arXiv subjects

Seunghun Lee

Publications and source records attributed to Seunghun Lee.

At least 55 records · Page 3Linked to original sources

arXiv2023

NuTrea: Neural Tree Search for Context-guided Multi-hop KGQA

Multi-hop Knowledge Graph Question Answering (KGQA) is a task that involves retrieving nodes from a knowledge graph (KG) to answer natural language questions. Recent GNN-based approaches formulate this task as a KG path searching problem, where messages are sequentially propagated from the seed node towards the answer nodes. However, these messages are past-oriented, and they do not consider the full KG context. To make matters worse, KG nodes often represent proper noun entities and are sometimes encrypted, being uninformative in selecting between paths. To address these problems, we propose Neural Tree Search (NuTrea), a tree search-based GNN model that incorporates the broader KG context. Our model adopts a message-passing scheme that probes the unreached subtree regions to boost the past-oriented embeddings. In addition, we introduce the Relation Frequency-Inverse Entity Frequency (RF-IEF) node embedding that considers the global KG context to better characterize ambiguous KG nodes. The general effectiveness of our approach is demonstrated through experiments on three major multi-hop KGQA benchmark datasets, and our extensive analyses further validate its expressiveness and robustness. Overall, NuTrea provides a powerful means to query the KG with complex natural language questions. Code is available at https://github.com/mlvlab/NuTrea.

cs.CL↗

arXiv2023

Transversals and colorings of simplicial spheres

Motivated from the surrounding property of a point set in $\mathbb{R}^d$ introduced by Holmsen, Pach and Tverberg, we consider the transversal number and chromatic number of a simplicial sphere. As an attempt to give a lower bound for the maximum transversal ratio of simplicial $d$-spheres, we provide two infinite constructions. The first construction gives infintely many $(d+1)$-dimensional simplicial polytopes with the transversal ratio exactly $\frac{2}{d+2}$ for every $d\geq 2$. In the case of $d=2$, this meets the previously well-known upper bound $1/2$ tightly. The second gives infinitely many simplicial 3-spheres with the transversal ratio greater than $1/2$. This was unexpected from what was previously known about the surrounding property. Moreover, we show that, for $d\geq 3$, the facet hypergraph $\mathcal{F}(\mathsf{K})$ of a $d$-dimensional simplicial sphere $\mathsf{K}$ has the chromatic number $χ(\mathcal{F}(\mathsf{K})) \in O(n^{\frac{\lceil d/2\rceil-1}{d}})$, where $n$ is the number of vertices of $\mathsf{K}$. This slightly improves the upper bound previously obtained by Heise, Panagiotou, Pikhurko, and Taraz.

math.CO↗

arXiv2023

Offline-to-Online Knowledge Distillation for Video Instance Segmentation

In this paper, we present offline-to-online knowledge distillation (OOKD) for video instance segmentation (VIS), which transfers a wealth of video knowledge from an offline model to an online model for consistent prediction. Unlike previous methods that having adopting either an online or offline model, our single online model takes advantage of both models by distilling offline knowledge. To transfer knowledge correctly, we propose query filtering and association (QFA), which filters irrelevant queries to exact instances. Our KD with QFA increases the robustness of feature matching by encoding object-centric features from a single frame supplemented by long-range global information. We also propose a simple data augmentation scheme for knowledge distillation in the VIS task that fairly transfers the knowledge of all classes into the online model. Extensive experiments show that our method significantly improves the performance in video instance segmentation, especially for challenging datasets including long, dynamic sequences. Our method also achieves state-of-the-art performance on YTVIS-21, YTVIS-22, and OVIS datasets, with mAP scores of 46.1%, 43.6%, and 31.1%, respectively.

cs.CV↗

arXiv2023

Observation of the superconducting proximity effect from surface states in SmB$_6$/YB$_6$ thin film heterostructures via terahertz spectroscopy

The AC conduction of epitaxially-grown SmB$_6$ thin films and superconducting heterostructures of SmB$_6$/YB$_6$ are investigated via time domain terahertz spectroscopy. A two-channel model of thickness-dependent bulk states and thickness-independent surface states accurately describes the measured conductance of bare SmB$_6$ thin films, demonstrating the presence of surface states in SmB$_6$. While the observed reductions in the simultaneously-measured superconducting gap, transition temperature, and superfluid density of SmB$_6$/YB$_6$ heterostructures relative to bare YB$_6$ indicate the penetration of proximity-induced superconductivity into the SmB$_6$ overlayer; the corresponding SmB$_6$-thickness independence between different heterostructures indicates that the induced superconductivity is predominantly confined to the interface surface state of the SmB$_6$. This study demonstrates the ability of terahertz spectroscopy to probe proximity-induced superconductivity at an interface buried within a heterostructure, and our results show that SmB$_6$ behaves as a predominantly insulating bulk surrounded by conducting surface states in both the normal and induced-superconducting states in both terahertz and DC responses, which is consistent with the topological Kondo insulator picture.

cond-mat.mtrl-sci↗

arXiv2022

Electronic Manipulation of Magnon Topology by Chirality Injection from Boundaries

Magnon bands are known to exhibit nontrivial topology in ordered magnets under suitable conditions, engendering topological phases referred to as magnonic topological insulators. Conventional methods to drive a magnonic topological phase transition are bulk magnetic or thermal operations such as changing the direction of an external magnetic field or varying the temperature of the system, which are undesired in device applications of magnon topology. In this work, we lift the limitation of the magnon topology control on the bulk non-electronic manipulation by proposing a scheme to manipulate magnonic topological phases by electronic boundary operations of spin chirality injection. More specifically, we consider a ferromagnetic honeycomb lattice and show that a finite spin chirality injected from the boundary of the system via the spin Hall effects introduces a tunable sublattice-symmetry-breaking mass term to the bosonic counterpart of the Haldane model for the Chern insulators and thereby allows us to electronically manipulate the bulk topology of magnons from the boundary. The "shoulder" in the thermal Hall conductivity profile is proposed as an experimental probe of the chirality-induced topological phase transition. The scheme for the boundary manipulation of the magnon topology is shown to work for a honeycomb antiferromagnet as well. We envisage that the interfacial chirality injection may offer a nonintrusive electronic means to tune the static and the dynamical bulk properties of general magnetic systems.

cond-mat.mes-hall↗

arXiv2022

Metropolis-Hastings Data Augmentation for Graph Neural Networks

Graph Neural Networks (GNNs) often suffer from weak-generalization due to sparsely labeled data despite their promising results on various graph-based tasks. Data augmentation is a prevalent remedy to improve the generalization ability of models in many domains. However, due to the non-Euclidean nature of data space and the dependencies between samples, designing effective augmentation on graphs is challenging. In this paper, we propose a novel framework Metropolis-Hastings Data Augmentation (MH-Aug) that draws augmented graphs from an explicit target distribution for semi-supervised learning. MH-Aug produces a sequence of augmented graphs from the target distribution enables flexible control of the strength and diversity of augmentation. Since the direct sampling from the complex target distribution is challenging, we adopt the Metropolis-Hastings algorithm to obtain the augmented samples. We also propose a simple and effective semi-supervised learning strategy with generated samples from MH-Aug. Our extensive experiments demonstrate that MH-Aug can generate a sequence of samples according to the target distribution to significantly improve the performance of GNNs.

cs.LG↗

arXiv2022

ADAS: A Direct Adaptation Strategy for Multi-Target Domain Adaptive Semantic Segmentation

In this paper, we present a direct adaptation strategy (ADAS), which aims to directly adapt a single model to multiple target domains in a semantic segmentation task without pretrained domain-specific models. To do so, we design a multi-target domain transfer network (MTDT-Net) that aligns visual attributes across domains by transferring the domain distinctive features through a new target adaptive denormalization (TAD) module. Moreover, we propose a bi-directional adaptive region selection (BARS) that reduces the attribute ambiguity among the class labels by adaptively selecting the regions with consistent feature statistics. We show that our single MTDT-Net can synthesize visually pleasing domain transferred images with complex driving datasets, and BARS effectively filters out the unnecessary region of training images for each target domain. With the collaboration of MTDT-Net and BARS, our ADAS achieves state-of-the-art performance for multi-target domain adaptation (MTDA). To the best of our knowledge, our method is the first MTDA method that directly adapts to multiple domains in semantic segmentation.

cs.CV↗

arXiv2022

Leray numbers of complexes of graphs with bounded matching number

Given a graph $G$ on the vertex set $V$, the non-matching complex of $G$, $\mathsf{NM}_k(G)$, is the family of subgraphs $G' \subset G$ whose matching number $ν(G')$ is strictly less than $k$. As an attempt to generalize the result by Linusson, Shareshian and Welker on the homotopy types of $\mathsf{NM}_k(K_n)$ and $\mathsf{NM}_k(K_{r,s})$ to arbitrary graphs $G$, we show that (i) $\mathsf{NM}_k(G)$ is $(3k-3)$-Leray, and (ii) if $G$ is bipartite, then $\mathsf{NM}_k(G)$ is $(2k-2)$-Leray. This result is obtained by analyzing the homology of the links of non-empty faces of the complex $\mathsf{NM}_k(G)$, which vanishes in all dimensions $d\geq 3k-4$, and all dimensions $d \geq 2k-3$ when $G$ is bipartite. As a corollary, we have the following rainbow matching theorem which generalizes the result by Aharoni, Berger, Chudnovsky, Howard and Seymour: Let $E_1, \dots, E_{3k-2}$ be non-empty edge subsets of a graph and suppose that $ν(E_i\cup E_j)\geq k$ for every $i\ne j$. Then $E=\bigcup E_i$ has a rainbow matching of size $k$. Furthermore, the number of edge sets $E_i$ can be reduced to $2k-1$ when $E$ is the edge set of a bipartite graph.

math.CO↗

arXiv2021

Generation of Nonreciprocity of Gapless Spin Waves by Chirality Injection

In chiral magnets with intrinsic inversion symmetry breaking, it has been known that two spin waves moving in opposite directions can propagate at different velocities, exhibiting a phenomenon called magnetochiral nonreciprocity which allows for realizations of certain spin logic devices such as a spin-wave diode. Here, we theoretically demonstrate that the spin-wave nonreciprocity can occur without intrinsic bulk chirality in certain magnets including easy-cone ferromagnets and easy-cone antiferromagnetic. Specifically, we show that nonlocal injection of a spin current from proximate normal metals to easy-cone magnets engenders a non-equilibrium chiral spin texture, on top of which spin waves exhibit nonreciprocity proportional to the injected spin current. One notable feature of the discovered nonreciprocal spin waves is its gapless nature, which can lead to a large thermal rectification effect at sufficiently low temperatures. We envision that nonlocal electric injection of chirality into otherwise nonchiral magnets may serve as a versatile route to realize electrically controllable magnetochiral phenomena in a wide class of materials.

cond-mat.mes-hall↗

arXiv2021

Graph Transformer Networks: Learning Meta-path Graphs to Improve GNNs

Graph Neural Networks (GNNs) have been widely applied to various fields due to their powerful representations of graph-structured data. Despite the success of GNNs, most existing GNNs are designed to learn node representations on the fixed and homogeneous graphs. The limitations especially become problematic when learning representations on a misspecified graph or a heterogeneous graph that consists of various types of nodes and edges. To address this limitations, we propose Graph Transformer Networks (GTNs) that are capable of generating new graph structures, which preclude noisy connections and include useful connections (e.g., meta-paths) for tasks, while learning effective node representations on the new graphs in an end-to-end fashion. We further propose enhanced version of GTNs, Fast Graph Transformer Networks (FastGTNs), that improve scalability of graph transformations. Compared to GTNs, FastGTNs are 230x faster and use 100x less memory while allowing the identical graph transformations as GTNs. In addition, we extend graph transformations to the semantic proximity of nodes allowing non-local operations beyond meta-paths. Extensive experiments on both homogeneous graphs and heterogeneous graphs show that GTNs and FastGTNs with non-local operations achieve the state-of-the-art performance for node classification tasks. The code is available: https://github.com/seongjunyun/Graph_Transformer_Networks

cs.LG↗

arXiv2021

DRANet: Disentangling Representation and Adaptation Networks for Unsupervised Cross-Domain Adaptation

In this paper, we present DRANet, a network architecture that disentangles image representations and transfers the visual attributes in a latent space for unsupervised cross-domain adaptation. Unlike the existing domain adaptation methods that learn associated features sharing a domain, DRANet preserves the distinctiveness of each domain's characteristics. Our model encodes individual representations of content (scene structure) and style (artistic appearance) from both source and target images. Then, it adapts the domain by incorporating the transferred style factor into the content factor along with learnable weights specified for each domain. This learning framework allows bi-/multi-directional domain adaptation with a single encoder-decoder network and aligns their domain shift. Additionally, we propose a content-adaptive domain transfer module that helps retain scene structure while transferring style. Extensive experiments show our model successfully separates content-style factors and synthesizes visually pleasing domain-transferred images. The proposed method demonstrates state-of-the-art performance on standard digit classification tasks as well as semantic segmentation tasks.

cs.CV↗

arXiv2021

Magnetoelastic Gilbert damping in magnetostrictive Fe$_{0.7}$Ga$_{0.3}$ thin films

We report an enhanced magnetoelastic contribution to the Gilbert damping in highly magnetostrictive Fe$_{0.7}$Ga$_{0.3}$ thin films. This effect is mitigated for perpendicular-to-plane fields, leading to a large anisotropy of the Gilbert damping in all of the films (up to a factor of 10 at room temperature). These claims are supported by broadband measurements of the ferromagnetic resonance linewidths over a range of temperatures (5 to 400 K), which serve to elucidate the effect of both the magnetostriction and phonon relaxation on the magnetoelastic Gilbert damping.

cond-mat.mtrl-sci↗

arXiv2020

The sandpile group of a trinity and a canonical definition for the planar Bernardi action

Baker and Wang define the so-called Bernardi action of the sandpile group of a ribbon graph on the set of its spanning trees. This potentially depends on a fixed vertex of the graph but it is independent of the base vertex if and only if the ribbon structure is planar, moreover, in this case the Bernardi action is compatible with planar duality. Earlier, Chan, Church and Grochow and Chan, Glass, Macauley, Perkinson, Werner and Yang proved analogous results about the rotor-routing action. Baker and Wang moreover showed that the Bernardi and rotor-routing actions coincide for plane graphs. We clarify this still confounding picture by giving a canonical definition for the planar Bernardi/rotor-routing action, and also a canonical isomorphism between sandpile groups of planar dual graphs. Our canonical definition implies the compatibility with planar duality via an extremely short argument. We also show hidden symmetries of the problem by proving our results in the slightly more general setting of balanced plane digraphs. Any balanced plane digraph gives rise to a trinity, i.e., a triangulation of the sphere with a three-coloring of the $0$-simplices. Our most important tool is a group associated to trinities, introduced by Cavenagh and Wanless, and a result of a subset of the authors characterizing the Bernardi bijection in terms of a dissection of a root polytope.

math.CO↗

arXiv2019

Spin gap and L modulated intensity at the low-energy incommensurate magnetic fluctuations in the superconducting state of Sr2RuO4

Low-energy incommensurate (IC) magnetic fluctuations in the multiband superconductor Sr$_2$RuO$_4$ is investigated by high-resolution inelastic neutron scattering measurements and random phase approximation (RPA) calculations. Below $T_\text{c}$, the substantial spin gap is observed at $\mathbf{Q}_\text{IC}=(0.3, 0.3, L)$ where the quasi-one-dimensional $α$ and $β$ sheets consisting of the Fermi surfaces are in good nesting conditions. $L$ modulated intensity of the low-energy IC magnetic fluctuations and our RPA calculations indicate that the superconducting gaps regarding the $α$ and $β$ sheets have the horizontal line nodes.

cond-mat.supr-con↗

arXiv2019

Microwave Meissner Screening of Proximity coupled Topological Insulator / Superconductor Bilayers

The proximity coupled topological insulator / superconductor (TI/SC) bilayer system is a representative system to realize topological superconductivity. In order to better understand this unique state and design devices from the TI/SC bilayer, a comprehensive understanding of the microscopic properties of the bilayer is required. In this work, a microwave Meissner screening study, which exploits a high-precision microwave resonator technique, is conducted on the SmB6/YB6 thin film bilayers as an example TI/SC system. The study reveals spatially dependent electrodynamic screening response of the TI/SC system that is not accessible to other techniques, from which the corresponding microscopic properties of a TI/SC bilayer can be obtained. The TI thickness dependence of the effective penetration depth suggests the existence of a bulk insulating region in the TI layer. The spatially dependent electrodynamic screening model analysis provides an estimate for the characteristic lengths of the TI/SC bilayer: normal penetration depth, normal coherence length, and the thickness of the surface states. We also discuss implications of these characteristic lengths on the design of a vortex Majorana device such as the radius of the vortex core, the energy splitting due to intervortex tunneling, and the minimum thickness required for a device.

cond-mat.supr-con↗

arXiv2019

Observation of perfect Andreev reflection due to Klein paradox in a topological superconducting state

In 1928, P. Dirac proposed a new wave equation to describe relativistic electrons. Shortly afterwards, O. Klein solved a simple potential step problem for the Dirac equation and stumbled upon an apparent paradox - the potential becomes transparent when the height is larger than the electron energy. For massless particles, backscattering is completely forbidden in Klein tunneling, leading to perfect transmission through any potential barrier. Recent advent of condensed matter systems with Dirac-like excitations, such as graphene and topological insulators (TIs), has opened the possibility of observing the Klein tunneling experimentally. In the surface states of TIs, fermions are bound by spin-momentum locking, and are thus immune to backscattering due to time-reversal symmetry. Here we report the observation of perfect Andreev reflection in point contact spectroscopy - a clear signature of Klein tunneling and a manifestation of the underlying relativistic physics of a proximity-induced superconducting state in a topological Kondo insulator.

cond-mat.mes-hall↗

arXiv2018

On a conjecture of Karasev

Karasev conjectured that for any set of $3k$ lines in general position in the plane, which is partitioned into $3$ color classes of equal size $k$, the set can be partitioned into $k$ colorful 3-subsets such that all the triangles formed by the subsets have a point in common. Although the general conjecture is false, we show that Karasev's conjecture is true for lines in convex position. We also discuss possible generalizations of this result.

math.CO↗

arXiv2018

Nerves, minors, and piercing numbers

We make the first step towards a "nerve theorem" for graphs. Let $G$ be a simple graph and let $\mathcal{F}$ be a family of induced subgraphs of $G$ such that the intersection of any members of $\mathcal{F}$ is either empty or connected. We show that if the nerve complex of $\mathcal{F}$ has non-vanishing homology in dimension three, then $G$ contains the complete graph on five vertices as a minor. As a consequence we confirm a conjecture of Goaoc concerning an extension of the planar $(p,q)$ theorem due to Alon and Kleitman: Let $\mathcal{F}$ be a finite family of open connected sets in the plane such that the intersection of any members of $\mathcal{F}$ is either empty or connected. If among any $p \geq 3$ members of $\mathcal{F}$ there are some three that intersect, then there is a set of $C$ points which intersects every member of $\mathcal{F}$, where $C$ is a constant depending only on $p$.

math.CO↗