcharxiv⌕ Search

arXiv subjects

Jaehoon Jeong

Publications and source records attributed to Jaehoon Jeong.

At least 19 recordsLinked to original sources
arXiv2026

OBJECTION! Lawyer Agents Mitigate Guilty Bias in Legal Judgment Prediction

Legal Judgment Prediction (LJP) models are typically trained on documents that describe facts from a prosecutorial perspective. Existing datasets further exhibit severe label imbalance toward guilty outcomes. Consequently, these models suffer from "Guilty Bias", blindly accepting the prosecution's narrative as objective truth. Previous studies employing three-step reasoning structures or training on synthetically generated innocence data improve overall accuracy, but they still fail to mitigate bias at inference time. In this paper, we introduce OBJECTION, an inference-time pipeline that integrates an Adversarial Lawyer Agent into each 3-step reasoning of offense, unlawfulness, and culpability. Unlike generic critics, our agent actively challenges the model's presumptions of guilt by injecting legal defense arguments at each reasoning stage. To thoroughly evaluate this, we present a new "Natural Innocent" dataset including 3.4k real-world cases, overcoming the limitations of synthetic innocence benchmarks. Test results show that OBJECTION drastically reduces the False Guilty Rate (FGR) from 82.93% (SOTA baseline) to 16.69%, proving its capability to perform substantive legal reasoning. This work denotes a key progress toward aligning Legal AI with the presumption of innocence.

cs.CL↗
arXiv2026

Partial Waves for Multipositivity

We develop a partial-wave formalism providing the missing spin-resolved framework for multipositivity among massless planar amplitudes. We present a systematic procedure for constructing kinematic charts realizing the complex-forward limit. Applying it to obtain the four-, five-, and six-point charts, we show how to derive the one-to-two and one-to-three wavefunctions and construct the associated partial waves. The resulting formalism places higher-point multipositivity on the same footing as conventional partial-wave analyses of two-to-two scattering.

hep-th↗
arXiv2026

Deep Learning-based Surrogate Modelling of the LOD Method for Multiscale Problems

Multiscale problems are notoriously difficult to tackle using traditional numerical methods, as accurately resolving fine-scale features often requires prohibitively fine discretizations. This challenge is particularly pronounced in applications such as materials science, fluid dynamics, climate systems, chemical processes, and complex networks. Recent neural operator models provide a promising data-driven alternative, but frequently struggle to achieve sufficient accuracy in the presence of strongly heterogeneous or oscillatory coefficients. In this work, we focus on the solution of elliptic PDEs with rough and high-contrast inputs. The Localized Orthogonal Decomposition (LOD) method is a well-established numerical approach for such problems, but it comes, however, at a substantial computational cost. We investigate the performance of popular neural operator architectures on these challenging multiscale problems and identify key limitations in their ability to resolve fine-scale structure. To overcome these challenges, we introduce LOD-MSNO (LOD-Multiscale Neural Operator), a hybrid approach that leverages the LOD method as a strong multiscale prior by building on its representation of the solution as a linear combination of problem-adapted basis functions, while addressing its main computational bottlenecks through data-driven operator learning. We further provide theoretical error estimates for the proposed coefficient-learning framework. Lastly, we demonstrate the potential of our proposed method to outperform current neural operator baselines in terms of accuracy for challenging multiscale inputs, while mainly retaining the computational efficiency of neural operator models.

math.NA↗
arXiv2026

Multipositivity Constrains the Chiral Lagrangian

The chiral Lagrangian is a cornerstone of modern particle physics, offering a systematic and quantitative description of low-energy pions. Using tools from the modern scattering amplitudes program, we show that consistent multiparticle dynamics impose novel constraints on the coupling constants of this theory. In the planar limit, these constraints imply that certain Wilson coefficients of the chiral Lagrangian are bounded from below by the chiral anomaly. Our results reveal a subtle connection between the anomalous and nonanomalous sectors of the underlying strong interactions, while introducing a novel formulation of multipositivity bounds that holds for any planar tree-level theory.

hep-th↗
arXiv2026

Explicit Conditions for Diagnosing Tree-Level Unitarity

We explicitly present all coupling conditions required for tree-level unitarity (tree unitarity) in theories with a finite number of massive and massless particles of spin up to 1. They allow us to diagnose tree unitarity of a system using only its particle content in the mass basis, without reconstructing the full Lagrangian. We show that all four-point amplitudes whose high-energy growth is canceled by tree unitarity conditions are on-shell constructible, thereby motivating the recursive construction of four-point amplitudes. By examining their high-energy growth, we derive tree unitarity conditions for four-point amplitudes. Imposing these conditions to simplify the Lagrangian structure, we use the Stückelberg formulation to derive the tree unitarity conditions arising from all higher-point amplitudes. We show that all tree unitarity conditions are fully captured up to five-point amplitudes, ensuring no necessity of examining higher-point ones. We apply our results to systematically examine tree unitarity conditions in the dark sector with a massive dark photon and dark matter particles of spin up to 1, and extract the essential features for mass generation in the massive dark photon case. In addition, we show that our results allow us to conclude that theories without scalars require an infinite tower of vectors and fermions for tree unitarity. Finally unitarity and related issues in the Higgs portal VDM are discussed in brief.

hep-ph↗
arXiv2025

Causal Bounds on EFTs with anomalies with a Pseudoscalar, Photons, and Gravitons

Theories with pseudoscalars that couple through anomalies (such as axion models) are of particular phenomenological interest. We carry out a comprehensive analysis of all bounds obtainable from bootstrapping the amplitudes when a pseudoscalar couples to photons and gravitons. This allows us to find new cutoff scales of theories with anomalies that are more restrictive than those obtained from naive perturbative analysis. Our results are especially relevant for holographic models, as the bounds determine the allowed region of the five-dimensional EFTs, for example, by imposing strong bounds on Chern-Simons terms. We also consider modifications of General Relativity in photon--graviton couplings and show that current experiments are sensitive to these effects only if new physics appears at $\sim 10^{-10}$ eV.

hep-th↗
arXiv2025

Temporal Grounding as a Learning Signal for Referring Video Object Segmentation

Referring Video Object Segmentation (RVOS) aims to segment and track objects in videos based on natural language expressions, requiring precise alignment between visual content and textual queries. However, existing methods often suffer from semantic misalignment, largely due to indiscriminate frame sampling and supervision of all visible objects during training -- regardless of their actual relevance to the expression. We identify the core problem as the absence of an explicit temporal learning signal in conventional training paradigms. To address this, we introduce MeViS-M, a dataset built upon the challenging MeViS benchmark, where we manually annotate temporal spans when each object is referred to by the expression. These annotations provide a direct, semantically grounded supervision signal that was previously missing. To leverage this signal, we propose Temporally Grounded Learning (TGL), a novel learning framework that directly incorporates temporal grounding into the training process. Within this frame- work, we introduce two key strategies. First, Moment-guided Dual-path Propagation (MDP) improves both grounding and tracking by decoupling language-guided segmentation for relevant moments from language-agnostic propagation for others. Second, Object-level Selective Supervision (OSS) supervises only the objects temporally aligned with the expression in each training clip, thereby reducing semantic noise and reinforcing language-conditioned learning. Extensive experiments demonstrate that our TGL framework effectively leverages temporal signal to establish a new state-of-the-art on the challenging MeViS benchmark. We will make our code and the MeViS-M dataset publicly available.

cs.CV↗
arXiv2025

Latest Object Memory Management for Temporally Consistent Video Instance Segmentation

In this paper, we present Latest Object Memory Management (LOMM) for temporally consistent video instance segmentation that significantly improves long-term instance tracking. At the core of our method is Latest Object Memory (LOM), which robustly tracks and continuously updates the latest states of objects by explicitly modeling their presence in each frame. This enables consistent tracking and accurate identity management across frames, enhancing both performance and reliability through the VIS process. Moreover, we introduce Decoupled Object Association (DOA), a strategy that separately handles newly appearing and already existing objects. By leveraging our memory system, DOA accurately assigns object indices, improving matching accuracy and ensuring stable identity consistency, even in dynamic scenes where objects frequently appear and disappear. Extensive experiments and ablation studies demonstrate the superiority of our method over traditional approaches, setting a new benchmark in VIS. Notably, our LOMM achieves state-of-the-art AP score of 54.0 on YouTube-VIS 2022, a dataset known for its challenging long videos. Project page: https://seung-hun-lee.github.io/projects/LOMM/

cs.CV↗
arXiv2025

No gauge cancellation at high energy in the five-vector $R_ξ$ gauge

We propose a novel $R_ξ$ gauge in the five-vector (5V) framework within the Abelian Higgs model. In the Cartesian basis of the complex Higgs field, the 5V $R_ξ$ gauge ensures non-divergent tree-level amplitudes for each Feynman diagram in the high-energy limit. This framework pinpoints the origin of high-energy divergences in tree-level amplitudes for each diagram, providing a criterion for quantifying the degree of divergences from other gauges. The 5V description necessitates treating the Goldstone field as the fifth gauge-field component, offering deeper insight into the dynamics of massive gauge bosons, particularly its longitudinal mode. The impact of this framework is demonstrated by rigorously comparing tree-level amplitudes from the 5V $R_ξ$ gauge with those from the conventional 4V $R_ξ$ gauge and the Feynman diagram gauge, the latter of which exhibits no gauge cancellation, similar to the 5V $R_ξ$ gauge.

hep-ph↗
arXiv2024

Vetoing all the Higgs imposters in $H\to \ell^-\ell^+ Z$

We develop an effective and methodical algorithm for the construction of general covariant four-point $H\ell\ell Z$ vertices, accommodating leptons $\ell=e, μ$, and designed to handle a boson $H$ of any integer spin, not merely confined to spins up to 2. While our numerical analysis assumes the $H$-boson mass to be $m_H=125\,{\rm GeV}$, the analytical framework we propose is versatile, enabling the examination of various mass as well as spin scenarios. These meticulously devised general covariant four-point $H\ell\ell Z$ vertices are pivotal in vetoing all the imposters of the Standard Model Higgs boson holding the spin-0 and even-parity quantum numbers, especially in one of its primary decay channels, the three-body decay process $H\to \ell^-\ell^+ Z$, observable at the Large Hadron Collider. Our innovative strategy encompasses the analysis of all the effectively allowed scenarios, extending beyond the limitations of previous investigations on the Higgs spin and parity determinations in the decay $H \to \ell^-\ell^+ Z$. Based on the significantly expanded scheme, we demonstrate that the Higgs boson imposter of any spin and parity can be definitively vetoed by leveraging threshold effects and angular correlations, even though achieving such conclusive results in practical and exhaustive analyses necessitates high event rates.

hep-ph↗
arXiv2024

Hunting for Hypercharge Anapole Dark Matter in All Spin Scenarios

We conduct a combined analysis to investigate dark matter (DM) with hypercharge anapole moments, focusing on scenarios where Majorana DM particles with spin 1/2, 1, 3/2, and 2 interact exclusively with Standard Model particles through U(1)$_{Y}$ hypercharge anapole terms for the first time. For completeness, we construct general effective U(1) gauge-invariant three-point vertices. These enable the generation of hypercharge gauge-invariant interaction vertices for both a virtual photon $γ$ and a virtual $Z$ boson with two identical massive Majorana particles of any non-zero spin $s$, after the spontaneous breaking of electroweak gauge symmetry. For complementarity, we adopt effective operators tailored to each dark matter spin allowing crossing symmetry. We calculate the relic abundance, analyze current constraints and future sensitivities from dark matter direct detection and collider experiments, and apply the conceptual naive perturbativity bound. Our estimations based on a generalized vertex calculation demonstrate that the scenario with a higher-spin DM is more stringently constrained than a lower-spin DM, primarily due to the reduced annihilation cross-section and/or the enhanced rate of LHC mono-jet events. As a remarkable outcome, the spin-2 anapole DM scenario is almost entirely excluded, while the high-luminosity LHC exhibits high sensitivities in probing spin-1 and 3/2 scenarios, except for a tiny parameter range of DM mass around 1 TeV. A significant portion of the remaining parameter space in the spin-1/2 DM scenario can be explored through upcoming Xenon experiments, with more than 20 ton-year exposure equivalent to approximately 5 years of running the XENONnT experiment.

hep-ph↗
arXiv2023

All the matrix elements of covariant tensor currents of massless particles in the covariant formulation

We present an efficient algorithm for constructing all the matrix elements of covariant tensor currents of massless particles of arbitrary spins in the covariant formulation. The construction of matrix elements can be taken simply by assembling the basic matrix elements which are derived from the basic three-point vertices. We obtain the selection rules for the decay of an off-shell massive particle into two identical massless particles which are the generalization of the Landau-Yang (LY) theorem. After showing how to identify fully the discontinuity between the matrix elements for the spacelike and lightlike momentum transfers, we derive all the matrix elements of conserved tensor currents of massless particles including high spins, from which the Weinberg and Witten (WW) theorem is automatically extracted with additional limits on the particles.

hep-ph↗
arXiv2017

Thermodynamic Volume and the Extended Smarr Relation

We continue to explore the scaling transformation in the reduced action formalism of gravity models. As an extension of our construction, we consider the extended forms of the Smarr relation for various black holes, adopting the cosmological constant as the bulk pressure as in some literatures on black holes. Firstly, by using the quasi-local formalism for charges, we show that, in a general theory of gravity, the volume in the black hole thermodynamics could be defined as the thermodynamic conjugate variable to the bulk pressure in such a way that the first law can be extended consistently. This, so called, thermodynamic volume can be expressed explicitly in terms of the metric and field variables. Then, by using the scaling transformation allowed in the reduced action formulation, we obtain the extended Smarr relation involving the bulk pressure and the thermodynamic volume. In our approach, we do not resort to Euler's homogeneous scaling of charges while incorporating the would-be hairy contribution without any difficulty.

hep-th↗
arXiv2015

Scaling symmetry and scalar hairy Lifshitz black holes

By utilizing the scaling symmetry of the reduced action for planar black holes, we obtain the corresponding conserved charge. We use the conserved charge to find the generalized Smarr relation of static hairy planar black holes in various dimensions. Our results not only reproduce the relation in the various known cases but also give the new relation in the Lifshitz planar black holes with the scalar hair.

hep-th↗
arXiv2015

Frame-independent holographic conserved charges

We propose the modified form of the conventional holographic conserved charges which provides us the frame-independent expressions for charges. This form is also preferable to the conventional one since it is independent of the holographic renormalization scheme. We show the frame and scheme independence through the matching of our holographic expression to the covariant bulk expression of conserved charges. As an explicit example, we consider five-dimensional AdS Kerr black holes and show that our form of holographic conserved charges gives us the identical expressions in the rotating and non-rotating frames.

hep-th↗
arXiv2014

Quasi-local conserved charges and holography

We construct a quasi-local formalism for conserved charges in a theory of gravity in the presence of matter fields which may have slow falloff behaviors at the asymptotic infinity. This construction depends only on equations of motion and so it is irrespective of ambiguities in the total derivatives of the Lagrangian. By using identically conserved currents, we show that this formalism leads to the same expressions of conserved charges as those in the covariant phase space approach. At the boundary of the asymptotic AdS space, we also introduce an identically conserved boundary current which has the same structure as the bulk current and then show that this boundary current gives us the holographic conserved charges identical with those from the boundary stress tensor method. In our quasi-local formalism we present a general proof that conserved charges from the bulk potential are identical with those from the boundary current. Our results can be regarded as the extension of the existing results on the equivalence of conserved charges by the covariant phase space approach and by the boundary stress tensor method.

hep-th↗
arXiv2014

SUSY properties of warped AdS$_3$

We examine supersymmetric properties of null-warped AdS$_3$, or alternatively Schrodinger geometries, dual to putative warped CFTs in two dimensions. We classify super Schrodinger subalgebras of the superalgebra psu(1, 1$|$2) $\oplus$ psu(1, 1$|$2), corresponding to the superconformal algebra of the AdS$_3 \times$ S$^3$ geometry. We comment on geometric realisations and provide a string theory description with enhanced supersymmetry in terms of intersecting D3-branes. For type IIB supergravity solutions based on T$^{1,1}$, we consider the relationship between five-dimensional Schrodinger solutions and their three-dimensional null-warped counterparts, corresponding to R symmetry twists. Finally, we study a family of null-warped AdS$_3$ solutions in a setting where there is an ambiguity over the R symmetry and confirm that, for examples admitting a Kaluza-Klein (KK) reduction to three dimensions, the minimisation of a real superpotential of the three-dimensional gauged supergravity captures the central charge and R symmetry.

hep-th↗