arXiv subjects

Yuhao Zhang

Publications and source records attributed to Yuhao Zhang.

At least 163 records · Page 9Linked to original sources

arXiv2021

Stacked Acoustic-and-Textual Encoding: Integrating the Pre-trained Models into Speech Translation Encoders

Encoder pre-training is promising in end-to-end Speech Translation (ST), given the fact that speech-to-translation data is scarce. But ST encoders are not simple instances of Automatic Speech Recognition (ASR) or Machine Translation (MT) encoders. For example, we find that ASR encoders lack the global context representation, which is necessary for translation, whereas MT encoders are not designed to deal with long but locally attentive acoustic sequences. In this work, we propose a Stacked Acoustic-and-Textual Encoding (SATE) method for speech translation. Our encoder begins with processing the acoustic sequence as usual, but later behaves more like an MT encoder for a global representation of the input sequence. In this way, it is straightforward to incorporate the pre-trained models into the system. Also, we develop an adaptor module to alleviate the representation inconsistency between the pre-trained ASR encoder and MT encoder, and develop a multi-teacher knowledge distillation method to preserve the pre-training knowledge. Experimental results on the LibriSpeech En-Fr and MuST-C En-De ST tasks show that our method achieves state-of-the-art BLEU scores of 18.3 and 25.2. To our knowledge, we are the first to develop an end-to-end ST system that achieves comparable or even better BLEU performance than the cascaded ST counterpart when large-scale ASR and MT data is available.

cs.CL↗

arXiv2021

Improving Factual Completeness and Consistency of Image-to-Text Radiology Report Generation

Neural image-to-text radiology report generation systems offer the potential to improve radiology reporting by reducing the repetitive process of report drafting and identifying possible medical errors. However, existing report generation systems, despite achieving high performances on natural language generation metrics such as CIDEr or BLEU, still suffer from incomplete and inconsistent generations. Here we introduce two new simple rewards to encourage the generation of factually complete and consistent radiology reports: one that encourages the system to generate radiology domain entities consistent with the reference, and one that uses natural language inference to encourage these entities to be described in inferentially consistent ways. We combine these with the novel use of an existing semantic equivalence metric (BERTScore). We further propose a report generation system that optimizes these rewards via reinforcement learning. On two open radiology report datasets, our system substantially improved the F1 score of a clinical information extraction performance by +22.1 (Delta +63.9%). We further show via a human evaluation and a qualitative analysis that our system leads to generations that are more factually complete and consistent compared to the baselines.

cs.CL↗

arXiv2021

Brain Tumors Classification for MR images based on Attention Guided Deep Learning Model

In the clinical diagnosis and treatment of brain tumors, manual image reading consumes a lot of energy and time. In recent years, the automatic tumor classification technology based on deep learning has entered people's field of vision. Brain tumors can be divided into primary and secondary intracranial tumors according to their source. However, to our best knowledge, most existing research on brain tumors are limited to primary intracranial tumor images and cannot classify the source of the tumor. In order to solve the task of tumor source type classification, we analyze the existing technology and propose an attention guided deep convolution neural network (CNN) model. Meanwhile, the method proposed in this paper also effectively improves the accuracy of classifying the presence or absence of tumor. For the brain MR dataset, our method can achieve the average accuracy of 99.18% under ten-fold cross-validation for identifying the presence or absence of tumor, and 83.38% for classifying the source of tumor. Experimental results show that our method is consistent with the method of medical experts. It can assist doctors in achieving efficient clinical diagnosis of brain tumors.

eess.IV↗

arXiv2021

Finding Matrix Sequences with a High Asymptotic Growth Rate for Linear Constrained Switching Systems

Linear constrained switching systems are linear switched systems whose switching sequences are constrained by a deterministic finite automaton. This work investigates how to generate a sequence of matrices with an asymptotic growth rate close to the constrained joint spectral radius (CJSR) for constrained switching systems, based on our previous result that reveals the equivalence of a constrained switching system and a lifted arbitrary switching system. By using the dual solution of a sum-of-squares optimization program, an algorithm is designed for the lifted arbitrary switching system to produce a sequence of matrices with an asymptotic growth rate that is close to the CJSR of the original constrained switching system. It is also shown that a type of existing algorithms designed for arbitrary switching systems can be applied to the lifted system such that the desired sequence of matrices can be generated for the constrained switching system. Several numerical examples are provided to illustrate the better performance of the proposed algorithms compared with existing ones.

math.OC↗

arXiv2021

Do Syntax Trees Help Pre-trained Transformers Extract Information?

Much recent work suggests that incorporating syntax information from dependency trees can improve task-specific transformer models. However, the effect of incorporating dependency tree information into pre-trained transformer models (e.g., BERT) remains unclear, especially given recent studies highlighting how these models implicitly encode syntax. In this work, we systematically study the utility of incorporating dependency trees into pre-trained transformers on three representative information extraction tasks: semantic role labeling (SRL), named entity recognition, and relation extraction. We propose and investigate two distinct strategies for incorporating dependency structure: a late fusion approach, which applies a graph neural network on the output of a transformer, and a joint fusion approach, which infuses syntax structure into the transformer attention layers. These strategies are representative of prior work, but we introduce additional model design elements that are necessary for obtaining improved performance. Our empirical analysis demonstrates that these syntax-infused transformers obtain state-of-the-art results on SRL and relation extraction tasks. However, our analysis also reveals a critical shortcoming of these models: we find that their performance gains are highly contingent on the availability of human-annotated dependency parses, which raises important questions regarding the viability of syntax-augmented transformers in real-world applications.

cs.CL↗

arXiv2020

Adwords in a Panorama

Three decades ago, Karp, Vazirani, and Vazirani (STOC 1990) defined the online matching problem and gave an optimal $1-\frac{1}{e} \approx 0.632$-competitive algorithm. Fifteen years later, Mehta, Saberi, Vazirani, and Vazirani (FOCS 2005) introduced the first generalization called AdWords driven by online advertising and obtained the optimal $1-\frac{1}{e}$ competitive ratio in the special case of small bids. It has been open ever since whether there is an algorithm for general bids better than the $0.5$-competitive greedy algorithm. This paper presents a $0.5016$-competitive algorithm for AdWords, answering this open question on the positive end. The algorithm builds on several ingredients, including a combination of the online primal dual framework and the configuration linear program of matching problems recently explored by Huang and Zhang (STOC 2020), a novel formulation of AdWords which we call the panorama view, and a generalization of the online correlated selection by Fahrbach, Huang, Tao, and Zadimorghaddam (FOCS 2020) which we call the panoramic online correlated selection.

cs.DS↗

arXiv2020

Stay Hungry, Stay Focused: Generating Informative and Specific Questions in Information-Seeking Conversations

We investigate the problem of generating informative questions in information-asymmetric conversations. Unlike previous work on question generation which largely assumes knowledge of what the answer might be, we are interested in the scenario where the questioner is not given the context from which answers are drawn, but must reason pragmatically about how to acquire new information, given the shared conversation history. We identify two core challenges: (1) formally defining the informativeness of potential questions, and (2) exploring the prohibitively large space of potential questions to find the good candidates. To generate pragmatic questions, we use reinforcement learning to optimize an informativeness metric we propose, combined with a reward function designed to promote more specific questions. We demonstrate that the resulting pragmatic questioner substantially improves the informativeness and specificity of questions generated over a baseline model, as evaluated by our metrics as well as humans.

cs.CL↗

arXiv2020

Robustness to Programmable String Transformations via Augmented Abstract Training

Deep neural networks for natural language processing tasks are vulnerable to adversarial input perturbations. In this paper, we present a versatile language for programmatically specifying string transformations -- e.g., insertions, deletions, substitutions, swaps, etc. -- that are relevant to the task at hand. We then present an approach to adversarially training models that are robust to such user-defined string transformations. Our approach combines the advantages of search-based techniques for adversarial training with abstraction-based techniques. Specifically, we show how to decompose a set of user-defined string transformations into two component specifications, one that benefits from search and another from abstraction. We use our technique to train models on the AG and SST2 datasets and show that the resulting models are robust to combinations of user-defined transformations mimicking spelling mistakes and other meaning-preserving transformations.

cs.LG↗

arXiv2020

Biomedical and Clinical English Model Packages in the Stanza Python NLP Library

We introduce biomedical and clinical English model packages for the Stanza Python NLP library. These packages offer accurate syntactic analysis and named entity recognition capabilities for biomedical and clinical text, by combining Stanza's fully neural architecture with a wide variety of open datasets as well as large-scale unsupervised biomedical and clinical text data. We show via extensive experiments that our packages achieve syntactic analysis and named entity recognition performance that is on par with or surpasses state-of-the-art results. We further show that these models do not compromise speed compared to existing toolkits when GPU acceleration is available, and are made easy to download and use with Stanza's Python interface. A demonstration of our packages is available at: http://stanza.run/bio.

cs.CL↗

arXiv2020

Learning Architectures from an Extended Search Space for Language Modeling

Neural architecture search (NAS) has advanced significantly in recent years but most NAS systems restrict search to learning architectures of a recurrent or convolutional cell. In this paper, we extend the search space of NAS. In particular, we present a general approach to learn both intra-cell and inter-cell architectures (call it ESS). For a better search result, we design a joint learning method to perform intra-cell and inter-cell NAS simultaneously. We implement our model in a differentiable architecture search system. For recurrent neural language modeling, it outperforms a strong baseline significantly on the PTB and WikiText data, with a new state-of-the-art on PTB. Moreover, the learned architectures show good transferability to other systems. E.g., they improve state-of-the-art systems on the CoNLL and WNUT named entity recognition (NER) tasks and CoNLL chunking task, indicating a promising line of research on large-scale pre-learned architectures.

cs.LG↗

arXiv2020

Fully Online Matching II: Beating Ranking and Water-filling

Karp, Vazirani, and Vazirani (STOC 1990) initiated the study of online bipartite matching, which has held a central role in online algorithms ever since. Of particular importance are the Ranking algorithm for integral matching and the Water-filling algorithm for fractional matching. Most algorithms in the literature can be viewed as adaptations of these two in the corresponding models. Recently, Huang et al.~(STOC 2018, SODA 2019) introduced a more general model called \emph{fully online matching}, which considers general graphs and allows all vertices to arrive online. They also generalized Ranking and Water-filling to fully online matching and gave some tight analysis: Ranking is $Ω\approx 0.567$-competitive on bipartite graphs where the $Ω$-constant satisfies $Ωe^Ω= 1$, and Water-filling is $2-\sqrt{2} \approx 0.585$-competitive on general graphs. We propose fully online matching algorithms strictly better than Ranking and Water-filling. For integral matching on bipartite graphs, we build on the online primal dual analysis of Ranking and Water-filling to design a $0.569$-competitive hybrid algorithm called Balanced Ranking. To our knowledge, it is the first integral algorithm in the online matching literature that successfully integrates ideas from Water-filling. For fractional matching on general graphs, we give a $0.592$-competitive algorithm called Eager Water-filling, which may match a vertex on its arrival. By contrast, the original Water-filling algorithm always matches vertices at their deadlines. Our result for fractional matching further shows a separation between fully online matching and the general vertex arrival model by Wang and Wong (ICALP 2015), due to an upper bound of $0.5914$ in the latter model by Buchbinder, Segev, and Tkach (ESA 2017).

cs.DS↗

arXiv2020

Optimizing the Factual Correctness of a Summary: A Study of Summarizing Radiology Reports

Neural abstractive summarization models are able to generate summaries which have high overlap with human references. However, existing models are not optimized for factual correctness, a critical metric in real-world applications. In this work, we develop a general framework where we evaluate the factual correctness of a generated summary by fact-checking it automatically against its reference using an information extraction module. We further propose a training strategy which optimizes a neural summarization model with a factual correctness reward via reinforcement learning. We apply the proposed method to the summarization of radiology reports, where factual correctness is a key requirement. On two separate datasets collected from hospitals, we show via both automatic and human evaluation that the proposed approach substantially improves the factual correctness and overall quality of outputs over a competitive neural summarization system, producing radiology summaries that approach the quality of human-authored ones.

cs.CL↗

arXiv2020

Stanza: A Python Natural Language Processing Toolkit for Many Human Languages

We introduce Stanza, an open-source Python natural language processing toolkit supporting 66 human languages. Compared to existing widely used toolkits, Stanza features a language-agnostic fully neural pipeline for text analysis, including tokenization, multi-word token expansion, lemmatization, part-of-speech and morphological feature tagging, dependency parsing, and named entity recognition. We have trained Stanza on a total of 112 datasets, including the Universal Dependencies treebanks and other multilingual corpora, and show that the same neural architecture generalizes well and achieves competitive performance on all languages tested. Additionally, Stanza includes a native Python interface to the widely used Java Stanford CoreNLP software, which further extends its functionality to cover other tasks such as coreference resolution and relation extraction. Source code, documentation, and pretrained models for 66 languages are available at https://stanfordnlp.github.io/stanza.

cs.CL↗

arXiv2020

Towards a Better Understanding of Randomized Greedy Matching

There has been a long history for studying randomized greedy matching algorithms since the work by Dyer and Frieze~(RSA 1991). We follow this trend and consider the problem formulated in the oblivious setting, in which the algorithm makes (random) decisions that are essentially oblivious to the input graph. We revisit the \textsf{Modified Randomized Greedy (MRG)} algorithm by Aronson et al.~(RSA 1995) that is proved to be $(0.5+ε)$-approximate. In particular, we study a weaker version of the algorithm named \textsf{Random Decision Order (RDO)} that in each step, randomly picks an unmatched vertex and matches it to an arbitrary neighbor if exists. We prove the \textsf{RDO} algorithm is $0.639$-approximate and $0.531$-approximate for bipartite graphs and general graphs respectively. As a corollary, we substantially improve the approximation ratio of \textsf{MRG}. Furthermore, we generalize the \textsf{RDO} algorithm to the edge-weighted case and prove that it achieves a $0.501$ approximation ratio. This result solves the open question by Chan et al.~(SICOMP 2018) about the existence of an algorithm that beats greedy in this setting. As a corollary, it also solves the open questions by Gamlath et al.~(SODA 2019) in the stochastic setting.

cs.DS↗

arXiv2020

A Simple 1-1/e Approximation for Oblivious Bipartite Matching

We study the oblivious matching problem, which aims at finding a maximum matching on a graph with unknown edge set. Any algorithm for the problem specifies an ordering of the vertex pairs. The matching is then produced by probing the pairs following the ordering, and including a pair if both of them are unmatched and there exists an edge between them. The unweighted (Chan et al. (SICOMP 2018)) and the vertex-weighted (Chan et al. (TALG 2018)) versions of the problem are well studied. In this paper, we consider the edge-weighted oblivious matching problem on bipartite graphs, which generalizes the stochastic bipartite matching problem. Very recently, Gamlath et al. (SODA 2019) studied the stochastic bipartite matching problem, and proposed an (1-1/e)-approximate algorithm. We give a very simple algorithm adapted from the Ranking algorithm by Karp et al. (STOC 1990), and show that it achieves the same (1-1/e) approximation ratio for the oblivious matching problem on bipartite graph.

cs.DS↗

arXiv2019

Polylogarithmic Approximation Algorithm for k-Connected Directed Steiner Tree on Quasi-Bipartite Graphs

In the k-Connected Directed Steiner Tree problem (k-DST), we are given a directed graph G=(V, E) with edge (or vertex) costs, a root vertex r, a set of q terminals T, and a connectivity requirement k>0; the goal is to find a minimum-cost subgraph H of G such that H has k internally disjoint paths from the root r to each terminal t . The k-DST problem is a natural generalization of the classical Directed Steiner Tree problem (DST) in the fault-tolerant setting in which the solution subgraph is required to have an r,t-path, for every terminal t, even after removing k-1 edges or vertices. Despite being a classical problem, there are not many positive results on the problem, especially for the case k >= 3. In this paper, we will present an O(log k log q)-approximation algorithm for k-DST when an input graph is quasi-bipartite, i.e., when there is no edge joining two non-terminal vertices. To the best of our knowledge, our algorithm is the only known non-trivial approximation algorithm for k-DST, for k >= 3, that runs in polynomial-time regardless of the structure of the optimal solution. In addition, our algorithm is tight for every constant k, due to the hardness result inherited from the Set Cover problem.

cs.DS↗

arXiv2019

Online Vertex-Weighted Bipartite Matching: Beating 1-1/e with Random Arrivals

We introduce a weighted version of the ranking algorithm by Karp et al. (STOC 1990), and prove a competitive ratio of 0.6534 for the vertex-weighted online bipartite matching problem when online vertices arrive in random order. Our result shows that random arrivals help beating the 1-1/e barrier even in the vertex-weighted case. We build on the randomized primal-dual framework by Devanur et al. (SODA 2013) and design a two dimensional gain sharing function, which depends not only on the rank of the offline vertex, but also on the arrival time of the online vertex. To our knowledge, this is the first competitive ratio strictly larger than 1-1/e for an online bipartite matching problem achieved under the randomized primal-dual framework. Our algorithm has a natural interpretation that offline vertices offer a larger portion of their weights to the online vertices as time goes by, and each online vertex matches the neighbor with the highest offer at its arrival.

cs.DS↗

arXiv2019

MULDEF: Multi-model-based Defense Against Adversarial Examples for Neural Networks

Despite being popularly used in many applications, neural network models have been found to be vulnerable to adversarial examples, i.e., carefully crafted examples aiming to mislead machine learning models. Adversarial examples can pose potential risks on safety and security critical applications. However, existing defense approaches are still vulnerable to attacks, especially in a white-box attack scenario. To address this issue, we propose a new defense approach, named MulDef, based on robustness diversity. Our approach consists of (1) a general defense framework based on multiple models and (2) a technique for generating these multiple models to achieve high defense capability. In particular, given a target model, our framework includes multiple models (constructed from the target model) to form a model family. The model family is designed to achieve robustness diversity (i.e., an adversarial example successfully attacking one model cannot succeed in attacking other models in the family). At runtime, a model is randomly selected from the family to be applied on each input example. Our general framework can inspire rich future research to construct a desirable model family achieving higher robustness diversity. Our evaluation results show that MulDef (with only up to 5 models in the family) can substantially improve the target model's accuracy on adversarial examples by 22-74% in a white-box attack scenario, while maintaining similar accuracy on legitimate examples.

cs.LG↗