May 21

16:00-16:20

Title Non-equilibrium steady state of quantum-mechanically coupled two-qubit system
Speaker Kwang Hyun Cho
Abstract The minimal bound of the thermodynamic uncertainty relation (TUR) is modulated from that of the classical counterpart when a quantumness is present in dynamical processes far from equilibrium. A recent study on a model of open quantum systems, specifically, an external field-driven transition dynamics of dissipative two-level system (TLS), indicates that the quantum coherence between ground and excited states can suppress the fluctuations of the irreversible current and loosens the TUR bound. Here, we extend on the previous study of field-driven single TLS (qubit) in a photonic bath to a quantum-mechanically coupled two-qubit system (TQS), and explore, in addition to the coherence, how the quantum coupling and entanglement between the two qubits affect the photon current and fluctuations and contribute to loosening or tightening the TUR bound.

16:20-16:40

Title Symphony in a Cell: Letting Bacteria Take the Stage
Speaker YeongKyu Lee
Abstract Inside each of us, a hidden orchestra plays — a biological clock that keeps time, guiding when we should be active and when we should rest. But this internal rhythm isn’t just a human phenomenon. From animals to plants, and even to microscopic life, biological rhythms are a universal feature of life. Among the most fascinating performers of this cellular symphony are cyanobacteria — simple, single-celled organisms that keep astonishingly accurate 24-hour rhythms. Despite their simplicity, these bacteria “play” a molecular symphony using just three proteins: KaiA, KaiB, and KaiC. Together, they orchestrate a self-sustained rhythm that echoes the day-night cycle of our planet. In this talk, we’ll explore how these proteins create a clock-like rhythm, and why timing is such a fundamental part of life.

17:00-17:20

Title Generalized parity-oblivious communication games: from foundational underpinnings to its applications in quantum information theory
Speaker Prabuddha Roy
Abstract The parity-oblivious random-access-code (PORAC) is a class of communication games involving a sender (Alice) and a receiver (Bob). In such games, Alice's amount of communication to Bob is constraint by the parity-oblivious (PO) conditions, so that the parity information of her inputs remains oblivious to Bob. In this talk, I will demonstrate that the quantum theory outperforms the classical model by predicting higher winning probability in our generalized PORAC thereby providing advantage in quantum information theoretic tasks.

17:20-17:40

Title Concurrence speed limit and its connection with bounds in many body physics
Speaker Shrobona Bagchi
Abstract Abstract: Quantum speed limit is a fundamental speed limit for the evolution of quantum states. It is the single-most important interpretation of the time energy uncertainty relation. Recently the speed limit of quantum correlations have been proposed like the concurrence for pure quantum states. In this direction, we derive a speed limit bound for a quantum correlation named the concurrence for the generally mixed quantum states of two qubits. By this we mean that we find an expression for the minimum time required to reach a given value of entanglement starting from an arbitrary initial generally mixed state. We discuss the connection of the findings of this article in the interdisciplinary area of the condensed matter physics or the many body physics and quantum information science such as on the topic of Lieb-Robinson bound in a quantitative manner

May 22

09:00-09:20

Title Rainbow problems with oriented matroids
Speaker Minho Cho
Abstract We list some rainbow problems on oriented matroids and other related mathematical objects. As a partial result, we obtain an oriented matroid version of Barany's colorful conic Caratheodory theorem. The key lemma is a common generalization of Sperner's theorem and Meshulam's lemma, each of which guarantees the existence of rainbow simplexes in vertex-colored simplicial complexes under different assumptions.

09:20-09:40

Title Over-smoothing vs. Over-squashing in Graph Neural Networks
Speaker Hyeyun Yang
Abstract I will introduce Graph Neural Networks (GNNs) and two problems which GNNs suffer, over-smoothing and over-squashing. I will briefly share a hypothesis about the relationship between these problems -“which one is the bottleneck”- by using curvature-based rewiring method and present the verification results.

09:40-10:00

Title Adaptively Secure Fully Homomorphic Message Authentication Code with Pre-processable Verification
Speaker Jeongsu Kim
Abstract There has been remarkable progress in fully homomorphic encryption, ever since Gentry's first scheme.  In contrast, fully homomorphic authentication primitives received relatively less attention, despite existence of some previous constructions. While there exist various schemes with different functionalities for fully homomorphic encryption, there are only a few options for fully homomorphic authentication. Moreover, there are even fewer options when considering two of the most important properties: adaptive security, and pre-processable verification. To our knowledge, except for some concurrent works, achieving both properties requires the use of nested construction, which involves homomorphically authenticating a homomorphic authentication tag of a message, making the scheme costly and complicated.

In this work, we propose a dedicated scheme for (leveled) fully homomorphic message authentication code that is adaptively secure and has pre-processable verification. Leveraging the secrecy of the primitive, we demonstrate that a slight modification of a selectively secure (leveled) fully homomorphic signature scheme yields an adaptively secure (leveled) fully homomorphic message authentication code with pre-processable verification. Additionally, we introduce a novel notion and generic transform to enhance the security of a homomorphic message authentication code, which also exploits the secrecy of the primitive.

10:00-10:20

Title How Can We Ensure Privacy of Training Data in Deep Learning Models?
Speaker Jinseong Park
Abstract Deep learning models are known to pose a risk of privacy leakage from training data samples. To safeguard against potential data exposure, various methods, such as anonymization and encryption, have been proposed. Among them, differential privacy (DP) offers a mathematical guarantee against adversaries with practical implementations in training neural networks through gradient modifications. However, training deep learning models with DP may lead to a degradation in prediction performance compared to models without DP. In this talk, we will review recent advancements in privacy-preserving deep learning models, particularly focusing on the recent evolution of generative models. We will end with a discussion on promising future directions in the field.

11:00-11:20

Title Hot carrier diffusion-assisted ideal carrier multiplication in monolayer MoSe2
Speaker Hongguk Min
Abstract One of the challenges in improving solar cell performance is how to make the most out of each photon of sunlight. A process called carrier multiplication (CM) offers a promising solution: it allows a single photon to generate multiple charge carriers, which leads to higher energy conversion efficiency. In this work, ultrafast optical measurements reveal that monolayer MoSe₂, an atomically thin two-dimensional semiconductor, can achieve a highly efficient CM process. We use computational modeling to uncover the physical mechanisms behind this process. Starting from the electronic structure of the material obtained through first-principles DFT calculations, we identify multiple efficient pathways that enable CM. Our results position monolayer MoSe₂ as an excellent candidate for studying hot carrier dynamics and developing next-generation optoelectronic devices.

11:20-11:40

Title Electronic structures of crystalline and amorphous GeSe and GeSbTe compounds using machine learning empirical pseudopotentials
Speaker Sungmo Kang
Abstract The newly developed machine learning (ML) empirical pseudopotential (EP) method overcomes the poor transferability of the traditional EP method with the help of ML techniques while preserving its formal simplicity and computational efficiency. We apply the new method to binary and ternary systems such as GeSe and Ge-Sb-Te (GST) compounds, well-known materials for non-volatile phase-change memory and related technologies. Using a training set of {it ab initio} electronic energy bands and rotation-covariant descriptors for various GeSe and GST compounds, we generate transferable EPs for Ge, Se, Sb, and Te. We demonstrate that the new ML model accurately reproduces the energy bands and wavefunctions of structures outside the training set, closely matching first-principles calculations. This accuracy is achieved with significantly lower computational costs due to the elimination of self-consistency iterations and the reduced size of the plane-wave basis set. Notably, the method maintains accuracy even for diverse local atomic environments, such as amorphous phases or larger systems not explicitly included in the training set.

11:40-12:00

Title Self-interaction error correction using ab initio extended Hubbard interactions
Speaker Wooil Yang
Abstract Density functional theory (DFT) is a widely used first-principles method mapping the many-body problem onto an effective single-particle framework. However, standard DFT approximations suffer from inherent self-interaction errors (SIE), leading to significant failures in describing certain systems, particularly those with localized electrons. To mitigate SIE, Hubbard corrections for on-site (U) and inter-site (V) interactions (DFT+U+V) have been introduced. Utilizing U and V parameters determined self-consistently via ab initio methods, rather than empirical values, has demonstrated improved agreement with higher-level theories like GW approximation and experiments. In this work, we further validate the robustness and broad applicability of our ab initio DFT+U+V methodology across diverse material conditions and interaction strengths. Notably, our approach offers significant computational advantages, potentially reducing computational cost by more than an order of magnitude compared to certain alternative correlated electron methods, rendering it suitable for high-throughput screening and data-driven discovery. We demonstrate its effectiveness for large-scale and low-dimensional systems through calculations on an eight-quintuple-layer Bi2Se3 structure. This work thus offers an efficient and robust framework for studying large-scale correlated materials, including those with pronounced relativistic effects.

May 23

09:20-09:40

Title Optimization of Adiabatic Evolution Schedules Using the Quantum Zeno Approach 
Speaker Mancheon Han
Abstract The adiabatic theorem, originally developed by Born and Fock in 1928, asserts that a quantum system remains in its instantaneous eigenstate if the perturbation acting upon it varies slowly enough. This principle underpins many modern quantum theories and applications, including the definition of distinct quantum phases and adiabatic quantum computation. While a gradual transformation of the Hamiltonian—from an initial easily prepared one to a complex target Hamiltonian—ensures eventual convergence, the performance of such adiabatic evolutions critically depends on the choice of the schedule function. We address the challenge of determining an optimal adiabatic schedule, which is inherently difficult due to the infinite number of promising schedules and the dependence on system-specific details like energy gaps that are typically unknown beforehand. We propose a method that leverages the Quantum Zeno effect to autonomously identify a constant speed schedule, optimizing an upper bound on the adiabatic infidelity. Notably, our procedure does not require detailed a priori knowledge of the system. We validate our method through numerical simulations on various quantum systems, demonstrating that it reliably discovers the optimal schedule and substantially reduces the total evolution time, especially in complex scenarios.

09:40-10:00

Title Quantum computing methods for ground state problem and image processing
Speaker Taehee Ko
Abstract Despite the promise of quantum computing, we are still in early stages where practical applications are yet readily available. Nevertheless, quantum computing is believed to provide computational advantages that are hard to be achieved using only classical resources. In this talk, we discuss well-known computational problems: finding an eigenvector corresponding to the smallest eigenvalue of a Hamiltonian matrix and computing image gradients of a pixel-based image. We present relevant quantum computing methods and show that the methods have computational advantages in some situations compared to the classical counterparts.

10:30-10:50

Title Effective Field Theory Analysis of Neural Network Ensembles and Neural Operators
Speaker Taeyoung Kim
Abstract In this presentation, we analyze the critical phenomena of ensembles of neural networks in their initialized state from the perspective of effective field theory. Furthermore, we briefly explore the case of applying this approach to neural operators.

10:50-11:10

Title Jibeom Choi
Speaker What do ants and cancers have in common?
Abstract I am an evolutionary biologist who majored in mathematics. During graduate school, I delved into the collective movements of ants and the general principles of evolution. Over time, I came to realize that the social structures of eusocial ant colonies and cancer clusters share diverse features such as monopoly of reproduction and cooperation among group members. This insight led me to investigate the dynamics of cancer through the lens of social evolution, longing for the development of transformative strategies to cure cancer. In this talk, I will introduce the motivations behind my research and share related publications.

11:10-11:30

Title On the difficulty of training autoregressive models on long time series data
Speaker Joon-Hyuk Ko
Abstract From parameterized differential equation models to deep learning architectures such as recurrent neural networks and neural ordinary differential equations, autoregressive models are ubiquitous in time-series forecasting. As these models can be trained on sequences of arbitrary length then deployed to produce long term predictions, superficially, there is a lot of flexibility in training these models. In reality, however, training is very sensitive to the length of the training data, with shorter data resulting in nearsighted predictions and long data causing exploding gradients.

In the first half of this talk, I will provide an overview of the unstable training problem of these models. The second half will then be devoted to how dynamical systems theory can be utilized to effectively train autoregressive models on long time series data while avoiding training instabilities.