arXiv subjects

Victor Zhang

Publications and source records attributed to Victor Zhang.

6 recordsLinked to original sources

arXiv2026

OpenZL: Using Graphs to Compress Smaller and Faster

In the last few decades, research techniques have improved lossless compression ratios by significantly increasing processing time. However, these techniques have not gained popularity in industry because production systems require high throughput and low resource utilization. Instead, real world improvements in compression are increasingly realized by building application-specific compressors which can exploit knowledge about the structure and semantics of the data being compressed. Application-specific compressor systems outperform even the best generic compressors, but these techniques have severe drawbacks -- they are inherently limited in applicability, are hard to develop, and are difficult to maintain and deploy. In this work, we show that these challenges can be overcome with a new compression strategy. We propose the "graph model" of compression, a new theoretical framework for representing compression as a directed acyclic graph of modular codecs. OpenZL implements this framework and compresses data into a self-describing wire format, any configuration of which can be decompressed by a universal decoder. OpenZL's design enables rapid development of application-specific compressors with minimal code. Experimental results demonstrate that OpenZL achieves superior compression ratios and speeds compared to state-of-the-art general-purpose compressors on a variety of real-world datasets. Compared to ratio-focused deep-learning compressors, OpenZL is competitive on ratio while being many orders of magnitude faster. Internal deployments at Meta have also shown consistent improvements in size and/or speed, with development timelines reduced from months to days. OpenZL thus represents a significant advance in practical, scalable, and maintainable data compression for modern data-intensive applications.

cs.IR↗

arXiv2025

OpenZL: A Graph-Based Model for Compression

Research techniques in the last decade have improved lossless compression ratios by significantly increasing processing time. These techniques have remained obscure because production systems require high throughput and low resource utilization. In practice, application-specific compression algorithms that leverage knowledge of the data structure and semantics are more popular. Application-specific compressor systems outperform even the best generic compressors, but these techniques have some drawbacks. Application-specific compressors are inherently limited in applicability, have high development costs, and are difficult to maintain and deploy. In this work, we show that these challenges can be overcome with a new compression strategy. We propose the "graph model" of compression, a new theoretical framework for representing compression as a directed acyclic graph of modular codecs. OpenZL compresses data into a self-describing wire format, any configuration of which can be decompressed by a universal decoder. OpenZL's design enables rapid development of tailored compressors with minimal code; its universal decoder eliminates deployment lag; and its investment in a well-vetted standard component library minimizes security risks. Experimental results demonstrate that OpenZL achieves superior compression ratios and speeds compared to state-of-the-art general-purpose compressors on a variety of real-world datasets. Internal deployments at Meta have also shown consistent improvements in size and/or speed, with development timelines reduced from months to days. OpenZL thus represents a significant advance in practical, scalable, and maintainable data compression for modern data-intensive applications.

cs.IR↗

arXiv2025

Big data comparison of quantum invariants

We apply big data techniques, including exploratory and topological data analysis, to investigate quantum invariants. More precisely, our study explores the Jones polynomial's structural properties and contrasts its behavior under four principal methods of enhancement: coloring, rank increase, categorification, and leaving the realm of Lie algebras.

math.GT↗

arXiv2024

Exploring the Landscape of Distributed Graph Sketching

Recent work has initiated the study of dense graph processing using graph sketching methods, which drastically reduce space costs by lossily compressing information about the input graph. In this paper, we explore the strange and surprising performance landscape of sketching algorithms. We highlight both their surprising advantages for processing dense graphs that were previously prohibitively expensive to study, as well as the current limitations of the technique. Most notably, we show how sketching can avoid bottlenecks that limit conventional graph processing methods. Single-machine streaming graph processing systems are typically bottlenecked by CPU performance, and distributed graph processing systems are typically bottlenecked by network latency. We present Landscape, a distributed graph-stream processing system that uses linear sketching to distribute the CPU work of computing graph properties to distributed workers with no need for worker-to-worker communication. As a result, it overcomes the CPU and network bottlenecks that limit other systems. In fact, for the connected components problem, Landscape achieves a stream ingestion rate one-fourth that of maximum sustained RAM bandwidth, and is four times faster than random access RAM bandwidth. Additionally, we prove that for any sequence of graph updates and queries Landscape consumes at most a constant factor more network bandwidth than is required to receive the input stream. We show that this system can ingest up to 332 million stream updates per second on a graph with $2^{17}$ vertices. We show that it scales well with more distributed compute power: given a cluster of 40 distributed worker machines, it can ingest updates 35 times as fast as with 1 distributed worker machine. Landscape uses heuristics to reduce its query latency by up to four orders of magnitude over the prior state of the art.

cs.DC↗

arXiv2022

GraphZeppelin: Storage-Friendly Sketching for Connected Components on Dynamic Graph Streams

Finding the connected components of a graph is a fundamental problem with uses throughout computer science and engineering. The task of computing connected components becomes more difficult when graphs are very large, or when they are dynamic, meaning the edge set changes over time subject to a stream of edge insertions and deletions. A natural approach to computing the connected components on a large, dynamic graph stream is to buy enough RAM to store the entire graph. However, the requirement that the graph fit in RAM is prohibitive for very large graphs. Thus, there is an unmet need for systems that can process dense dynamic graphs, especially when those graphs are larger than available RAM. We present a new high-performance streaming graph-processing system for computing the connected components of a graph. This system, which we call GraphZeppelin, uses new linear sketching data structures (CubeSketches) to solve the streaming connected components problem and as a result requires space asymptotically smaller than the space required for a lossless representation of the graph. GraphZeppelin is optimized for massive dense graphs: GraphZeppelin can process millions of edge updates (both insertions and deletions) per second, even when the underlying graph is far too large to fit in available RAM. As a result GraphZeppelin vastly increases the scale of graphs that can be processed.

cs.DS↗

arXiv2019

Optimizing Multi-GPU Parallelization Strategies for Deep Learning Training

Deploying deep learning (DL) models across multiple compute devices to train large and complex models continues to grow in importance because of the demand for faster and more frequent training. Data parallelism (DP) is the most widely used parallelization strategy, but as the number of devices in data parallel training grows, so does the communication overhead between devices. Additionally, a larger aggregate batch size per step leads to statistical efficiency loss, i.e., a larger number of epochs are required to converge to a desired accuracy. These factors affect overall training time and beyond a certain number of devices, the speedup from leveraging DP begins to scale poorly. In addition to DP, each training step can be accelerated by exploiting model parallelism (MP). This work explores hybrid parallelization, where each data parallel worker is comprised of more than one device, across which the model dataflow graph (DFG) is split using MP. We show that at scale, hybrid training will be more effective at minimizing end-to-end training time than exploiting DP alone. We project that for Inception-V3, GNMT, and BigLSTM, the hybrid strategy provides an end-to-end training speedup of at least 26.5%, 8%, and 22% respectively compared to what DP alone can achieve at scale.

cs.LG↗