arXiv subjects

Hyunmin Choi

Publications and source records attributed to Hyunmin Choi.

11 recordsLinked to original sources

arXiv2026

Simthesizer: An Agent-Driven Simulation Framework for LLM Serving Systems

System-level simulation is an essential tool for exploring the rapidly expanding design space of LLM serving systems, where real deployments remain costly and often infeasible. However, modern LLM serving now evolves faster than human-driven simulator development can track, and emerging workloads and mechanisms, from agentic workflows to disaggregated serving, no longer fit the monolithic simulation pipeline that existing simulators assume. Each new mechanism therefore demands an invasive rewrite, leaving a widening development gap between deployed serving systems and the simulators that model them. To close this gap, we present Simthesizer, a framework that realizes agent-driven simulator development. Simthesizer introduces a composable simulator infrastructure that uniformly expresses the complete serving workflow, including the control decisions that coordinate it, and realizes it as a unified dynamic graph in Simthesizer simulator. Synthesizer agent, a harnessed coding agent, then lowers natural-language feature requests onto this abstraction under simulator-specific guardrails and fidelity validation, evolving one shared simulator instead of building a new one for every feature. Under the same coding agent and harnesses, extensions built on Simthesizer follow a vLLM-based real system with 2.51% average throughput error, versus 6.03% for extensions built on existing simulators. On identical workloads, Simthesizer also simulates up to 284.96x and 23.19x faster than two state-of-the-art simulators, LLMServingSim2.0 and Vidur, respectively.

cs.AR↗

arXiv2026

Poisoned Playbooks: Demystifying Knowledge Poisoning Effects on AI Security Agents

AI security agents increasingly rely on Retrieval-Augmented Generation (RAG) to use external security knowledge for vulnerability analysis and exploit reasoning. This creates a new risk: poisoned write-ups can be operationalized into incorrect exploit behavior. Yet, prior work on RAG poisoning has mostly studied answer corruption in QA settings, much less is known about action-taking security agents. This paper aims to reveal such characteristics with crafted poisons about real-world challenges and AI agents. First, we demonstrate how a crafted single poisoned write-up injected into public-style security knowledge sources which we denote as Poisoned Playbooks, alters the behavior of RAG-based AI security agents. Across 11 CTF challenges, 3 frontier LLM families, 2 model generations, and 11 real-world CVEs, we find that poison adoption is systematic rather than random. To explain this pattern, we introduce the Verification Boundary (VB), a 3-level empirical classification based on what evidence the agent can use to refute a retrieved claim. Finally, we evaluate verification prompting and multi-source retrieval, showing that both help when stronger evidence exists, but weaken under sparse-evidence and zero-day conditions.

cs.CR↗

arXiv2026

HE-DAP: Homomorphic Encryption-based Dynamic Adaptive Parameter Optimization for Statistical Computation

Homomorphic encryption (HE) enables privacy-preserving analytics but remains hindered by high computational overhead. We find that the inverse square root-a key primitive in many statistical and machine learning workloads-exhibits inconsistent and often suboptimal performance across HE libraries and hardware. This stems from a core trade-off between two costly HE operations: evaluating high-degree Chebyshev polynomials to speed up Newton's method versus performing bootstrapping to manage ciphertext noise. Because their relative costs vary by up to 6x across environments, any fixed configuration proves inherently inefficient. To address this challenge, we present HE-DAP, a cross-platform optimization framework that automatically navigates this trade-off. By profiling an environment's unique performance characteristics, HE-DAP finds the optimal balance between polynomial degree and iteration count to accelerate the encrypted inverse square root computation for a given accuracy target. Our evaluation on Lattigo, HEaaN-CPU, and HEaaN-GPU shows that HE-DAP's adaptive approach yields significant performance gains. It accelerates the core inverse square root computation by up to 2.35x over the fixed configuration in PP-STAT while maintaining high numerical accuracy (MRE <= 3.1 x 10^-8). We further demonstrate that optimizing this fundamental building block directly enhances the end-to-end performance of complex statistical analyses, confirming the practical benefits of our environment-aware approach. By automatically adapting to heterogeneous execution environments, HE-DAP demonstrates that principled parameter optimization can make privacy-preserving statistical analytics practical at scale.

cs.CR↗

arXiv2026

LLMServingSim 2.0: A Unified Simulator for Heterogeneous and Disaggregated LLM Serving Infrastructure

Large language model (LLM) serving infrastructures are undergoing a shift toward heterogeneity and disaggregation. Modern deployments increasingly integrate diverse accelerators and near-memory processing technologies, introducing significant hardware heterogeneity, while system software increasingly separates computation, memory, and model components across distributed resources to improve scalability and efficiency. As a result, LLM serving performance is no longer determined by hardware or software choices in isolation, but by their runtime interaction through scheduling, data movement, and interconnect behavior. However, understanding these interactions remains challenging, as existing simulators lack the ability to jointly model heterogeneous hardware and disaggregated serving techniques within a unified, runtime-driven framework. This paper presents LLMServingSim 2.0, a unified system-level simulator designed to make runtime-driven hardware-software interactions in heterogeneous and disaggregated LLM serving infrastructures explicit and analyzable. LLMServingSim 2.0 embeds serving decisions and hardware behavior into a single runtime loop, enabling interaction-aware modeling of batching, routing, offloading, memory, and power. The simulator supports extensible integration of emerging accelerators and memory systems through profile-based modeling, while capturing dynamic serving behavior and system-level effects. We validate LLMServingSim 2.0 against real deployments, showing that it reproduces key performance, memory, and power metrics with an average error of 0.95%, while maintaining simulation times of around 10 minutes even for complex configurations. These results demonstrate that LLMServingSim 2.0 provides a practical bridge between hardware innovation and serving-system design, enabling systematic exploration and co-design for next-generation LLM serving infrastructures.

cs.DC↗

arXiv2025

LLMServingSim2.0: A Unified Simulator for Heterogeneous Hardware and Serving Techniques in LLM Infrastructure

This paper introduces LLMServingSim2.0, a system simulator designed for exploring heterogeneous hardware in large-scale LLM serving systems. LLMServingSim2.0 addresses two key limitations of its predecessor: (1) integrating hardware models into system-level simulators is non-trivial due to the lack of a clear abstraction, and (2) existing simulators support only a narrow subset of serving techniques, leaving no infrastructure that captures the breadth of approaches in modern LLM serving. To overcome these issues, LLMServingSim2.0 adopts trace-driven performance modeling, accompanied by an operator-level latency profiler, enabling the integration of new accelerators with a single command. It further embeds up-to-date serving techniques while exposing flexible interfaces for request routing, cache management, and scheduling policies. In a TPU case study, our profiler requires 18.5x fewer LoC and outperforms the predecessor's hardware-simulator integration, demonstrating LLMServingSim2.0's low-effort hardware extensibility. Our experiments further show that LLMServingSim2.0 reproduces GPU-based LLM serving with 1.9% error, while maintaining practical simulation time, making it a comprehensive platform for both hardware developers and LLM service providers.

cs.DC↗

arXiv2025

PP-STAT: An Efficient Privacy-Preserving Statistical Analysis Framework using Homomorphic Encryption

With the widespread adoption of cloud computing, the need for outsourcing statistical analysis to third-party platforms is growing rapidly. However, handling sensitive data such as medical records and financial information in cloud environments raises serious privacy concerns. In this paper, we present PP-STAT, a novel and efficient Homomorphic Encryption (HE)-based framework for privacy-preserving statistical analysis. HE enables computations to be performed directly on encrypted data without revealing the underlying plaintext. PP-STAT supports advanced statistical measures, including Z-score normalization, skewness, kurtosis, coefficient of variation, and Pearson correlation coefficient, all computed securely over encrypted data. To improve efficiency, PP-STAT introduces two key optimizations: (1) a Chebyshev-based approximation strategy for initializing inverse square root operations, and (2) a pre-normalization scaling technique that reduces multiplicative depth by folding constant scaling factors into mean and variance computations. These techniques significantly lower computational overhead and minimize the number of expensive bootstrapping procedures. Our evaluation on real-world datasets demonstrates that PP-STAT achieves high numerical accuracy, with mean relative error (MRE) below 2.4x10-4. Notably, the encrypted Pearson correlation coefficient between the smoker attribute and charges reaches 0.7873, with an MRE of 2.86x10-4. These results confirm the practical utility of PP-STAT for secure and precise statistical analysis in privacy-sensitive domains.

cs.CR↗

arXiv2024

Blind-Match: Efficient Homomorphic Encryption-Based 1:N Matching for Privacy-Preserving Biometric Identification

We present Blind-Match, a novel biometric identification system that leverages homomorphic encryption (HE) for efficient and privacy-preserving 1:N matching. Blind-Match introduces a HE-optimized cosine similarity computation method, where the key idea is to divide the feature vector into smaller parts for processing rather than computing the entire vector at once. By optimizing the number of these parts, Blind-Match minimizes execution time while ensuring data privacy through HE. Blind-Match achieves superior performance compared to state-of-the-art methods across various biometric datasets. On the LFW face dataset, Blind-Match attains a 99.63% Rank-1 accuracy with a 128-dimensional feature vector, demonstrating its robustness in face recognition tasks. For fingerprint identification, Blind-Match achieves a remarkable 99.55% Rank-1 accuracy on the PolyU dataset, even with a compact 16-dimensional feature vector, significantly outperforming the state-of-the-art method, Blind-Touch, which achieves only 59.17%. Furthermore, Blind-Match showcases practical efficiency in large-scale biometric identification scenarios, such as Naver Cloud's FaceSign, by processing 6,144 biometric samples in 0.74 seconds using a 128-dimensional feature vector.

cs.CV↗

arXiv2024

LLMServingSim: A HW/SW Co-Simulation Infrastructure for LLM Inference Serving at Scale

Recently, there has been an extensive research effort in building efficient large language model (LLM) inference serving systems. These efforts not only include innovations in the algorithm and software domains but also constitute developments of various hardware acceleration techniques. Nevertheless, there is a lack of simulation infrastructure capable of accurately modeling versatile hardware-software behaviors in LLM serving systems without extensively extending the simulation time. This paper aims to develop an effective simulation tool, called LLMServingSim, to support future research in LLM serving systems. In designing LLMServingSim, we focus on two limitations of existing simulators: (1) they lack consideration of the dynamic workload variations of LLM inference serving due to its autoregressive nature, and (2) they incur repetitive simulations without leveraging algorithmic redundancies in LLMs. To address these limitations, LLMServingSim simulates the LLM serving in the granularity of iterations, leveraging the computation redundancies across decoder blocks and reusing the simulation results from previous iterations. Additionally, LLMServingSim provides a flexible framework that allows users to plug in any accelerator compiler-and-simulation stacks for exploring various system designs with heterogeneous processors. Our experiments demonstrate that LLMServingSim produces simulation results closely following the performance behaviors of real GPU-based LLM serving system with less than 14.7% error rate, while offering 91.5x faster simulation speed compared to existing accelerator simulators.

cs.DC↗

arXiv2024

NeuPIMs: NPU-PIM Heterogeneous Acceleration for Batched LLM Inferencing

Modern transformer-based Large Language Models (LLMs) are constructed with a series of decoder blocks. Each block comprises three key components: (1) QKV generation, (2) multi-head attention, and (3) feed-forward networks. In batched processing, QKV generation and feed-forward networks involve compute-intensive matrix-matrix multiplications (GEMM), while multi-head attention requires bandwidth-heavy matrix-vector multiplications (GEMV). Machine learning accelerators like TPUs or NPUs are proficient in handling GEMM but are less efficient for GEMV computations. Conversely, Processing-in-Memory (PIM) technology is tailored for efficient GEMV computation, while it lacks the computational power to handle GEMM effectively. Inspired by this insight, we propose NeuPIMs, a heterogeneous acceleration system that jointly exploits a conventional GEMM-focused NPU and GEMV-optimized PIM devices. The main challenge in efficiently integrating NPU and PIM lies in enabling concurrent operations on both platforms, each addressing a specific kernel type. First, existing PIMs typically operate in a "blocked" mode, allowing only either NPU or PIM to be active at any given time. Second, the inherent dependencies between GEMM and GEMV in LLMs restrict their parallel processing. To tackle these challenges, NeuPIMs is equipped with dual row buffers in each bank, facilitating the simultaneous management of memory read/write operations and PIM commands. Further, NeuPIMs employs a runtime sub-batch interleaving technique to maximize concurrent execution, leveraging batch parallelism to allow two independent sub-batches to be pipelined within a single NeuPIMs device. Our evaluation demonstrates that compared to GPU-only, NPU-only, and a na\"ive NPU+PIM integrated acceleration approaches, NeuPIMs achieves 3$\times$, 2.4$\times$ and 1.6$\times$ throughput improvement, respectively.

cs.AR↗

arXiv2024

UniHENN: Designing Faster and More Versatile Homomorphic Encryption-based CNNs without im2col

Homomorphic encryption (HE) enables privacy-preserving deep learning by allowing computations on encrypted data without decryption. However, deploying convolutional neural networks (CNNs) with HE is challenging due to the need to convert input data into a two-dimensional matrix for convolution using the im2col technique, which rearranges the input for efficient computation. This restricts the types of CNN models that can be used since the encrypted data structure must be compatible with the specific model. UniHENN is a novel HE-based CNN architecture that eliminates the need for im2col, enhancing its versatility and compatibility with a broader range of CNN models. UniHENN flattens input data to one dimension without using im2col. The kernel performs convolutions by traversing the image, using incremental rotations and structured multiplication on the flattened input, with results spaced by the stride interval. Experimental results show that UniHENN significantly outperforms the state-of-the-art 2D CNN inference architecture named PyCrCNN in terms of inference time. For example, on the LeNet-1 model, UniHENN achieves an average inference time of 30.089 seconds, about 26.6 times faster than PyCrCNN's 800.591 seconds. Furthermore, UniHENN outperforms TenSEAL, an im2col-optimized CNN model, in concurrent image processing. For ten samples, UniHENN (16.247 seconds) was about 3.9 times faster than TenSEAL (63.706 seconds), owing to its support for batch processing of up to 10 samples. We demonstrate UniHENN's adaptability to various CNN architectures, including a 1D CNN and six 2D CNNs, highlighting its flexibility and efficiency for privacy-preserving cloud-based CNN services.

cs.CR↗

arXiv2023

Blind-Touch: Homomorphic Encryption-Based Distributed Neural Network Inference for Privacy-Preserving Fingerprint Authentication

Fingerprint authentication is a popular security mechanism for smartphones and laptops. However, its adoption in web and cloud environments has been limited due to privacy concerns over storing and processing biometric data on servers. This paper introduces Blind-Touch, a novel machine learning-based fingerprint authentication system leveraging homomorphic encryption to address these privacy concerns. Homomorphic encryption allows computations on encrypted data without decrypting. Thus, Blind-Touch can keep fingerprint data encrypted on the server while performing machine learning operations. Blind-Touch combines three strategies to efficiently utilize homomorphic encryption in machine learning: (1) It optimizes the feature vector for a distributed architecture, processing the first fully connected layer (FC-16) in plaintext on the client side and the subsequent layer (FC-1) post-encryption on the server, thereby minimizing encrypted computations; (2) It employs a homomorphic encryption compatible data compression technique capable of handling 8,192 authentication results concurrently; and (3) It utilizes a clustered server architecture to simultaneously process authentication results, thereby enhancing scalability with increasing user numbers. Blind-Touch achieves high accuracy on two benchmark fingerprint datasets, with a 93.6% F1- score for the PolyU dataset and a 98.2% F1-score for the SOKOTO dataset. Moreover, Blind-Touch can match a fingerprint among 5,000 in about 0.65 seconds. With its privacy focused design, high accuracy, and efficiency, Blind-Touch is a promising alternative to conventional fingerprint authentication for web and cloud applications.

cs.CR↗