arXiv subjects

Steven Xia

Publications and source records attributed to Steven Xia.

3 recordsLinked to original sources

arXiv2025

Spectrally symmetric orientations of graphs

The Hermitian adjacency matrices of digraphs based on the sixth root of unity were introduced in [B. Mohar, A new kind of Hermitian matrices for digraphs, Linear Alg. Appl. (2020)]. They appear to be the most natural choice for the spectral theory of digraphs. Undirected graphs have adjacency spectrum symmetric about 0 if and only if they are bipartite. The situation is more complex for the Hermitian spectra of digraphs. In this paper we study non-bipartite oriented graphs with symmetric Hermitian spectra. Our main result concerns the extremal problem of maximizing the density of spectrally symmetric oriented graphs. The maximum possible density is shown to be between 31/36} and 10/11. Furthermore, we give a necessary condition for an oriented graph to be spectrally symmetric based on the adjacency spectrum of the underlying graph. This allows us to show that line graphs of sufficiently dense graphs do not admit spectrally symmetric orientations. We also show how to construct infinite families of spectrally symmetric graphs using 1-sums.

math.CO↗

arXiv2025

Hermitian adjacency matrices with at most three distinct eigenvalues

We study oriented graphs whose Hermitian adjacency matrices of the second kind have few eigenvalues. We give a complete characterization of the oriented graphs with two distinct eigenvalues, showing that there are only four such graphs. We extend this result to mixed graphs. We show that there are infinitely many regular tournaments with three distinct eigenvalues. We extend our main results to Hermitian adjacency matrices defined over other roots of unity.

math.CO↗

arXiv2023

Optimal Pricing for Data-Augmented AutoML Marketplaces

Organizations often lack sufficient data to effectively train machine learning (ML) models, while others possess valuable data that remains underutilized. Data markets promise to unlock substantial value by matching data suppliers with demand from ML consumers. However, market design involves addressing intricate challenges, including data pricing, fairness, robustness, and strategic behavior. In this paper, we propose a pragmatic data-augmented AutoML market that seamlessly integrates with existing cloud-based AutoML platforms such as Google's Vertex AI and Amazon's SageMaker. Unlike standard AutoML solutions, our design automatically augments buyer-submitted training data with valuable external datasets, pricing the resulting models based on their measurable performance improvements rather than computational costs as the status quo. Our key innovation is a pricing mechanism grounded in the instrumental value - the marginal model quality improvement - of externally sourced data. This approach bypasses direct dataset pricing complexities, mitigates strategic buyer behavior, and accommodates diverse buyer valuations through menu-based options. By integrating automated data and model discovery, our solution not only enhances ML outcomes but also establishes an economically sustainable framework for monetizing external data.

cs.LG↗