arXiv subjects

Zeyu Zhang

Publications and source records attributed to Zeyu Zhang.

294 records · Page 17Linked to original sources

arXiv2019

ScienceExamCER: A High-Density Fine-Grained Science-Domain Corpus for Common Entity Recognition

Named entity recognition identifies common classes of entities in text, but these entity labels are generally sparse, limiting utility to downstream tasks. In this work we present ScienceExamCER, a densely-labeled semantic classification corpus of 133k mentions in the science exam domain where nearly all (96%) of content words have been annotated with one or more fine-grained semantic class labels including taxonomic groups, meronym groups, verb/action groups, properties and values, and synonyms. Semantic class labels are drawn from a manually-constructed fine-grained typology of 601 classes generated through a data-driven analysis of 4,239 science exam questions. We show an off-the-shelf BERT-based named entity recognition model modified for multi-label classification achieves an accuracy of 0.85 F1 on this task, suggesting strong utility for downstream tasks in science domain question answering requiring densely-labeled semantic classification.

cs.CL↗

arXiv2019

Visualizing How Embeddings Generalize

Deep metric learning is often used to learn an embedding function that captures the semantic differences within a dataset. A key factor in many problem domains is how this embedding generalizes to new classes of data. In observing many triplet selection strategies for Metric Learning, we find that the best performance consistently arises from approaches that focus on a few, well selected triplets.We introduce visualization tools to illustrate how an embedding generalizes beyond measuring accuracy on validation data, and we illustrate the behavior of a range of triplet selection strategies.

cs.LG↗

arXiv2019

Lasserre Integrality Gaps for Graph Spanners and Related Problems

There has been significant recent progress on algorithms for approximating graph spanners, i.e., algorithms which approximate the best spanner for a given input graph. Essentially all of these algorithms use the same basic LP relaxation, so a variety of papers have studied the limitations of this approach and proved integrality gaps for this LP in a variety of settings. We extend these results by showing that even the strongest lift-and-project methods cannot help significantly, by proving polynomial integrality gaps even for $n^{Ω(ε)}$ levels of the Lasserre hierarchy, for both the directed and undirected spanner problems. We also extend these integrality gaps to related problems, notably Directed Steiner Network and Shallow-Light Steiner Network.

cs.CC↗

arXiv2018

Characterizing Demand Graphs for (Fixed-Parameter) Shallow-Light Steiner Network

We consider the Shallow-Light Steiner Network problem from a fixed-parameter perspective. Given a graph $G$, a distance bound $L$, and $p$ pairs of vertices $(s_1,t_1),\cdots,(s_p,t_p)$, the objective is to find a minimum-cost subgraph $G'$ such that $s_i$ and $t_i$ have distance at most $L$ in $G'$ (for every $i \in [p]$). Our main result is on the fixed-parameter tractability of this problem with parameter $p$. We exactly characterize the demand structures that make the problem "easy", and give FPT algorithms for those cases. In all other cases, we show that the problem is W$[1]$-hard. We also extend our results to handle general edge lengths and costs, precisely characterizing which demands allow for good FPT approximation algorithms and which demands remain W$[1]$-hard even to approximate.

cs.DS↗

arXiv2016

Approximating Approximate Distance Oracles

Given a finite metric space $(V,d)$, an approximate distance oracle is a data structure which, when queried on two points $u,v \in V$, returns an approximation to the the actual distance between $u$ and $v$ which is within some bounded stretch factor of the true distance. There has been significant work on the tradeoff between the important parameters of approximate distance oracles (and in particular between the size, stretch, and query time), but in this paper we take a different point of view, that of per-instance optimization. If we are given an particular input metric space and stretch bound, can we find the smallest possible approximate distance oracle for that particular input? Since this question is not even well-defined, we restrict our attention to well-known classes of approximate distance oracles, and study whether we can optimize over those classes. In particular, we give an $O(\log n)$-approximation to the problem of finding the smallest stretch $3$ Thorup-Zwick distance oracle, as well as the problem of finding the smallest Pǎtraşcu-Roditty distance oracle. We also prove a matching $Ω(\log n)$ lower bound for both problems, and an $Ω(n^{\frac{1}{k}-\frac{1}{2^{k-1}}})$ integrality gap for the more general stretch $(2k-1)$ Thorup-Zwick distance oracle. We also consider the problem of approximating the best TZ or PR approximate distance oracle \emph{with outliers}, and show that more advanced techniques (SDP relaxations in particular) allow us to optimize even in the presence of outliers.

cs.DS↗

arXiv2013

Ranking with Diverse Intents and Correlated Contents

We consider the following document ranking problem: We have a collection of documents, each containing some topics (e.g. sports, politics, economics). We also have a set of users with diverse interests. Assume that user u is interested in a subset I_u of topics. Each user u is also associated with a positive integer K_u, which indicates that u can be satisfied by any K_u topics in I_u. Each document s contains information for a subset C_s of topics. The objective is to pick one document at a time such that the average satisfying time is minimized, where a user's satisfying time is the first time that at least K_u topics in I_u are covered in the documents selected so far. Our main result is an O(ρ)-approximation algorithm for the problem, where ρ is the algorithmic integrality gap of the linear programming relaxation of the set cover instance defined by the documents and topics. This result generalizes the constant approximations for generalized min-sum set cover and ranking with unrelated intents and the logarithmic approximation for the problem of ranking with submodular valuations (when the submodular function is the coverage function), and can be seen as an interpolation between these results. We further extend our model to the case when each user may interest in more than one sets of topics and when the user's valuation function is XOS, and obtain similar results for these models.

cs.DS↗