IJCAI 2020 Papers

Learning pseudo-Boolean (PB) constraints in PB solvers exploiting cutting planes based inference is not as well understood as clause learning in conflict-driven clause learning solvers. In this paper, we show that PB constraints derived using cutting planes may contain irrelevant literals, i.e., literals whose assigned values (whatever they are) never change the truth value of the constraint. Such literals may lead to infer constraints that are weaker than they should be, impacting the size of the proof built by the solver, and thus also affecting its performance. This suggests that current implementations of PB solvers based on cutting planes should be reconsidered to prevent the generation of irrelevant literals. Indeed, detecting and removing irrelevant literals is too expensive in practice to be considered as an option (the associated problem is NP-hard).

Unsupervised active learning has attracted increasing attention in recent years, where its goal is to select representative samples in an unsupervised setting for human annotating. Most existing works are based on shallow linear models by assuming that each sample can be well approximated by the span (i.e., the set of all linear combinations) of certain selected samples, and then take these selected samples as representative ones to label. However, in practice, the data do not necessarily conform to linear models, and how to model nonlinearity of data often becomes the key point to success. In this paper, we present a novel Deep neural network framework for Unsupervised Active Learning, called DUAL. DUAL can explicitly learn a nonlinear embedding to map each input into a latent space through an encoder-decoder architecture, and introduce a selection block to select representative samples in the the learnt latent space. In the selection block, DUAL considers to simultaneously preserve the whole input patterns as well as the cluster structure of data. Extensive experiments are performed on six publicly available datasets, and experimental results clearly demonstrate the efficacy of our method, compared with state-of-the-arts.

Iterated belief change aims to determine how the belief state of a rational agent evolves given a sequence of change formulae. Several families of iterated belief change operators (revision operators, improvement operators) have been pointed out so far, and characterized from an axiomatic point of view. This paper focuses on the inference problem for iterated belief change, when belief states are represented as a special kind of stratified belief bases. The computational complexity of the inference problem is identified and shown to be identical for all revision operators satisfying Darwiche and Pearl's (R*1-R*6) postulates. In addition, some complexity bounds for the inference problem are provided for the family of soft improvement operators. We also show that a revised belief state can be computed in a reasonable time for large-sized instances using SAT-based algorithms, and we report empirical results showing the feasibility of iterated belief change for bases of significant sizes.

This research is an endeavor to combine deep-learning-based language modeling with classical topic modeling techniques to produce interpretable topics for a given set of documents in Urdu, a low resource language. The existing topic modeling techniques produce a collection of words, often un-interpretable, as suggested topics without integrat-ing them into a semantically correct phrase/sentence. The proposed approach would first build an accurate Part of Speech (POS) tagger for the Urdu Language using a publicly available corpus of many million sentences. Using semanti-cally rich feature extraction approaches including Word2Vec and BERT, the proposed approach, in the next step, would experiment with different clus-tering and topic modeling techniques to produce a list of potential topics for a given set of documents. Finally, this list of topics would be sent to a labeler module to produce syntactically correct phrases that will represent interpretable topics.

Spatio-temporal video grounding aims to retrieve the spatio-temporal tube of a queried object according to the given sentence. Currently, most existing grounding methods are restricted to well-aligned segment-sentence pairs. In this paper, we explore spatio-temporal video grounding on unaligned data and multi-form sentences. This challenging task requires to capture critical object relations to identify the queried target. However, existing approaches cannot distinguish notable objects and remain in ineffective relation modeling between unnecessary objects. Thus, we propose a novel object-aware multi-branch relation network for object-aware relation discovery. Concretely, we first devise multiple branches to develop object-aware region modeling, where each branch focuses on a crucial object mentioned in the sentence. We then propose multi-branch relation reasoning to capture critical object relationships between the main branch and auxiliary branches. Moreover, we apply a diversity loss to make each branch only pay attention to its corresponding object and boost multi-branch learning. The extensive experiments show the effectiveness of our proposed method.

The minimum connected dominating set (MCDS) problem is an important extension of the minimum dominating set problem, with wide applications, especially in wireless networks. Despite its practical importance, there are few works on solving MCDS for massive graphs, mainly due to the complexity of maintaining connectivity. In this paper, we propose two novel ideas, and develop a new local search algorithm for MCDS called NuCDS. First, a hybrid dynamic connectivity maintenance method is designed to switch alternately between a novel fast connectivity maintenance method based on spanning tree and its previous counterpart. Second, we define a new vertex property called \emph{safety} to make the algorithm more considerate when selecting vertices. Experiments show that NuCDS significantly outperforms the state-of-the-art MCDS algorithms on both massive graphs and classic benchmarks.

Many real-world problems can often be cast as the optimization of DR-submodular functions defined over a convex domain. These functions play an important role with applications in many areas of applied mathematics, such as machine learning, computer vision, operation research, communication systems or economics. In addition, they capture a subclass of non-convex optimization that provides both practical and theoretical guarantees. In this paper, we show that for maximizing non-monotone DR-submodular functions over a general convex set (such as up-closed convex sets, conic convex set, etc) the Frank-Wolfe algorithm achieves an approximation guarantee which depends on the convex set. To the best of our knowledge, this is the first approximation guarantee. Finally we benchmark our algorithm on problems arising in machine learning domain with the real-world datasets.

Most image captioning models are autoregressive, i.e. they generate each word by conditioning on previously generated words, which leads to heavy latency during inference. Recently, non-autoregressive decoding has been proposed in machine translation to speed up the inference time by generating all words in parallel. Typically, these models use the word-level cross-entropy loss to optimize each word independently. However, such a learning process fails to consider the sentence-level consistency, thus resulting in inferior generation quality of these non-autoregressive models. In this paper, we propose a Non-Autoregressive Image Captioning (NAIC) model with a novel training paradigm: Counterfactuals-critical Multi-Agent Learning (CMAL). CMAL formulates NAIC as a multi-agent reinforcement learning system where positions in the target sequence are viewed as agents that learn to cooperatively maximize a sentence-level reward. Besides, we propose to utilize massive unlabeled images to boost captioning performance. Extensive experiments on MSCOCO image captioning benchmark show that our NAIC model achieves a performance comparable to state-of-the-art autoregressive models, while brings 13.9x decoding speedup.

Recommender systems often involve multi-aspect factors. For example, when shopping for shoes online, consumers usually look through their images, ratings, and product's reviews before making their decisions. To learn multi-aspect factors, many context-aware models have been developed based on tensor factorizations. However, existing models assume multilinear structures in the tensor data, thus failing to capture nonlinear feature interactions. To fill this gap, we propose a novel nonlinear tensor machine, which combines deep neural networks and tensor algebra to capture nonlinear interactions among multi-aspect factors. We further consider adversarial learning to assist the training of our model. Extensive experiments demonstrate the effectiveness of the proposed model.

Multi-Criteria Decision Making (MCDM) aims at modelling expert preferences and assisting decision makers in identifying options best accommodating expert criteria. An instance of MCDM model, the Choquet integral is widely used in real-world applications, due to its ability to capture interactions between criteria while retaining interpretability. Aimed at a better scalability and modularity, hierarchical Choquet integrals involve intermediate aggregations of the interacting criteria, at the cost of a more complex elicitation. The paper presents a machine learning-based approach for the automatic identification of hierarchical MCDM models, composed of 2-additive Choquet integral aggregators and of marginal utility functions on the raw features from data reflecting expert preferences. The proposed NEUR-HCI framework relies on a specific neural architecture, enforcing by design the Choquet model constraints and supporting its end-to-end training. The empirical validation of NEUR-HCI on real-world and artificial benchmarks demonstrates the merits of the approach compared to state-of-art baselines.

Neural machine translation (NMT) generates the next target token given as input the previous ground truth target tokens during training while the previous generated target tokens during inference, which causes discrepancy between training and inference as well as error propagation, and affects the translation accuracy. In this paper, we introduce an error correction mechanism into NMT, which corrects the error information in the previous generated tokens to better predict the next token. Specifically, we introduce two-stream self-attention from XLNet into NMT decoder, where the query stream is used to predict the next token, and meanwhile the content stream is used to correct the error information from the previous predicted tokens. We leverage scheduled sampling to simulate the prediction errors during training. Experiments on three IWSLT translation datasets and two WMT translation datasets demonstrate that our method achieves improvements over Transformer baseline and scheduled sampling. Further experimental analyses also verify the effectiveness of our proposed error correction mechanism to improve the translation quality.

In a large-scale knowledge graph (KG), an entity is often described by a large number of triple-structured facts. Many applications require abridged versions of entity descriptions, called entity summaries. Existing solutions to entity summarization are mainly unsupervised. In this paper, we present a supervised approach NEST that is based on our novel neural model to jointly encode graph structure and text in KGs and generate high-quality diversified summaries. Since it is costly to obtain manually labeled summaries for training, our supervision is weak as we train with programmatically labeled data which may contain noise but is free of manual work. Evaluation results show that our approach significantly outperforms the state of the art on two public benchmarks.

This paper proposes a new generative model called neural belief reasoner (NBR). It differs from previous models in that it specifies a belief function rather than a probability distribution. Its implementation consists of neural networks, fuzzy-set operations and belief-function operations, and query-answering, sample-generation and training algorithms are presented. This paper studies NBR in two tasks. The first is a synthetic unsupervised-learning task, which demonstrates NBR's ability to perform multi-hop reasoning, reasoning with uncertainty and reasoning about conflicting information. The second is supervised learning: a robust MNIST classifier for 4 and 9, which is the most challenging pair of digits. This classifier needs no adversarial training, and it substantially exceeds the state of the art in adversarial robustness as measured by the L2 metric, while at the same time maintains 99.1% accuracy on natural images.

Attentional, RNN-based encoder-decoder architectures have obtained impressive performance on abstractive summarization of news articles. However, these methods fail to account for long term dependencies within the sentences of a document. This problem is exacerbated in multi-document summarization tasks such as summarizing the popular opinion in threads present in community question answering (CQA) websites such as Yahoo! Answers and Quora. These threads contain answers which often overlap or contradict each other. In this work, we present a hierarchical encoder based on structural attention to model such inter-sentence and inter-document dependencies. We set the popular pointer-generator architecture and some of the architectures derived from it as our baselines and show that they fail to generate good summaries in a multi-document setting. We further illustrate that our proposed model achieves significant improvement over the baseline in both single and multi-document summarization settings -- in the former setting, it beats the baseline by 1.31 and 7.8 ROUGE-1 points on CNN and CQA datasets, respectively; in the latter setting, the performance is further improved by 1.6 ROUGE-1 points on the CQA dataset.

We present NeurASP, a simple extension of answer set programs by embracing neural networks. By treating the neural network output as the probability distribution over atomic facts in answer set programs, NeurASP provides a simple and effective way to integrate sub-symbolic and symbolic computation. We demonstrate how NeurASP can make use of a pre-trained neural network in symbolic computation and how it can improve the neural network's perception result by applying symbolic reasoning in answer set programming. Also, NeurASP can make use of ASP rules to train a neural network better so that a neural network not only learns from implicit correlations from the data but also from the explicit complex semantic constraints expressed by the rules.

As heterogeneous networks have become increasingly ubiquitous, Heterogeneous Information Network (HIN) embedding, aiming to project nodes into a low-dimensional space while preserving the heterogeneous structure, has drawn increasing attention in recent years. Many of the existing HIN embedding methods adopt meta-path guided random walk to retain both the semantics and structural correlations between different types of nodes. However, the selection of meta-paths is still an open problem, which either depends on domain knowledge or is learned from label information. As a uniform blueprint of HIN, the network schema comprehensively embraces the high-order structure and contains rich semantics. In this paper, we make the first attempt to study network schema preserving HIN embedding, and propose a novel model named NSHE. In NSHE, a network schema sampling method is first proposed to generate sub-graphs (i.e., schema instances), and then multi-task learning task is built to preserve the heterogeneous structure of each schema instance. Besides preserving pairwise structure information, NSHE is able to retain high-order structure (i.e., network schema). Extensive experiments on three real-world datasets demonstrate that our proposed model NSHE significantly outperforms the state-of-the-art methods.

Graph convolutional networks (GCNs) have been widely used to process graph-structured data. However, existing GNN methods do not explicitly extract critical structures, which reflect the intrinsic property of a graph. In this work, we propose a novel GCN module named Neighbor Combinatorial ATtention (NCAT) to find critical structure in graph-structured data. NCAT attempts to match combinatorial neighbors with learnable patterns and assigns different weights to each combination based on the matching degree between the patterns and combinations. By stacking several NCAT modules, we can extract hierarchical structures that is helpful for down-stream tasks. Our experimental results show that NCAT achieves state-of-the-art performance on several benchmark graph classification datasets. In addition, we interpret what kind of features our model learned by visualizing the extracted critical structures.

In this paper, we study the problem of stochastic linear bandits with finite action sets. Most of existing work assume the payoffs are bounded or sub-Gaussian, which may be violated in some scenarios such as financial markets. To settle this issue, we analyze the linear bandits with heavy-tailed payoffs, where the payoffs admit finite 1+epsilon moments for some epsilon in (0,1]. Through median of means and dynamic truncation, we propose two novel algorithms which enjoy a sublinear regret bound of widetilde{O}(d^(1/2)T^(1/(1+epsilon))), where d is the dimension of contextual information and T is the time horizon. Meanwhile, we provide an Omega(d^(epsilon/(1+epsilon))T^(1/(1+epsilon))) lower bound, which implies our upper bound matches the lower bound up to polylogarithmic factors in the order of d and T when epsilon=1. Finally, we conduct numerical experiments to demonstrate the effectiveness of our algorithms and the empirical results strongly support our theoretical guarantees.

Convolutional neural networks (CNNs) are the backbones of deep learning paradigms for numerous vision tasks. Early advancements in CNN architectures are primarily driven by human expertise and elaborate design. Recently, neural architecture search (NAS) was proposed with the aim of automating the network design process and generating task-dependent architectures. This paper introduces NSGA-Net -- an evolutionary search algorithm that explores a space of potential neural network architectures in three steps, namely, a population initialization step that is based on prior-knowledge from hand-crafted architectures, an exploration step comprising crossover and mutation of architectures, and finally an exploitation step that utilizes the hidden useful knowledge stored in the entire history of evaluated neural architectures in the form of a Bayesian Network. The integration of these components allows an efficient design of architectures that are competitive and in many cases outperform both manually and automatically designed architectures on CIFAR-10 classification task. The flexibility provided from simultaneously obtaining multiple architecture choices for different compute requirements further differentiates our approach from other methods in the literature.