arXiv subjects

Jinuk Kim

Publications and source records attributed to Jinuk Kim.

18 recordsLinked to original sources

arXiv2026

Rule2DRC: Benchmarking LLM Agents for DRC Script Synthesis with Execution-Guided Test Generation

Manufacturable chip layouts must satisfy thousands of geometry-based design rules, and design rule checking (DRC) enforces them by running executable DRC scripts on layouts. Translating natural language rules into correct DRC scripts is labor-intensive and requires specialized expertise, motivating LLM agents for DRC script synthesis and debugging. However, existing benchmarks have small evaluation sets and often evaluate scripts by code similarity rather than execution correctness, and prior machine learning-based methods either ignore execution feedback or require labeled test layouts as agent's input. To this end, we introduce Rule2DRC, a large-scale benchmark for DRC script coding agents with 1,000 rule-to-script tasks and 13,921 evaluation chip layouts for execution-based scoring. Rule2DRC provides an evaluation pipeline that measures functional correctness via DRC execution outcomes without requiring evaluation layouts as input to the agent. We also propose SplitTester, a tester agent for program selection that uses execution feedback to generate discriminative test cases and separate previously indistinguishable candidate scripts, substantially improving Best-of-N selection performance in this domain. We release the code at https://github.com/snu-mllab/Rule2DRC.

cs.LG↗

arXiv2025

In-situ tuning of cavity dissipation and a topological transition in an atom-nanotip-cavity system

We theoretically investigate a method for tuning a dissipation rate in an atom-cavity system. By inserting a nanotip into the cavity mode, we estimate that the cavity dissipation rate increases by a factor of approximately 20, due to the scattering loss at the tip. As applications of our in-situ technique, we demonstrate that an exceptional line can be obtained when a single atom or ion is coupled to the resonator. Moreover, the position of an exceptional point is tuned by adjusting the decay rate, enabling a topological transition through dissipation control alone.

quant-ph↗

arXiv2025

KVzip: Query-Agnostic KV Cache Compression with Context Reconstruction

Transformer-based large language models (LLMs) cache context as key-value (KV) pairs during inference. As context length grows, KV cache sizes expand, leading to substantial memory overhead and increased attention latency. This paper introduces KVzip, a query-agnostic KV cache eviction method enabling effective reuse of compressed KV caches across diverse queries. KVzip quantifies the importance of a KV pair using the underlying LLM to reconstruct original contexts from cached KV pairs, subsequently evicting pairs with lower importance. Extensive empirical evaluations demonstrate that KVzip reduces KV cache size by $3$-$4\times$ and FlashAttention decoding latency by approximately $2\times$, with negligible performance loss in question-answering, retrieval, reasoning, and code comprehension tasks. Evaluations include various models such as LLaMA3.1, Qwen2.5, and Gemma3, with context lengths reaching up to 170K tokens. KVzip significantly outperforms existing query-aware KV eviction methods, which suffer from performance degradation even at a 90% cache budget ratio under multi-query scenarios.

cs.DB↗

arXiv2025

GuidedQuant: Large Language Model Quantization via Exploiting End Loss Guidance

Post-training quantization is a key technique for reducing the memory and inference latency of large language models by quantizing weights and activations without requiring retraining. However, existing methods either (1) fail to account for the varying importance of hidden features to the end loss or, when incorporating end loss, (2) neglect the critical interactions between model weights. To address these limitations, we propose GuidedQuant, a novel quantization approach that integrates gradient information from the end loss into the quantization objective while preserving cross-weight dependencies within output channels. GuidedQuant consistently boosts the performance of state-of-the-art quantization methods across weight-only scalar, weight-only vector, and weight-and-activation quantization. Additionally, we introduce a novel non-uniform scalar quantization algorithm, which is guaranteed to monotonically decrease the quantization objective value, and outperforms existing methods in this category. We release the code at https://github.com/snu-mllab/GuidedQuant.

cs.LG↗

arXiv2024

Topological Switching via Exceptional Point Pairs in an Optical Microcavity Laser

Conventional mode switching mechanisms, which rely on dynamically encircling exceptional points (EPs) through non-adiabatic transitions (NATs), suffer from intrinsic nonlinear dynamics that hinder precise control and reproducibility in experimental settings. Additionally, these methods exhibit low transmission efficiencies due to path-dependent attenuation, limiting their effectiveness in optical switching and logic gate applications. To overcome these limitations, we propose a novel mode switching approach that leverages a pair of EPs configuration in an optical microcavity, characterized by superradiance and subradiance. This mechanism exploits the topological structure of the Riemann surface to enable robust mode switching control and tunable Q-factor through purely adiabatic encircling. Furthermore, topological protection validated via braid isotopy ensures robustness against noise and parametric perturbations, facilitating a compact, robust, and adaptive non-Hermitian system.

physics.optics↗

arXiv2024

Numerical analysis of a superradiance-sideband-assisted laser with a zero frequency pulling and a narrow linewidth

Numerical simulations based on the quantum Langevin equations have been performed for a large number of two-level atoms in a beam interacting with a low-Q cavity with the atomic initial superposition states close to the north pole of the Bloch sphere. When the pump Rabi frequency was modulated at $Δ_{pa}$ with zero pump-atom detuning for various cavity-atom detunings, we obtained a lasing peak at the atomic resonance and superradiant lasing peaks at $\pmΔ_{pa}$ simultaneously while the central peak exhibiting a zero frequency pulling coefficient. The linewidth of the central peak was reduced beyond the gain narrowing as the mean number of atoms was increased, resulting in a minimum linewidth as small as a millionth of the atomic or cavity linewdith. A pump carrier detuning caused asymmetric heights for the side superradiant peaks, the height difference of which can be used to lock the pump laser to the atom within the linewidth of the central lasing peak. Our results may lead to development of a new type of ultra-stable active optical clocks for future frequency standards when applied to proper atomic systems.

quant-ph↗

arXiv2024

LayerMerge: Neural Network Depth Compression through Layer Pruning and Merging

Recent works show that reducing the number of layers in a convolutional neural network can enhance efficiency while maintaining the performance of the network. Existing depth compression methods remove redundant non-linear activation functions and merge the consecutive convolution layers into a single layer. However, these methods suffer from a critical drawback; the kernel size of the merged layers becomes larger, significantly undermining the latency reduction gained from reducing the depth of the network. We show that this problem can be addressed by jointly pruning convolution layers and activation functions. To this end, we propose LayerMerge, a novel depth compression method that selects which activation layers and convolution layers to remove, to achieve a desired inference speed-up while minimizing performance loss. Since the corresponding selection problem involves an exponential search space, we formulate a novel surrogate optimization problem and efficiently solve it via dynamic programming. Empirical results demonstrate that our method consistently outperforms existing depth compression and layer pruning methods on various network architectures, both on image classification and generation tasks. We release the code at https://github.com/snu-mllab/LayerMerge.

cs.LG↗

arXiv2024

A photonic quantum engine driven by superradiance

Performance of nano- and micro-scale heat engines can be improved with a help from quantum mechanical phenomena. Recently, heat reservoirs with quantum coherence have been proposed to enhance engine performance beyond the Carnot limit even with a single reservoir. However, no physical realizations have been achieved so far. Here, we report the first proof-of-principle experimental demonstration of a photonic quantum engine driven by superradiance employing a single heat reservoir composed of atoms and photonic vacuum. Reservoir atoms prepared in a quantum coherent superposition state underwent superradiance while traversing the cavity. This led to about 40-fold increase of the effective engine temperature, resulting in a near-unity engine efficiency. Moreover, the observed engine output power grew quadratically with respect to the atomic injection rate. Our work can be utilized in quantum mechanical heat transfer as well as in boosting engine powers, opening a pathway to development of photomechanical devices that run on quantum coherence embedded in heat baths.

quant-ph↗

arXiv2024

HyperCLOVA X Technical Report

We introduce HyperCLOVA X, a family of large language models (LLMs) tailored to the Korean language and culture, along with competitive capabilities in English, math, and coding. HyperCLOVA X was trained on a balanced mix of Korean, English, and code data, followed by instruction-tuning with high-quality human-annotated datasets while abiding by strict safety guidelines reflecting our commitment to responsible AI. The model is evaluated across various benchmarks, including comprehensive reasoning, knowledge, commonsense, factuality, coding, math, chatting, instruction-following, and harmlessness, in both Korean and English. HyperCLOVA X exhibits strong reasoning capabilities in Korean backed by a deep understanding of the language and cultural nuances. Further analysis of the inherent bilingual nature and its extension to multilingualism highlights the model's cross-lingual proficiency and strong generalization ability to untargeted languages, including machine translation between several language pairs and cross-lingual inference tasks. We believe that HyperCLOVA X can provide helpful guidance for regions or countries in developing their sovereign LLMs.

cs.CL↗

arXiv2023

Third-order exceptional point in an ion-cavity system

We investigate a scheme for observing the third-order exceptional point (EP3) in an ion-cavity setting. In the lambda-type level configuration, the ion is driven by a pump field, and the resonator is probed with another weak laser field. We exploit the highly asymmetric branching ratio of an ion's excited state to satisfy the weak-excitation limit, which allows us to construct the non-Hermitian Hamiltonian $(H_{\textrm{nH}})$. Via fitting the cavity-transmission spectrum, the eigenvalues of $H_{\textrm{nH}}$ are obtained. The EP3 appears at a point where the Rabi frequency of the pump laser and the atom-cavity coupling constant balance the loss rates of the system. Feasible experimental parameters are provided.

quant-ph↗

arXiv2023

Efficient Latency-Aware CNN Depth Compression via Two-Stage Dynamic Programming

Recent works on neural network pruning advocate that reducing the depth of the network is more effective in reducing run-time memory usage and accelerating inference latency than reducing the width of the network through channel pruning. In this regard, some recent works propose depth compression algorithms that merge convolution layers. However, the existing algorithms have a constricted search space and rely on human-engineered heuristics. In this paper, we propose a novel depth compression algorithm which targets general convolution operations. We propose a subset selection problem that replaces inefficient activation layers with identity functions and optimally merges consecutive convolution operations into shallow equivalent convolution operations for efficient end-to-end inference latency. Since the proposed subset selection problem is NP-hard, we formulate a surrogate optimization problem that can be solved exactly via two-stage dynamic programming within a few seconds. We evaluate our methods and baselines by TensorRT for a fair inference latency comparison. Our method outperforms the baseline method with higher accuracy and faster inference speed in MobileNetV2 on the ImageNet dataset. Specifically, we achieve $1.41\times$ speed-up with $0.11$\%p accuracy gain in MobileNetV2-1.0 on the ImageNet.

cs.LG↗

arXiv2023

Realization of superabsorption by time reversal of superradiance

Emission and absorption of light lie at the heart of light-matter interaction. Although emission and absorption rates are regarded as intrinsic properties of atoms and molecules, various ways to modify these rates have been sought in applications such as quantum information processing, metrology and light-energy harvesting. One promising approach is to utilize collective behaviour of emitters in the same way as in superradiance5. Although superradiance has been observed in diverse systems, its conceptual counterpart in absorption has never been realized11 until now. Here we demonstrate enhanced cooperative absorption - superabsorption - by implementing a time-reversal process of superradiance. The observed superabsorption rate is much higher than that of ordinary absorption, with the number of absorbed photons scaling with the square of the number of atoms, exhibiting the cooperative nature of superabsorption. The present superabsorption - which performs beyond the limitations of conventional absorption - can facilitate weak-signal sensing, light-energy harvesting and light-matter quantum interfaces

quant-ph↗

arXiv2022

Double exceptional points generated by the strong imaginary coupling of a non-Hermitian Hamiltonian in an optical microcavity

Exceptional points (EPs) have recently attracted considerable attention in the study of non-Hermitian systems and in applications such as sensors and mode switching. In particular, nontrivial topological structures of EPs have been studied intensively in relation to encircling EPs. Thus, EP generation is currently an important issue in several fields. To generate multiple EPs, multiple levels or composite physical systems have been employed with Hermitian couplings. In this study, we generate multiple EPs on two-level systems in a single microcavity by adopting the non-Hermitian coupling of a non-Hermitian Hamiltonian under the imaginary (dominant) coupling. The topological structures of Riemann surfaces generated by non-Hermitian coupling exhibit features that are different from those of Riemann surfaces generated by Hermitian coupling. The features of these topological structures of Riemann surfaces were verified by encircling multiple EPs and using a Riemann sphere.

quant-ph↗

arXiv2022

Dataset Condensation via Efficient Synthetic-Data Parameterization

The great success of machine learning with massive amounts of data comes at a price of huge computation costs and storage for training and tuning. Recent studies on dataset condensation attempt to reduce the dependence on such massive data by synthesizing a compact training dataset. However, the existing approaches have fundamental limitations in optimization due to the limited representability of synthetic datasets without considering any data regularity characteristics. To this end, we propose a novel condensation framework that generates multiple synthetic data with a limited storage budget via efficient parameterization considering data regularity. We further analyze the shortcomings of the existing gradient matching-based condensation methods and develop an effective optimization technique for improving the condensation of training data information. We propose a unified algorithm that drastically improves the quality of condensed data against the current state-of-the-art on CIFAR-10, ImageNet, and Speech Commands.

cs.LG↗

arXiv2021

What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers

GPT-3 shows remarkable in-context learning ability of large-scale language models (LMs) trained on hundreds of billion scale data. Here we address some remaining issues less reported by the GPT-3 paper, such as a non-English LM, the performances of different sized models, and the effect of recently introduced prompt optimization on in-context learning. To achieve this, we introduce HyperCLOVA, a Korean variant of 82B GPT-3 trained on a Korean-centric corpus of 560B tokens. Enhanced by our Korean-specific tokenization, HyperCLOVA with our training configuration shows state-of-the-art in-context zero-shot and few-shot learning performances on various downstream tasks in Korean. Also, we show the performance benefits of prompt-based learning and demonstrate how it can be integrated into the prompt engineering pipeline. Then we discuss the possibility of materializing the No Code AI paradigm by providing AI prototyping capabilities to non-experts of ML by introducing HyperCLOVA studio, an interactive prompt engineering interface. Lastly, we demonstrate the potential of our methods with three successful in-house applications.

cs.CL↗

arXiv2019

Behavior of Shannon entropy around an exceptional point in an open microcavity

We have investigated the Shannon entropy around an exceptional point (EP) in an open elliptical microcavity as a non-Hermitian system. The Shannon entropy had an extreme value at the EP in the parameter space. The Shannon entropies showed discontinuity across a specific line in the parameter space, directly related to the occurrence of exchange of the Shannon entropy as well as the mode patterns with that line as a boundary. This feature results in a nontrivial topological structure of the Shannon entropy surfaces.

quant-ph↗

arXiv2018

The Shannon entropy and avoided crossings in closed and open quantum billiards

The relation between the Shannon entropy and avoided crossings is investigated in dielectric microcavities. The Shannon entropy of probability density for eigenfunctions in an open elliptic billiard as well as a closed quadrupole billiard increases as the center of avoided crossing is approached. These results are opposite to those of atomic physics for electrons. It is found that the collective Lamb shift of the open quantum system and the symmetry breaking in the closed chaotic quantum system give equivalent effects to the Shannon entropy.

quant-ph↗

arXiv2017

Observation of a non-equilibrium steady state of cold atoms in a moving optical lattice

We investigated non-equilibrium atomic dynamics in a moving optical lattice via observation of atomic resonance fluorescence spectrum. A three-dimensional optical lattice was generated in a phase-stabilized magneto-optical trap (MOT) and the lattice was made to move by introducing a detuning between the counter-propagating trap lasers. A non-equilibrium steady states (NESS's) of atoms was then established in the hybrid of the moving optical lattice and the surrounding MOT. A part of atoms were localized and transported in the moving optical lattice and the rest were not localized in the lattice while trapped as a cold gas in the MOT. These motional states coexisted with continuous transition between them. As the speed of the lattice increased, the population of the non-localized state increased in a stepwise fashion due to the existence of bound states at the local minima of the lattice potential. A deterministic rate-equation model for atomic populations in those motional states was introduced in order to explain the experimental results. The model calculations then well reproduced the key features of the experimental observations, confirming the existence of an NESS in the cold atom system.

physics.atom-ph↗