Support vector machine (SVM) has attracted great attentions for the last two decades due to its extensive applications, and thus numerous optimization models have been proposed. To distinguish all of them, in this paper, we introduce a new model equipped with an $L_{0/1}$ soft-margin loss (dubbed as $L_{0/1}$-SVM) which well captures the nature of the binary classification. Many of the existing convex/non-convex soft-margin losses can be viewed as a surrogate of the $L_{0/1}$ soft-margin loss. Despite the discrete nature of $L_{0/1}$, we manage to establish the existence of global minimizer of the new model as well as revealing the relationship among its minimizers and KKT/P-stationary points. These theoretical properties allow us to take advantage of the alternating direction method of multipliers. In addition, the $L_{0/1}$-support vector operator is introduced as a filter to prevent outliers from being support vectors during the training process. Hence, the method is expected to be relatively robust. Finally, numerical experiments demonstrate that our proposed method generates better performance in terms of much shorter computational time with much fewer number of support vectors when against with some other leading methods in areas of SVM. When the data size gets bigger, its advantage becomes more evident.
We recorded and preprocessed ZuCo 2.0, a new dataset of simultaneous eye-tracking and electroencephalography during natural reading and during annotation. This corpus contains gaze and brain activity data of 739 sentences, 349 in a normal reading paradigm and 390 in a task-specific paradigm, in which the 18 participants actively search for a semantic relation type in the given sentences as a linguistic annotation task. This new dataset complements ZuCo 1.0 by providing experiments designed to analyze the differences in cognitive processing between natural reading and annotation. The data is freely available here: url{https://osf.io/2urht/
This paper focuses on valuating training data for supervised learning tasks and studies the Shapley value, a data value notion originated in cooperative game theory. The Shapley value defines a unique value distribution scheme that satisfies a set of appealing properties desired by a data value notion. However, the Shapley value requires exponential complexity to calculate exactly. Existing approximation algorithms, although achieving great improvement over the exact algorithm, relies on retraining models for multiple times, thus remaining limited when applied to larger-scale learning tasks and real-world datasets. In this work, we develop a simple and efficient heuristic for data valuation based on the Shapley value with complexity independent with the model size. The key idea is to approximate the model via a $K$-nearest neighbor ($K$NN) classifier, which has a locality structure that can lead to efficient Shapley value calculation. We evaluate the utility of the values produced by the $K$NN proxies in various settings, including label noise correction, watermark detection, data summarization, active data acquisition, and domain adaption. Extensive experiments demonstrate that our algorithm achieves at least comparable utility to the values produced by existing algorithms while significant efficiency improvement. Moreover, we theoretically analyze the Shapley value and justify its advantage over the leave-one-out error as a data value measure.
Translating document renderings (e.g. PDFs, scans) into hierarchical structures is extensively demanded in the daily routines of many real-world applications, and is often a prerequisite step of many downstream NLP tasks. Earlier attempts focused on different but simpler tasks such as the detection of table or cell locations within documents; however, a holistic, principled approach to inferring the complete hierarchical structure in documents is missing. As a remedy, we developed "DocParser": an end-to-end system for parsing the complete document structure - including all text elements, figures, tables, and table cell structures. To the best of our knowledge, DocParser is the first system that derives the full hierarchical document compositions. Given the complexity of the task, annotating appropriate datasets is costly. Therefore, our second contribution is to provide a dataset for evaluating hierarchical document structure parsing. Our third contribution is to propose a scalable learning framework for settings where domain-specific data is scarce, which we address by a novel approach to weak supervision. Our computational experiments confirm the effectiveness of our proposed weak supervision: Compared to the baseline without weak supervision, it improves the mean average precision for detecting document entities by 37.1%. When classifying hierarchical relations between entity pairs, it improves the F1 score by 27.6%.
An interesting method of evaluating word representations is by how much they reflect the semantic representations in the human brain. However, most, if not all, previous works only focus on small datasets and a single modality. In this paper, we present the first multi-modal framework for evaluating English word representations based on cognitive lexical semantics. Six types of word embeddings are evaluated by fitting them to 15 datasets of eye-tracking, EEG and fMRI signals recorded during language processing. To achieve a global score over all evaluation hypotheses, we apply statistical significance testing accounting for the multiple comparisons problem. This framework is easily extensible and available to include other intrinsic and extrinsic evaluation methods. We find strong correlations in the results between cognitive datasets, across recording modalities and to their performance on extrinsic NLP tasks.
Graph Convolutional Network (GCN) has attracted intensive interests recently. One major limitation of GCN is that it often cannot benefit from using a deep architecture, while traditional CNN and an alternative Graph Neural Network architecture, namely GraphCNN, often achieve better quality with a deeper neural architecture. How can we explain this phenomenon? In this paper, we take the first step towards answering this question. We first conduct a systematic empirical study on the accuracy of GCN, GraphCNN, and ResNet-18 on 2D images and identified relative importance of different factors in architectural design. This inspired a novel theoretical analysis on the mutual information between the input and the output after $l$ GCN and GraphCNN layers. We identified regimes in which GCN suffers exponentially fast information lose and show that GraphCNN requires a much weaker condition for similar behavior to happen.
Deep neural networks have recently advanced the state-of-the-art in image compression and surpassed many traditional compression algorithms. The training of such networks involves carefully trading off entropy of the latent representation against reconstruction quality. The term quality crucially depends on the observer of the images which, in the vast majority of literature, is assumed to be human. In this paper, we go beyond this notion of quality and look at human visual perception and machine perception simultaneously. To that end, we propose a family of loss functions that allows to optimize deep image compression depending on the observer and to interpolate between human perceived visual quality and classification accuracy. Our experiments show that our proposed training objectives result in compression systems that, when trained with machine friendly loss, preserve accuracy much better than the traditional codecs BPG, WebP and JPEG, without requiring fine-tuning of inference algorithms on decoded images and independent of the classifier architecture. At the same time, when using the human friendly loss, we achieve competitive performance in terms of MS-SSIM.