arXiv subjects

Jaemin Choi

Publications and source records attributed to Jaemin Choi.

6 recordsLinked to original sources

arXiv2026

SparSTAR: Sparse Attention for SpaceTime AutoRegressive Video Synthesis

InfinityStar extends visual autoregressive generation to video through a sequence of image and clip pyramids. Its changing scale and cross-clip context, however, leave late-scale attention costly and make sparse patterns reused from diffusion or image VAR models unreliable. We introduce SparSTAR, a training-free block-sparse attention method tailored to this setting. At each expensive scale and attention head, SparSTAR scores contiguous key blocks from the current query and key activations, retains required conditioning context, and executes the selected blocks through a forward-only sparse path. We analyze cross-scale consistency within a clip, pattern persistence across clip boundaries, and quality degradation as reuse spans increasingly distant scales. Across these analyses, important key blocks shift, showing that recomputing block selection at each target scale is more reliable than reusing a transferred mask. On 720p text-to-video and image-to-video generation, SparSTAR preserves every token and refinement scale while providing about a 1.6x end-to-end speedup and maintaining VBench and paired-output reconstruction fidelity close to dense InfinityStar.

cs.CV↗

arXiv2026

LiveWeb-IE: A Benchmark For Online Web Information Extraction

Web information extraction (WIE) is the task of automatically extracting data from web pages, offering high utility for various applications. The evaluation of WIE systems has traditionally relied on benchmarks built from HTML snapshots captured at a single point in time. However, this offline evaluation paradigm fails to account for the temporally evolving nature of the web; consequently, performance on these static benchmarks often fails to generalize to dynamic real-world scenarios. To bridge this gap, we introduce \dataset, a new benchmark designed for evaluating WIE systems directly against live websites. Based on trusted and permission-granted websites, we curate natural language queries that require information extraction of various data categories, such as text, images, and hyperlinks. We further design these queries to represent four levels of complexity, based on the number and cardinality of attributes to be extracted, enabling a granular assessment of WIE systems. In addition, we propose Visual Grounding Scraper (VGS), a novel multi-stage agentic framework that mimics human cognitive processes by visually narrowing down web page content to extract desired information. Extensive experiments across diverse backbone models demonstrate the effectiveness and robustness of VGS. We believe that this study lays the foundation for developing practical and robust WIE systems.

cs.CL↗

arXiv2023

Formal Verification of Chase-Lev Deque in Concurrent Separation Logic

Chase-Lev deque is a concurrent data structure designed for efficient load balancing in multiprocessor scheduling. It employs a work-stealing strategy, where each thread possesses its own work-stealing deque to store tasks, and idle threads steal tasks from other threads. However, given the inherent risk of bugs in software, particularly in a multiprocessor environment, it is crucial to formally establish the correctness of programs and data structures. To our knowledge, no formal verification work for the Chase-Lev deque has met three key criteria: (1) utilizing a minimal trusted computing base, (2) using a realistic and unrestricted implementation, and (3) proving a strong specification. In this thesis, we address this gap by presenting the formal verification of the Chase-Lev deque using a concurrent separation logic. Our work is mechanized in the Coq proof assistant, and our verified implementation is both realistic and unbounded in terms of the number of tasks it can handle. Also, we adopt linearizability as the specification, as it is widely recognized as a strong specification for concurrent data structures. Consequently, our work satisfies all three aforementioned criteria for formal verification. Additionally, we extend our verification to support safe memory reclamation, and provide a basis for verifying the Chase-Lev deque in the relaxed memory model.

cs.LO↗

arXiv2022

Accelerating Communication for Parallel Programming Models on GPU Systems

As an increasing number of leadership-class systems embrace GPU accelerators in the race towards exascale, efficient communication of GPU data is becoming one of the most critical components of high-performance computing. For developers of parallel programming models, implementing support for GPU-aware communication using native APIs for GPUs such as CUDA can be a daunting task as it requires considerable effort with little guarantee of performance. In this work, we demonstrate the capability of the Unified Communication X (UCX) framework to compose a GPU-aware communication layer that serves multiple parallel programming models of the Charm++ ecosystem: Charm++, Adaptive MPI (AMPI), and Charm4py. We demonstrate the performance impact of our designs with microbenchmarks adapted from the OSU benchmark suite, obtaining improvements in latency of up to 10.1x in Charm++, 11.7x in AMPI, and 17.4x in Charm4py. We also observe increases in bandwidth of up to 10.1x in Charm++, 10x in AMPI, and 10.5x in Charm4py. We show the potential impact of our designs on real-world applications by evaluating a proxy application for the Jacobi iterative method, improving the communication performance by up to 12.4x in Charm++, 12.8x in AMPI, and 19.7x in Charm4py.

cs.DC↗

arXiv2022

Improving Scalability with GPU-Aware Asynchronous Tasks

Asynchronous tasks, when created with over-decomposition, enable automatic computation-communication overlap which can substantially improve performance and scalability. This is not only applicable to traditional CPU-based systems, but also to modern GPU-accelerated platforms. While the ability to hide communication behind computation can be highly effective in weak scaling scenarios, performance begins to suffer with smaller problem sizes or in strong scaling due to fine-grained overheads and reduced room for overlap. In this work, we integrate GPU-aware communication into asynchronous tasks in addition to computation-communication overlap, with the goal of reducing time spent in communication and further increasing GPU utilization. We demonstrate the performance impact of our approach using a proxy application that performs the Jacobi iterative method, Jacobi3D. In addition to optimizations to minimize synchronizations between the host and GPU devices and increase the concurrency of GPU operations, we explore techniques such as kernel fusion and CUDA Graphs to mitigate fine-grained overheads at scale.

cs.DC↗

arXiv2021

Performance Evaluation of Python Parallel Programming Models: Charm4Py and mpi4py

Python is rapidly becoming the lingua franca of machine learning and scientific computing. With the broad use of frameworks such as Numpy, SciPy, and TensorFlow, scientific computing and machine learning are seeing a productivity boost on systems without a requisite loss in performance. While high-performance libraries often provide adequate performance within a node, distributed computing is required to scale Python across nodes and make it genuinely competitive in large-scale high-performance computing. Many frameworks, such as Charm4Py, DaCe, Dask, Legate Numpy, mpi4py, and Ray, scale Python across nodes. However, little is known about these frameworks' relative strengths and weaknesses, leaving practitioners and scientists without enough information about which frameworks are suitable for their requirements. In this paper, we seek to narrow this knowledge gap by studying the relative performance of two such frameworks: Charm4Py and mpi4py. We perform a comparative performance analysis of Charm4Py and mpi4py using CPU and GPU-based microbenchmarks other representative mini-apps for scientific computing.

cs.DC↗