arXiv subjects

Sanghyeon Lee

Publications and source records attributed to Sanghyeon Lee.

At least 19 recordsLinked to original sources

arXiv2026

Landscape-Similarity-Guided Optimization in Divide-and-Conquer QAOA

Divide-and-conquer strategies mitigate hardware constraints for the Quantum Approximate Optimization Algorithm (QAOA) on Noisy Intermediate-Scale Quantum (NISQ) devices by partitioning large interaction graphs into smaller, hardware-compatible sub-problems. However, this approach introduces a severe classical training bottleneck: a decomposition across $m$ boundary nodes generates $2^m$ distinct sub-problems that typically require independent optimization. In this work, we demonstrate that across diverse synthetic and real-world interaction graphs, the variational landscapes of these reduced QAOA instances actually exhibit a robust universality. Adapting the replica-overlap framework of spin-glass physics, we define a landscape-overlap order parameter $q$ to quantify geometric correlations between energy landscapes, revealing a sharp landscape-similarity transition as graph connectivity is tuned. Exploiting this, we introduce Doubly Optimized QAOA (DO-QAOA), an adaptive pipeline that collapses the sub-problems from $2^m$ distinct sub-problems into $K=\mathcal{O}(1)$ effective landscape classes. By performing optimization on a single representative sub-problem and dynamically transferring parameters to remaining sub-problems, DO-QAOA lowers runtime and quantum measurement overhead by orders of magnitude while maintaining a competitive Approximation Ratio Gap (ARG).

quant-ph↗

arXiv2026

Blow-up formulas of Gromov-Witten invariants and virtual cycles under positivity conditions

Let $\widetilde{X}$ be the blow-up of a smooth projective variety $X$ along a smooth subvariety $Z \subset X$ satisfying a numerical positivity condition. We study virtual cycles of moduli spaces of stable maps to $\widetilde{X}$. We lift the numerical blow-up formula of Chen and Du to a pushforward identity for virtual cycles of moduli spaces of genus-zero stable maps to $\widetilde{X}$ and $X$. Using virtual dimension calculations for the fixed loci and a free-leaf contraction morphism, we show that all correction terms in the master space localization formula vanish. Variants of this method yields further blow-up formulas, including a generalization of Gathmann's formula. We also give a counterexample to the point-blow-up conjecture in higher genus (proposed by Hu), and establish an absolute--relative correspondence in all genera under positivity and dimension assumptions, comparing the Gromov--Witten theories of $(\widetilde{X},E)$ and $X$, where $E$ is the exceptional divisor.

math.AG↗

arXiv2026

SUTURE: Syndrome-Guided Repair for Segmented Feasibility-Preserving VQAs on Noisy Hardware

Constrained binary optimization is a representative class of NP-hard problems in scheduling, resource allocation, and finance. Segmented feasibility-preserving variational quantum algorithms (VQAs) are a promising approach that restricts ideal circuit evolution to feasible assignments and executes short measure-and-reseed segments. The existing boundary runtime enforces feasibility through purification, which discards measurements that violate the constraints. As problem size and noise increase, feasible measurements become rare; if no shot survives, the execution chain terminates. In our 72-qubit graph-coloring experiment on a real-device IBM Heron, 11 of 12 purification chains terminate before completing all segments. We propose SUTURE: syndrome-guided repair, a runtime that repairs infeasible measurements instead of discarding them and is deployable on current quantum devices. SUTURE exploits a parity-check-like structure induced by the problem constraints: violated constraints form a syndrome that detects and localizes corruption and, under suitable structural conditions, often identifies a correction. SUTURE combines three stages: (1) a compile-time profiler that uses the compiled constraints and feasible initialization samples to predict single-flip recovery, with a maximum error of 0.011 across four constraint families; (2) a bounded runtime decoder that replaces purification inside the segmented execution loop; and (3) a regime analysis that identifies when repair is preferable to purification. In simulation up to 120 qubits, SUTURE continues the execution chain far beyond the noise level at which purification collapses. In the same 72-qubit experiment on IBM Heron hardware, SUTURE completes all segments in all 12 runs, and hardware timing experiment, decoding adds 1.6% to execution-path latency and less than 1% to total pipeline latency.

quant-ph↗

arXiv2026

LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning

Communication is a key component in multi-agent reinforcement learning (MARL) for mitigating partial observability, yet prior approaches often rely on inefficient information exchange or fail to transmit sufficient state information. To address this, we propose LLM-driven Multi-Agent Communication (LMAC), which leverages an LLM's reasoning capability to design a communication protocol that enables all agents to reconstruct the underlying state as accurately and uniformly as possible. LMAC iteratively refines the protocol using an explicit state-awareness criterion, improving state recovery while narrowing differences in agents' knowledge. Experiments on diverse MARL benchmarks show that LMAC improves state reconstruction across agents and yields substantial performance gains over prior communication baselines.

cs.AI↗

arXiv2026

Self-Improving Skill Learning for Robust Skill-based Meta-Reinforcement Learning

Meta-reinforcement learning (Meta-RL) facilitates rapid adaptation to unseen tasks but faces challenges in long-horizon environments. Skill-based approaches tackle this by decomposing state-action sequences into reusable skills and employing hierarchical decision-making. However, these methods are highly susceptible to noisy offline demonstrations, leading to unstable skill learning and degraded performance. To address this, we propose Self-Improving Skill Learning (SISL), which performs self-guided skill refinement using decoupled high-level and skill improvement policies, while applying skill prioritization via maximum return relabeling to focus updates on task-relevant trajectories, resulting in robust and stable adaptation even under noisy and suboptimal data. By mitigating the effect of noise, SISL achieves reliable skill learning and consistently outperforms other skill-based meta-RL methods on diverse long-horizon tasks. Our code is available at https://epsilog.github.io/SISL.

cs.LG↗

arXiv2026

Strict Subgoal Execution: Reliable Long-Horizon Planning in Hierarchical Reinforcement Learning

Long-horizon goal-conditioned tasks pose fundamental challenges for reinforcement learning (RL), particularly when goals are distant and rewards are sparse. While hierarchical and graph-based methods offer partial solutions, their reliance on conventional hindsight relabeling often fails to correct subgoal infeasibility, leading to inefficient high-level planning. To address this, we propose Strict Subgoal Execution (SSE), a graph-based hierarchical RL framework that integrates Frontier Experience Replay (FER) to separate unreachable from admissible subgoals and streamline high-level decision making. FER delineates the reachability frontier using failure and partial-success transitions, which identifies unreliable subgoals, increases subgoal reliability, and reduces unnecessary high-level decisions. Additionally, SSE employs a decoupled exploration policy to cover underexplored regions of the goal space and a path refinement that adjusts edge costs using observed low-level failures. Experimental results across diverse long-horizon benchmarks show that SSE consistently outperforms existing goal-conditioned and hierarchical RL methods in both efficiency and success rate. Our code is available at https://jaebak1996.github.io/SSE/

cs.LG↗

arXiv2026

Infinitesimal automorphisms and obstruction theory on the moduli of $L$-valued $G$-Higgs bundles

For an arbitrary reductive group $G$, we compute the infinitesimal automorphisms of $L$-valued principal $G$-Higgs bundles over a compact Kähler manifold $X$, extending known results for $Ω_X^{1}$-valued $G$-Higgs bundles. Using this computation, when $G$ is semisimple and $X$ is a smooth projective variety, we show that the moduli stack of stable $L$-valued $G$-Higgs bundles is a Deligne-Mumford (DM) stack. Furthermore, when $X$ is a smooth projective surface and $L=K_X$, we construct a symmetric perfect obstruction theory on this stable locus. We expect this will provide a foundation for defining Vafa-Witten invariants for reductive groups $G$.

math.AG↗

arXiv2026

PoseBridge: Bridging the Skeletonization Gap for Zero-Shot Skeleton-Based Action Recognition

Zero-shot skeleton-based action recognition (ZSSAR) is typically treated as a skeleton-text alignment problem: encode joint-coordinate sequences, align them with language, and classify unseen actions. We argue that this alignment is often too late. Skeletons are not complete action observations, but compressed outputs of human pose estimation (HPE); by the time alignment begins, human-object interactions and pose-relative visual cues may no longer be explicit. We call this upstream semantic loss. To address it, we propose PoseBridge, an HPE-aware ZSSAR framework that bridges intermediate HPE representations to skeleton-text alignment. Rather than adding an RGB action branch or object detector, PoseBridge extracts pose-anchored semantic cues from the same HPE process that produces skeletons, then transfers them through skeleton-conditioned bridging and semantic prototype adaptation. Across NTU-RGB+D 60/120, PKU-MMD, and Kinetics-200/400, PoseBridge improves ZSSAR performance under the evaluated protocols. On the Kinetics-200/400 PURLS benchmark, which contains in-the-wild videos with diverse scenes and action contexts, PoseBridge shows the clearest separation, improving the strongest compared baseline by 13.3-17.4 points across all eight splits. Our code will be publicly released.

cs.CV↗

arXiv2026

Adaptive Parallel Monte Carlo Tree Search for Efficient Test-time Compute Scaling

Monte Carlo Tree Search (MCTS) is an effective test-time compute scaling (TTCS) method for improving the reasoning performance of large language models, but its highly variable execution time leads to severe long-tail latency in practice. Existing optimizations such as positive early exit, reduce latency in favorable cases but are less effective when search continues without meaningful progress. We introduce {\it negative early exit}, which prunes unproductive MCTS trajectories, and an {\it adaptive boosting mechanism} that reallocates reclaimed computation to reduce resource contention among concurrent searches. Integrated into vLLM, these techniques substantially reduce p99 end-to-end latency while improving throughput and maintaining reasoning accuracy.

cs.AI↗

arXiv2026

OPRO: Orthogonal Panel-Relative Operators for Panel-Aware In-Context Image Generation

We introduce a parameter-efficient adaptation method for panel-aware in-context image generation with pre-trained diffusion transformers. The key idea is to compose learnable, panel-specific orthogonal operators onto the backbone's frozen positional encodings. This design provides two desirable properties: (1) isometry, which preserves the geometry of internal features, and (2) same-panel invariance, which maintains the model's pre-trained intra-panel synthesis behavior. Through controlled experiments, we demonstrate that the effectiveness of our adaptation method is not tied to a specific positional encoding design but generalizes across diverse positional encoding regimes. By enabling effective panel-relative conditioning, the proposed method consistently improves in-context image-based instructional editing pipelines, including state-of-the-art approaches.

cs.CV↗

arXiv2025

Irregular vanishing on $\mathbb{P}^2 \times \mathbb{P}^2$

In this paper, we describe Mixed-Spin-P(MSP) fields for a smooth CY 3-fold $X_{3,3} \subset \mathbb{P}^2 \times \mathbb{P}^2$. Then we describe $\mathbb{C}^* -$fixed loci of the moduli space of these MSP fields. We prove that any virtual localization term coming from the fixed locus corresponding to an irregular graph does not contribute to the invariant if the graph is not a pure loop, and also prove this vanishing property for the moduli space of N-MSP fields.

math.AG↗

arXiv2025

Enhancing the Yield of Bucket Brigade Quantum Random Access Memory using Redundancy Repair

Quantum Random Access Memory (qRAM) is an essential computing element for running oracle-based quantum algorithms. qRAM exploits quantum superposition to access all data stored in the memory cells simultaneously and guarantees the superior performance of quantum algorithms. A qRAM memory cell comprises logical qubits encoded through quantum error correction technology for successful operation against various quantum noises. In addition to quantum noise, the low-technology nodes based on silicon technology can increase the qubit density and may introduce defective qubits. As qRAM comprises many qubits, its yield will be reduced by defective qubits; these qubits must be handled using QEC scheme. However, the QEC scheme requires numerous physical qubits, which burdens resource overhead. In this paper, to resolve this overhead problem, we propose a novel quantum memory architecture that compensates for defective qubits by introducing redundant qubits. We also analyze the yield improvement offered by our proposed quantum memory architecture by varying the ideal fabrication error rate from 0.5% to 1% for different numbers of logical qubits in the qRAM. We demonstrate that for the qRAM comprising 1,024 logical qubits, eight redundant logical qubits improved the yield by 95.92% from that of qRAM not employing the redundant repair scheme.

quant-ph↗

arXiv2025

Hardware-based Heterogeneous Memory Management for Large Language Model Inference

A large language model (LLM) is one of the most important emerging machine learning applications nowadays. However, due to its huge model size and runtime increase of the memory footprint, LLM inferences suffer from the lack of memory capacity in conventional systems consisting of multiple GPUs with a modest amount of high bandwidth memory. Moreover, since LLM contains many bandwidthintensive kernels, only focusing on the memory capacity without considering the bandwidth incurs a serious performance degradation. To handle such conflicting memory capacity and bandwidth demands in a cost-effective way, this study investigates the potential of heterogeneous memory systems, proposing H2M2. It uses an asymmetric memory architecture consisting of capacity-centric and bandwidthcentric memory with computation units attached to each memory device. With the asymmetric memory, we first analyze the effect of kernel-memory mapping for the asymmetric memory. Second, we propose a dynamic runtime algorithm that finds a mapping solution considering the characteristics of LLM operations and the change of footprint during LLM inference. Third, we advocate the need for memory abstraction for the efficient management of the asymmetric memory. H2M2 outperforms the conventional homogeneous memory system with LPDDR by 1.46x, 1.55x, and 2.94x speedup in GPT3-175B, Chinchilla-70B, and Llama2-70B, respectively.

cs.AR↗

arXiv2025

Enabling Region-Specific Control via Lassos in Point-Based Colorization

Point-based interactive colorization techniques allow users to effortlessly colorize grayscale images using user-provided color hints. However, point-based methods often face challenges when different colors are given to semantically similar areas, leading to color intermingling and unsatisfactory results-an issue we refer to as color collapse. The fundamental cause of color collapse is the inadequacy of points for defining the boundaries for each color. To mitigate color collapse, we introduce a lasso tool that can control the scope of each color hint. Additionally, we design a framework that leverages the user-provided lassos to localize the attention masks. The experimental results show that using a single lasso is as effective as applying 4.18 individual color hints and can achieve the desired outcomes in 30% less time than using points alone.

cs.CV↗

arXiv2025

Efficient LLM Inference with Activation Checkpointing and Hybrid Caching

Recent large language models (LLMs) with enormous model sizes use many GPUs to meet memory capacity requirements incurring substantial costs for token generation. To provide cost-effective LLM inference with relaxed latency constraints, extensive research has focused on expanding GPU memory by leveraging the host memory. However, LLM inference engines that utilize the host memory often face underutilization of GPU compute units, as a considerable portion of inference time is spent in loading the model onto the GPU via host-GPU interconnect. To tackle these challenges of the host memory offloading for LLM, we introduce HybridServe, an LLM inference system with activation checkpointing based on activation caching. The activation cache stores activation checkpoints generated during intermediate inference stages, allowing the fast recomputation of KV cache while model parameters are transferred to GPU from host memory. Unlike conventional methods that recompute the KV cache from scratch using token IDs, the activation cache allows bypassing projection and FFN operations. To balance between the activation recomputation and parameter loading overhead, this study proposes a KV-activation hybrid caching scheme which finds the best ratio of the key-value and activation caches to adjust the recomputation time. Our system achieves 2.19x throughput improvement over the state-of-the-art prior work for offloading both model weights and KV cache.

cs.DC↗

arXiv2024

Anomaly Detection for People with Visual Impairments Using an Egocentric 360-Degree Camera

Recent advancements in computer vision have led to a renewed interest in developing assistive technologies for individuals with visual impairments. Although extensive research has been conducted in the field of computer vision-based assistive technologies, most of the focus has been on understanding contexts in images, rather than addressing their physical safety and security concerns. To address this challenge, we propose the first step towards detecting anomalous situations for visually impaired people by observing their entire surroundings using an egocentric 360-degree camera. We first introduce a novel egocentric 360-degree video dataset called VIEW360 (Visually Impaired Equipped with Wearable 360-degree camera), which contains abnormal activities that visually impaired individuals may encounter, such as shoulder surfing and pickpocketing. Furthermore, we propose a new architecture called the FDPN (Frame and Direction Prediction Network), which facilitates frame-level prediction of abnormal events and identifying of their directions. Finally, we evaluate our approach on our VIEW360 dataset and the publicly available UCF-Crime and Shanghaitech datasets, demonstrating state-of-the-art performance.

cs.CV↗

arXiv2024

Exclusively Penalized Q-learning for Offline Reinforcement Learning

Constraint-based offline reinforcement learning (RL) involves policy constraints or imposing penalties on the value function to mitigate overestimation errors caused by distributional shift. This paper focuses on a limitation in existing offline RL methods with penalized value function, indicating the potential for underestimation bias due to unnecessary bias introduced in the value function. To address this concern, we propose Exclusively Penalized Q-learning (EPQ), which reduces estimation bias in the value function by selectively penalizing states that are prone to inducing estimation errors. Numerical results show that our method significantly reduces underestimation bias and improves performance in various offline control tasks compared to other offline RL methods

cs.LG↗

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 naïve NPU+PIM integrated acceleration approaches, NeuPIMs achieves 3$\times$, 2.4$\times$ and 1.6$\times$ throughput improvement, respectively.

cs.AR↗