arXiv subjects

Patrick Wang

Publications and source records attributed to Patrick Wang.

6 recordsLinked to original sources

arXiv2026

The "I" in FAIR: Translating from Interoperability in Principle to Interoperation in Practice

The FAIR (Findable, Accessible, Interoperable, and Reusable) data principles [1] promote the interoperability of scientific data by encouraging the use of persistent identifiers, standardized vocabularies, and formal metadata structures. Many resources are created using vocabularies that are FAIR-compliant and well-annotated, yet the collective ecosystem of these resources often fails to interoperate effectively in practice. This continued challenge is mainly due to variation in identifier schemas and data models used in these resources. We have created two tools to bridge the chasm between interoperability in principle and interoperation in practice. Babel solves the problem of multiple identifier schemes by producing a curated set of identifier mappings to create cliques of equivalent identifiers that are exposed through high-performance APIs. ORION solves the problems of multiple data models by ingesting knowledge bases and transforming them into a common, community-managed data model. Here, we describe Babel and ORION and demonstrate their ability to support data interoperation. A library of fully interoperable knowledge bases created through the application of Babel and ORION is available for download and use at https://robokop.renci.org.

cs.DB↗

arXiv2024

CREST: Effectively Compacting a Datastore For Retrieval-Based Speculative Decoding

We present CREST (Compact Retrieval-Based Speculative Decoding), a redesign of REST that allows it to be effectively "compacted". REST is a drafting technique for speculative decoding based on retrieving exact n-gram matches of the most recent n tokens generated by the target LLM from a datastore. The key idea of CREST is to only store a subset of the smallest and most common n-grams in the datastore with the hope of achieving comparable performance with less storage space. We found that storing a subset of n-grams both reduces storage space and improves performance. CREST matches REST's accepted token length with 10.6-13.5x less storage space and achieves a 16.5-17.1% higher acceptance length than REST using the same storage space on the HumanEval and MT Bench benchmarks.

cs.CL↗

arXiv2022

Detection of large exact subgraph isomorphisms with a topology-only graphlet index built using deterministic walks

We introduce the first algorithm to perform topology-only local graph matching (a.k.a. local network alignment or subgraph isomorphism): BLANT, for Basic Local Alignment of Network Topology. BLANT first creates a limited, high-specificity index of a single graph containing connected k-node induced subgraphs called k-graphlets, for k=6-15. The index is constructed in a deterministic way such that, if significant common network topology exists between two networks, their indexes are likely to overlap. This is the key insight which allows BLANT to discover alignments using only topological information. To align two networks, BLANT queries their respective indexes to form large, high quality local alignments. BLANT is able to discover highly topologically similar alignments (S3 >= 0.95) of up to 150 node-pairs for which up to 50% of node pairs differ from their "assigned" global counterpart. These results compare favorably against the baseline, a state-of-the-art local alignment algorithm which was adapted to be topology-only. Such alignments are 3x larger and differ 30% more (additive) more from the global alignment than alignments of similar topological similarity (S3 >= 0.95) discovered by the baseline. We hope that such regions of high local similarity and low global similarity may provide complementary insights to global alignment algorithms.

cs.SI↗

arXiv2022

BLANT: Basic Local Alignment of Network Topology, Part 1: Seeding local alignments with unambiguous 8-node graphlets

BLAST is a standard tool in bioinformatics for creating local sequence alignments using a "seed-and-extend" approach. Here we introduce an analogous seed-and-extend algorithm that produces local network alignments: BLANT, for Basic Local Alignment of Network Topology. This paper introduces BLANT-seed: given an input graph, BLANT-seed uses network topology alone to create a limited, high-specificity index of k-node induced subgraphs called k-graphlets (analogous to BLASTS's k-mers). The index is constructed so that, if significant common network topology exists between two graphs, their indexes are likely to overlap. BLANT-seed then queries the indexes of two networks to generate a list of common k-graphlets which, when paired, form a seed pair. Our companion paper (submitted elsewhere) describes BLANT-extend, which "grows" these seeds to larger local alignments, again using only topological information.

q-bio.MN↗

arXiv2021

Integer-only Zero-shot Quantization for Efficient Speech Recognition

End-to-end neural network models achieve improved performance on various automatic speech recognition (ASR) tasks. However, these models perform poorly on edge hardware due to large memory and computation requirements. While quantizing model weights and/or activations to low-precision can be a promising solution, previous research on quantizing ASR models is limited. In particular, the previous approaches use floating-point arithmetic during inference and thus they cannot fully exploit efficient integer processing units. Moreover, they require training and/or validation data during quantization, which may not be available due to security or privacy concerns. To address these limitations, we propose an integer-only, zero-shot quantization scheme for ASR models. In particular, we generate synthetic data whose runtime statistics resemble the real data, and we use it to calibrate models during quantization. We apply our method to quantize QuartzNet, Jasper, and Conformer and show negligible WER degradation as compared to the full-precision baseline models, even without using any data. Moreover, we achieve up to 2.35x speedup on a T4 GPU and 4x compression rate, with a modest WER degradation of <1% with INT8 quantization.

eess.AS↗

arXiv2017

Viewpoint Adaptation for Rigid Object Detection

An object detector performs suboptimally when applied to image data taken from a viewpoint different from the one with which it was trained. In this paper, we present a viewpoint adaptation algorithm that allows a trained single-view object detector to be adapted to a new, distinct viewpoint. We first illustrate how a feature space transformation can be inferred from a known homography between the source and target viewpoints. Second, we show that a variety of trained classifiers can be modified to behave as if that transformation were applied to each testing instance. The proposed algorithm is evaluated on a person detection task using images from the PETS 2007 and CAVIAR datasets, as well as from a new synthetic multi-view person detection dataset. It yields substantial performance improvements when adapting single-view person detectors to new viewpoints, and simultaneously reduces computational complexity. This work has the potential to improve detection performance for cameras viewing objects from arbitrary viewpoints, while simplifying data collection and feature extraction.

cs.CV↗