arXiv subjects

Junhyun Lee

Publications and source records attributed to Junhyun Lee.

At least 37 records · Page 2Linked to original sources

arXiv2024

MolPLA: A Molecular Pretraining Framework for Learning Cores, R-Groups and their Linker Joints

Molecular core structures and R-groups are essential concepts in drug development. Integration of these concepts with conventional graph pre-training approaches can promote deeper understanding in molecules. We propose MolPLA, a novel pre-training framework that employs masked graph contrastive learning in understanding the underlying decomposable parts inmolecules that implicate their core structure and peripheral R-groups. Furthermore, we formulate an additional framework that grants MolPLA the ability to help chemists find replaceable R-groups in lead optimization scenarios. Experimental results on molecular property prediction show that MolPLA exhibits predictability comparable to current state-of-the-art models. Qualitative analysis implicate that MolPLA is capable of distinguishing core and R-group sub-structures, identifying decomposable regions in molecules and contributing to lead optimization scenarios by rationally suggesting R-group replacements given various query core templates. The code implementation for MolPLA and its pre-trained model checkpoint is available at https://github.com/dmis-lab/MolPLA

cs.LG↗

arXiv2023

Spin fractionalization and zero modes in the spin-$\frac{1}{2}$ XXZ chain with boundary fields

In this work we argue that the antiferromagnetic spin $\frac{1}{2}$ XXZ chain in the gapped phase with boundary magnetic fields hosts fractional spin $\frac{1}{4}$ at its edges. Using a combination of Bethe ansatz and the density matrix renormalization group we show that these fractional spins are sharp quantum observables in both the ground and the first excited state as the associated fractional spin operators have zero variance. In the limit of zero edge fields, we argue that these fractional spin operators once projected onto the low energy subspace spanned by the ground state and the first excited state, identify with the strong zero energy mode discovered by P. Fendley \cite{Fendley}.

cond-mat.str-el↗

arXiv2023

Co-attention Graph Pooling for Efficient Pairwise Graph Interaction Learning

Graph Neural Networks (GNNs) have proven to be effective in processing and learning from graph-structured data. However, previous works mainly focused on understanding single graph inputs while many real-world applications require pair-wise analysis for graph-structured data (e.g., scene graph matching, code searching, and drug-drug interaction prediction). To this end, recent works have shifted their focus to learning the interaction between pairs of graphs. Despite their improved performance, these works were still limited in that the interactions were considered at the node-level, resulting in high computational costs and suboptimal performance. To address this issue, we propose a novel and efficient graph-level approach for extracting interaction representations using co-attention in graph pooling. Our method, Co-Attention Graph Pooling (CAGPool), exhibits competitive performance relative to existing methods in both classification and regression tasks using real-world datasets, while maintaining lower computational complexity.

cs.LG↗

arXiv2023

Towards Flexible Time-to-event Modeling: Optimizing Neural Networks via Rank Regression

Time-to-event analysis, also known as survival analysis, aims to predict the time of occurrence of an event, given a set of features. One of the major challenges in this area is dealing with censored data, which can make learning algorithms more complex. Traditional methods such as Cox's proportional hazards model and the accelerated failure time (AFT) model have been popular in this field, but they often require assumptions such as proportional hazards and linearity. In particular, the AFT models often require pre-specified parametric distributional assumptions. To improve predictive performance and alleviate strict assumptions, there have been many deep learning approaches for hazard-based models in recent years. However, representation learning for AFT has not been widely explored in the neural network literature, despite its simplicity and interpretability in comparison to hazard-focused methods. In this work, we introduce the Deep AFT Rank-regression model for Time-to-event prediction (DART). This model uses an objective function based on Gehan's rank statistic, which is efficient and reliable for representation learning. On top of eliminating the requirement to establish a baseline event time distribution, DART retains the advantages of directly predicting event time in standard AFT models. The proposed method is a semiparametric approach to AFT modeling that does not impose any distributional assumptions on the survival time distribution. This also eliminates the need for additional hyperparameters or complex model architectures, unlike existing neural network-based AFT models. Through quantitative analysis on various benchmark datasets, we have shown that DART has significant potential for modeling high-throughput censored time-to-event data.

cs.LG↗

arXiv2023

Non-equilibrium steady state phases of the interacting Aubry-Andre-Harper model

Here we study the phase diagram of the Aubry-Andre-Harper model in the presence of strong interactions as the strength of the quasiperiodic potential is varied. Previous work has established the existence of many-body localized phase at large potential strength; here, we find a rich phase diagram in the delocalized regime characterized by spin transport and unusual correlations. We calculate the non-equilibrium steady states of a boundary-driven strongly interacting Aubry-Andre-Harper model by employing the time-evolving block decimation algorithm on matrix product density operators. From these steady states, we extract spin transport as a function of system size and quasiperiodic potential strength. This data shows spin transport going from superdiffusive to subdiffusive well before the localization transition; comparing to previous results, we also find that the transport transition is distinct from a transition observed in the speed of operator growth in the model. We also investigate the correlation structure of the steady state and find an unusual oscillation pattern for intermediate values of the potential strength. The unusual spin transport and quantum correlation structure suggest multiple dynamical phases between the much-studied thermal and many-body-localized phases.

cond-mat.str-el↗

arXiv2022

Boundary Quantum Phase Transitions in the Spin $\frac{1}{2}$ Heisenberg Chain with Boundary Magnetic Fields

We consider the spin $\frac{1}{2}$ Heisenberg chain with boundary magnetic fields and analyze it using a combination of Bethe ansatz and density matrix renormalization group (DMRG) techniques. We show that the system exhibits several different ground states which depend on the orientation of the boundary magnetic fields. When both the boundary fields take equal values greater than a critical field strength, each edge in the ground state accumulates a fractional spin which saturates to spin $\frac{1}{4}$, which is similar to systems exhibiting symmetry protected topological phases (SPT). Unlike in SPT systems, the fractional boundary spin in the Heisenberg spin chain is not a genuine quantum number since the variance of the associated operator does not vanish, this is due to the absence of a bulk gap. The system exhibits high energy bound states when the boundary fields take values greater than the critical field. All the excitations in the system can be sorted out into towers whose number depends on the number of bound states exhibited by the system. As the boundary fields are varied, in addition to the ground state phase transition, we find that the system may undergo an eigenstate phase transition (EPT) where the number of towers of the Hilbert space changes. We further inquire how the EPT reflects itself on local ground state properties by computing the magnetization profile $\langle S^z_j \rangle$ using DMRG. We identify a clear qualitative change from low edge fields to high edge fields when crossing the critical field. We though are unable to conclude on the basis of our data that EPT corresponds to a genuine phase transition in the ground state.

cond-mat.str-el↗

arXiv2022

Rydberg dressed spin-1/2 Fermi gases in one dimension

The emergent phases of strongly correlated spin-1/2 Fermi gases of Rydberg dressed atoms in a one dimensional optical lattice are theoretically investigated. At weak coupling a bosonization description is used to demonstrate the ability to drive alternating quantum phase transitions between distinct Luttinger liquids. At strong coupling the ground state develops non-trivial phase separation exhibiting Luttinger liquid ''puddles'' separated by magnetic domain walls due to the interplay of the incommensurate filling and the Rydberg core length scale. These phases can be detected in ultracold gases of Rydberg atoms made from $^6$Li.

cond-mat.quant-gas↗

arXiv2022

Neo-GNNs: Neighborhood Overlap-aware Graph Neural Networks for Link Prediction

Graph Neural Networks (GNNs) have been widely applied to various fields for learning over graph-structured data. They have shown significant improvements over traditional heuristic methods in various tasks such as node classification and graph classification. However, since GNNs heavily rely on smoothed node features rather than graph structure, they often show poor performance than simple heuristic methods in link prediction where the structural information, e.g., overlapped neighborhoods, degrees, and shortest paths, is crucial. To address this limitation, we propose Neighborhood Overlap-aware Graph Neural Networks (Neo-GNNs) that learn useful structural features from an adjacency matrix and estimate overlapped neighborhoods for link prediction. Our Neo-GNNs generalize neighborhood overlap-based heuristic methods and handle overlapped multi-hop neighborhoods. Our extensive experiments on Open Graph Benchmark datasets (OGB) demonstrate that Neo-GNNs consistently achieve state-of-the-art performance in link prediction. Our code is publicly available at https://github.com/seongjunyun/Neo_GNNs.

cs.LG↗

arXiv2022

MSTR: Multi-Scale Transformer for End-to-End Human-Object Interaction Detection

Human-Object Interaction (HOI) detection is the task of identifying a set of triplets from an image. Recent work proposed transformer encoder-decoder architectures that successfully eliminated the need for many hand-designed components in HOI detection through end-to-end training. However, they are limited to single-scale feature resolution, providing suboptimal performance in scenes containing humans, objects and their interactions with vastly different scales and distances. To tackle this problem, we propose a Multi-Scale TRansformer (MSTR) for HOI detection powered by two novel HOI-aware deformable attention modules called Dual-Entity attention and Entity-conditioned Context attention. While existing deformable attention comes at a huge cost in HOI detection performance, our proposed attention modules of MSTR learn to effectively attend to sampling points that are essential to identify interactions. In experiments, we achieve the new state-of-the-art performance on two HOI detection benchmarks.

cs.CV↗

arXiv2021

Emulating twisted double bilayer graphene with a multiorbital optical lattice

This work theoretically explores how to emulate twisted double bilayer graphene with ultracold atoms in multiorbital optical lattices. In particular, the quadratic band touching of Bernal stacked bilayer graphene is emulated using a square optical lattice with $p_x$, $p_y$, and $d_{x^2-y^2}$ orbitals on each site, while the effects of a twist are captured through the application of an incommensurate potential. The quadratic band touching is stable until the system undergoes an Anderson like delocalization transition in momentum space, which occurs concomitantly with a strongly renormalized single particle spectrum inducing flat bands, which is a generalization of the magic-angle condition realized in Dirac semimetals. The band structure is described perturbatively in the quasiperiodic potential strength, which captures miniband formation and the existence of magic-angles that qualitatively agrees with the exact numerical results in the appropriate regime. We identify several magic-angle conditions that can either have part or all of the quadratic band touching point become flat. In each case, these are accompanied by a diverging density of states and the delocalization of plane wave eigenstates. It is discussed how these transitions and phases can be observed in ultracold atom experiments.

cond-mat.quant-gas↗

arXiv2021

Anomalous Quantum Oscillations in Spin-3/2 Topological Semimetal YPtBi

The proposed high-spin superconductivity in the half-Heusler compounds changes the landscape of superconductivity research. While superconducting instability is possible only in systems with quantum mechanically coherent quasiparticles, it has not been verified for any proposed high-spin Fermi surfaces. Here we report an observation of anomalous Shubnikov-de Haas effect in half-Heusler YPtBi, which is compatible with a coherent $j=3/2$ Fermi surface. The quantum oscillation (QO) signal in cubic YPtBi manifests extreme anisotropy upon rotation of the magnetic field from [100] to [110] crystallographic direction where the QO signal drastically vanishes near [110]. This radical anisotropy for a cubic system cannot be explained by trivial scenarios for QO involving effective mass or impurity scattering, but it is naturally explained by the warping feature of the $j=3/2$ Fermi surface YPtBi. Our results prove the high-spin nature of the quasiparticle in the half-Heusler compounds, which makes the realization of the unprecedented high-spin superconductivity more plausible.

cond-mat.str-el↗

arXiv2021

HOTR: End-to-End Human-Object Interaction Detection with Transformers

Human-Object Interaction (HOI) detection is a task of identifying "a set of interactions" in an image, which involves the i) localization of the subject (i.e., humans) and target (i.e., objects) of interaction, and ii) the classification of the interaction labels. Most existing methods have indirectly addressed this task by detecting human and object instances and individually inferring every pair of the detected instances. In this paper, we present a novel framework, referred to by HOTR, which directly predicts a set of triplets from an image based on a transformer encoder-decoder architecture. Through the set prediction, our method effectively exploits the inherent semantic relationships in an image and does not require time-consuming post-processing which is the main bottleneck of existing methods. Our proposed algorithm achieves the state-of-the-art performance in two HOI detection benchmarks with an inference time under 1 ms after object detection.

cs.CV↗

arXiv2020

Robust Long-Term Object Tracking via Improved Discriminative Model Prediction

We propose an improved discriminative model prediction method for robust long-term tracking based on a pre-trained short-term tracker. The baseline pre-trained short-term tracker is SuperDiMP which combines the bounding-box regressor of PrDiMP with the standard DiMP classifier. Our tracker RLT-DiMP improves SuperDiMP in the following three aspects: (1) Uncertainty reduction using random erasing: To make our model robust, we exploit an agreement from multiple images after erasing random small rectangular areas as a certainty. And then, we correct the tracking state of our model accordingly. (2) Random search with spatio-temporal constraints: we propose a robust random search method with a score penalty applied to prevent the problem of sudden detection at a distance. (3) Background augmentation for more discriminative feature learning: We augment various backgrounds that are not included in the search area to train a more robust model in the background clutter. In experiments on the VOT-LT2020 benchmark dataset, the proposed method achieves comparable performance to the state-of-the-art long-term trackers. The source code is available at: https://github.com/bismex/RLT-DIMP.

cs.CV↗

arXiv2020

Wess-Zumino-Witten Terms in Graphene Landau Levels

We consider the interplay between the antiferromagnetic and Kekulé valence bond solid orderings in the zero energy Landau levels of neutral monolayer and bilayer graphene. We establish the presence of Wess-Zumino-Witten terms between these orders: this implies that their quantum fluctuations are described by the deconfined critical theories of quantum spin systems. We present implications for experiments, including the possible presence of excitonic superfluidity in bilayer graphene.

cond-mat.str-el↗

arXiv2019

Strongly interacting spin-orbit coupled Bose-Einstein condensates in one dimension

We theoretically study dilute superfluidity of spin-1 bosons with antiferromagnetic interactions and synthetic spin-orbit coupling (SOC) in a one-dimensional lattice. Employing a combination of density matrix renormalization group and quantum field theoretical techniques we demonstrate the appearance of a robust superfluid spin-liquid phase in which the spin-sector of this spinor Bose-Einstein condensate remains quantum disordered even after introducing quadratic Zeeman and helical magnetic fields. Despite remaining disordered, the presence of these symmetry breaking fields lifts the perfect spin-charge separation and thus the nematic correlators obey power-law behavior. We demonstrate that, at strong coupling, the SOC induces a charge density wave state that is not accessible in the presence of linear and quadratic Zeeman fields alone. In addition, the SOC induces oscillations in the spin and nematic expectation values as well as the bosonic Green's function. These non-trivial effects of a SOC are suppressed under the application of a large quadratic Zeeman field. We discuss how our results could be observed in experiments on ultracold gases of $^{23}$Na in an optical lattice.

cond-mat.quant-gas↗

arXiv2019

Lattice vibration as a knob for novel quantum criticality: Emergence of supersymmetry from spin-lattice coupling

Control of quantum coherence in many-body system is one of the key issues in modern condensed matter. Conventional wisdom is that lattice vibration is an innate source of decoherence, and amounts of research have been conducted to eliminate lattice effects. Challenging this wisdom, here we show that lattice vibration may not be a decoherence source but an impetus of a novel coherent quantum many-body state. We demonstrate the possibility by studying the transverse-field Ising model on a chain with renormalization group and density-matrix renormalization group method, and theoretically discover a stable $\mathcal{N}=1$ supersymmetric quantum criticality with central charge $c=3/2$. Thus, we propose an Ising spin chain with strong spin-lattice coupling as a candidate to observe supersymmetry. Generic precursor conditions of novel quantum criticality are obtained by generalizing the Larkin-Pikin criterion of thermal transitions. Our work provides a new perspective that lattice vibration may be a knob for exotic quantum many-body states.

cond-mat.str-el↗

arXiv2019

Self-Attention Graph Pooling

Advanced methods of applying deep learning to structured data such as graphs have been proposed in recent years. In particular, studies have focused on generalizing convolutional neural networks to graph data, which includes redefining the convolution and the downsampling (pooling) operations for graphs. The method of generalizing the convolution operation to graphs has been proven to improve performance and is widely used. However, the method of applying downsampling to graphs is still difficult to perform and has room for improvement. In this paper, we propose a graph pooling method based on self-attention. Self-attention using graph convolution allows our pooling method to consider both node features and graph topology. To ensure a fair comparison, the same training procedures and model architectures were used for the existing pooling methods and our method. The experimental results demonstrate that our method achieves superior graph classification performance on the benchmark datasets using a reasonable number of parameters.

cs.LG↗

arXiv2019

Chiral anomaly without Landau levels: From the quantum to the classical regime

We study the chiral anomaly in disordered Weyl semimetals, where the broken translational symmetry prevents the direct application of Nielsen and Ninomiya's mechanism and disorder is strong enough that quantum effects are important. In the weak disorder regime, there exist rare regions of the random potential where the disorder strength is locally strong, which gives rise to quasilocalized resonances and their effect on the chiral anomaly is unknown. We numerically show that these resonant states do not affect the chiral anomaly only in the case of a single Weyl node. At energies away from the Weyl point, or with strong disorder where one is deep in the diffusive regime, the chiral Landau level itself is not well defined and the semiclassical treatment is not justified. In this limit, we analytically use the supersymmetry method and find that the Chern-Simons term in the effective action which is not present in nontopological systems gives rise to a nonzero average level velocity which implies chiral charge pumping. We numerically establish that the nonzero average level velocity serves as an indicator of the chiral anomaly in the diffusive limit.

cond-mat.dis-nn↗