IJCAI 2020 Papers

Learning representations in an unsupervised or self-supervised manner is a growing area of research. Current approaches in representation learning seek to maximize the mutual information between the learned representation and original data. One of the most popular ways to estimate mutual information (MI) is based on Noise Contrastive Estimation (NCE). This MI estimate exhibits low variance, but it is upper-bounded by log(N), where N is the number of samples. In an ideal scenario, we would use the entire dataset to get the most accurate estimate. However, using such a large number of samples is computationally prohibitive. Our proposed solution is to decouple the upper-bound for the MI estimate from the sample size. Instead, we estimate the partition function of the NCE loss function for the entire dataset using importance sampling (IS). In this paper, we use locality-sensitive hashing (LSH) as an adaptive sampler and propose an unbiased estimator that accurately approximates the partition function in sub-linear (near-constant) time. The samples are correlated and non-normalized, but the derived estimator is unbiased without any assumptions. We show that our LSH sampling estimate provides a superior bias-variance trade-off when compared to other state-of-the-art approaches.

Multivariate probability calibration is the problem of predicting class membership probabilities from classification scores of multiple classifiers. To achieve better performance, the calibrating function is often required to be coordinate-wise non-decreasing; that is, for every classifier, the higher the score, the higher the probability of the class labeling being positive. To this end, we propose a multivariate regression method based on shape-restricted Bernstein polynomials. This method is universally flexible: it can approximate any continuous calibrating function with any specified error, as the polynomial degree increases to infinite. Moreover, it is universally consistent: the estimated calibrating function converges to any continuous calibrating function, as the training size increases to infinity. Our empirical study shows that the proposed method achieves better calibrating performance than benchmark methods.

The R, G and B channels of a color image generally have different noise statistical properties or noise strengths. It is thus problematic to apply grayscale image denoising algorithms to color image denoising. In this paper, based on the non-local self-similarity of an image and the different noise strength across each channel, we propose a MultiChannel Weighted Schatten p-Norm Minimization (MCWSNM) model for RGB color image denoising. More specifically, considering a small local RGB patch in a noisy image, we first find its nonlocal similar cubic patches in a search window with an appropriate size. These similar cubic patches are then vectorized and grouped to construct a noisy low-rank matrix, which can be recovered using the Schatten p-norm minimization framework. Moreover, a weight matrix is introduced to balance each channel’s contribution to the final denoising results. The proposed MCWSNM can be solved via the alternating direction method of multipliers. Convergence property of the proposed method are also theoretically analyzed . Experiments conducted on both synthetic and real noisy color image datasets demonstrate highly competitive denoising performance, outperforming comparison algorithms, including several methods based on neural networks.

Stock Trend Prediction(STP) has drawn wide attention from various fields, especially Artificial Intelligence. Most previous studies are single-scale oriented which results in information loss from a multi-scale perspective. In fact, multi-scale behavior is vital for making intelligent investment decisions. A mature investor will thoroughly investigate the state of a stock market at various time scales. To automatically learn the multi-scale information in stock data, we propose a Multi-scale Two-way Deep Neural Network. It learns multi-scale patterns from two types of scale-information, wavelet-based and downsampling-based, by eXtreme Gradient Boosting and Recurrent Convolutional Neural Network, respectively. After combining the learned patterns from the two-way, our model achieves state-of-the-art performance on FI-2010 and CSI-2016, where the latter is our published long-range stock dataset to help future studies for STP task. Extensive experimental results on the two datasets indicate that multi-scale information can significantly improve the STP performance and our model is superior in capturing such information.

Information theoretical based methods have attracted a great attention in recent years, and gained promising results to deal with multi-label data with high dimensionality. However, most of the existing methods are either directly transformed from heuristic single-label feature selection methods or inefficient in exploiting labeling information. Thus, they may not be able to get an optimal feature selection result shared by multiple labels. In this paper, we propose a general global optimization framework, in which feature relevance, label relevance (i.e., label correlation), and feature redundancy are taken into account, thus facilitating multi-label feature selection. Moreover, the proposed method has an excellent mechanism for utilizing inherent properties of multi-label learning. Specially, we provide a formulation to extend the proposed method with label-specific features. Empirical studies on twenty multi-label data sets reveal the effectiveness and efficiency of the proposed method. Our implementation of the proposed method is available online at: https://jiazhang-ml.pub/GRRO-master.zip.

Multi-hop reading comprehension across multiple documents attracts much attentions recently. In this paper, we propose a novel approach to tackle this multi-hop reading comprehension problem. Inspired by the human reasoning processing, we introduce a path-based graph with reasoning paths which extracted from supporting documents. The path-based graph can combine both the idea of the graph-based and path-based approaches, so it is better for multi-hop reasoning. Meanwhile, we propose Gated-GCN to accumulate evidences on the path-based graph, which contains a new question-aware gating mechanism to regulate the usefulness of information propagating across documents and add question information during reasoning. We evaluate our approach on WikiHop dataset, and our approach achieves the the-state-of-art accuracy against previous published approaches. Especially, our ensemble model surpasses the human performance by 4.2%.

Brain functional connectivity analysis on fMRI data could improve the understanding of human brain function. However, due to the influence of the inter-subject variability and the heterogeneity across subjects, previous methods of functional connectivity analysis are often insufficient in capturing disease-related representation so that decreasing disease diagnosis performance. In this paper, we first propose a new multi-graph fusion framework to fine-tune the original representation derived from Pearson correlation analysis, and then employ L1-SVM on fine-tuned representations to conduct joint brain region selection and disease diagnosis for avoiding the issue of the curse of dimensionality on high-dimensional data. The multi-graph fusion framework automatically learns the connectivity number for every node (i.e., brain region) and integrates all subjects in a unified framework to output homogenous and discriminative representations of all subjects. Experimental results on two real data sets, i.e., fronto-temporal dementia (FTD) and obsessive-compulsive disorder (OCD), verified the effectiveness of our proposed framework, compared to state-of-the-art methods.

The goal of few-shot image recognition is to distinguish different categories with only one or a few training samples. Previous works of few-shot learning mainly work on general object images. And current solutions usually learn a global image representation from training tasks to adapt novel tasks. However, fine-gained categories are distinguished by subtle and local parts, which could not be captured by global representations effectively. This may hinder existing few-shot learning approaches from dealing with fine-gained categories well. In this work, we propose a multi-attention meta-learning (MattML) method for few-shot fine-grained image recognition (FSFGIR). Instead of using only base learner for general feature learning, the proposed meta-learning method uses attention mechanisms of the base learner and task learner to capture discriminative parts of images. The base learner is equipped with two convolutional block attention modules (CBAM) and a classifier. The two CBAM can learn diverse and informative parts. And the initial weights of classifier are attended by the task learner, which gives the classifier a task-related sensitive initialization. For adaptation, the gradient-based meta-learning approach is employed by updating the parameters of two CBAM and the attended classifier, which facilitates the updated base learner to adaptively focus on discriminative parts. We experimentally analyze the different components of our method, and experimental results on four benchmark datasets demonstrate the effectiveness and superiority of our method.

The increasing amount of urban data enable us to investigate urban dynamics, assist urban planning, and eventually, make our cities more livable and sustainable. In this paper, we focus on learning an embedding space from urban data for urban regions. For the first time, we propose a multi-view joint learning model to learn comprehensive and representative urban region embeddings. We first model different types of region correlations based on both human mobility and inherent region properties. Then, we apply a graph attention mechanism in learning region representations from each view of the built correlations. Moreover, we introduce a joint learning module that boosts the region embedding learning by sharing cross-view information and fuses multi-view embeddings by learning adaptive weights. Finally, we exploit the learned embeddings in the downstream applications of land usage classification and crime prediction in urban areas with real-world data. Extensive experiment results demonstrate that by exploiting our proposed joint learning model, the performance is improved by a large margin on both tasks compared with the state-of-the-art methods.

Graph neural networks (GNNs) have made considerable achievements in processing graph-structured data. However, existing methods can not allocate learnable weights to different nodes in the neighborhood and lack of robustness on account of neglecting both node attributes and graph reconstruction. Moreover, most of multi-view GNNs mainly focus on the case of multiple graphs, while designing GNNs for solving graph-structured data of multi-view attributes is still under-explored. In this paper, we propose a novel Multi-View Attribute Graph Convolution Networks (MAGCN) model for the clustering task. MAGCN is designed with two-pathway encoders that map graph embedding features and learn the view-consistency information. Specifically, the first pathway develops multi-view attribute graph attention networks to reduce the noise/redundancy and learn the graph embedding features for each multi-view graph data. The second pathway develops consistent embedding encoders to capture the geometric relationship and probability distribution consistency among different views, which adaptively finds a consistent clustering embedding space for multi-view attributes. Experiments on three benchmark graph datasets show the superiority of our method compared with several state-of-the-art algorithms.

Applying multi-scale representations leads to consistent performance improvements on a wide range of image recognition tasks. However, with the addition of the temporal dimension in video domain, directly obtaining layer-wise multi-scale spatial-temporal features will add a lot extra computational cost. In this work, we propose a novel and efficient Multi-Scale Spatial-Temporal Integration Convolutional Tube (MSTI) aiming at achieving accurate recognition of actions with lower computational cost. It firstly extracts multi-scale spatial and temporal features through the multi-scale convolution block. Considering the interaction of different-scales representations and the interaction of spatial appearance and temporal motion, we employ the cross-scale attention weighted blocks to perform feature recalibration by integrating multi-scale spatial and temporal features. An end-to-end deep network, MSTI-Net, is also presented based on the proposed MSTI tube for human action recognition. Extensive experimental results show that our MSTI-Net significantly boosts the performance of existing convolution networks and achieves state-of-the-art accuracy on three challenging benchmarks, i.e., UCF-101, HMDB-51 and Kinetics-400, with much fewer parameters and FLOPs.

In this paper, we introduce Transformer to the time-domain methods for single-channel speech separation. Transformer has the potential to boost speech separation performance because of its strong sequence modeling capability. However, its computational complexity, which grows quadratically with the sequence length, has made it largely inapplicable to speech applications. To tackle this issue, we propose a novel variation of Transformer, named multi-scale group Transformer (MSGT). The key ideas are group self-attention, which significantly reduces the complexity, and multi-scale fusion, which retains Transform's ability to capture long-term dependency. We implement two versions of MSGT with different complexities, and apply them to a well-known time-domain speech separation method called Conv-TasNet. By simply replacing the original temporal convolutional network (TCN) with MSGT, our approach called MSGT-TasNet achieves a large gain over Conv-TasNet on both WSJ0-2mix and WHAM! benchmarks. Without bells and whistles, the performance of MSGT-TasNet is already on par with the SOTA methods.

In full-knowledge multi-robot adversarial patrolling, a group of robots have to detect an adversary who knows the robots' strategy. The adversary can easily take advantage of any deterministic patrolling strategy, which necessitates the employment of a randomised strategy. While the Markov decision process has been the dominant methodology in computing the penetration detection probabilities, we apply enumerative combinatorics to characterise the penetration detection probabilities. It allows us to provide the closed formulae of these probabilities and facilitates characterising optimal random defence strategies. Comparing to iteratively updating the Markov transition matrices, our methods significantly reduces the time and space complexity of solving the problem. We use this method to tackle four penetration configurations.

Contextual bandit is a classic multi-armed bandit setting, where side information (i.e., context) is available before arm selection. A standard assumption is that exact contexts are perfectly known prior to arm selection and only single feedback is returned. In this work, we focus on multi-feedback bandit learning with probabilistic contexts, where a bundle of contexts are revealed to the agent along with their corresponding probabilities at the beginning of each round. This models such scenarios as where contexts are drawn from the probability output of a neural network and the reward function is jointly determined by multiple feedback signals. We propose a kernelized learning algorithm based on upper confidence bound to choose the optimal arm in reproducing kernel Hilbert space for each context bundle. Moreover, we theoretically establish an upper bound on the cumulative regret with respect to an oracle that knows the optimal arm given probabilistic contexts, and show that the bound grows sublinearly with time. Our simula- tion on machine learning model recommendation further validates the sub-linearity of our cumulative regret and demonstrates that our algorithm outper- forms the approach that selects arms based on the most probable context.

In the Multi-Agent Meeting problem (MAM), the task is to find a meeting location for multiple agents, as well as a path for each agent to that location. In this paper, we introduce MM*, a Multi-Directional Heuristic Search algorithm that finds the optimal meeting location under different cost functions. MM* generalizes the Meet in the Middle (MM) bidirectional search algorithm to the case of finding an optimal meeting location for multiple agents. Several admissible heuristics are proposed, and experiments demonstrate the benefits of MM*.

Networked data often demonstrate the Pareto principle (i.e., 80/20 rule) with skewed class distributions, where most vertices belong to a few majority classes and minority classes only contain a handful of instances. When presented with imbalanced class distributions, existing graph embedding learning tends to bias to nodes from majority classes, leaving nodes from minority classes under-trained. In this paper, we propose Dual-Regularized Graph Convolutional Networks (DR-GCN) to handle multi-class imbalanced graphs, where two types of regularization are imposed to tackle class imbalanced representation learning. To ensure that all classes are equally represented, we propose a class-conditioned adversarial training process to facilitate the separation of labeled nodes. Meanwhile, to maintain training equilibrium (i.e., retaining quality of fit across all classes), we force unlabeled nodes to follow a similar latent distribution to the labeled nodes by minimizing their difference in the embedding space. Experiments on real-world imbalanced graphs demonstrate that DR-GCN outperforms the state-of-the-art methods in node classification, graph clustering, and visualization.

The classification of graph-structured data has be-come increasingly crucial in many disciplines. It has been observed that the implicit or explicit hierarchical community structures preserved in real-world graphs could be useful for downstream classification applications. A straightforward way to leverage the hierarchical structure is to make use the pooling algorithms to cluster nodes into fixed groups, and shrink the input graph layer by layer to learn the pooled graphs.However, the pool shrinking discards the graph details to make it hard to distinguish two non-isomorphic graphs, and the fixed clustering ignores the inherent multiple characteristics of nodes. To compensate the shrinking loss and learn the various nodes’ characteristics, we propose the multi-channel graph neural networks (MuchGNN). Motivated by the underlying mechanisms developed in convolutional neural networks, we define the tailored graph convolutions to learn a series of graph channels at each layer, and shrink the graphs hierarchically to en-code the pooled structures. Experimental results on real-world datasets demonstrate the superiority of MuchGNN over the state-of-the-art methods.

Fact-based Visual Question Answering (FVQA) requires external knowledge beyond the visible content to answer questions about an image. This ability is challenging but indispensable to achieve general VQA. One limitation of existing FVQA solutions is that they jointly embed all kinds of information without fine-grained selection, which introduces unexpected noises for reasoning the final answer. How to capture the question-oriented and information-complementary evidence remains a key challenge to solve the problem. In this paper, we depict an image by a multi-modal heterogeneous graph, which contains multiple layers of information corresponding to the visual, semantic and factual features. On top of the multi-layer graph representations, we propose a modality-aware heterogeneous graph convolutional network to capture evidence from different layers that is most relevant to the given question. Specifically, the intra-modal graph convolution selects evidence from each modality and cross-modal graph convolution aggregates relevant information across different graph layers. By stacking this process multiple times, our model performs iterative reasoning across three modalities and predicts the optimal answer by analyzing all question-oriented evidence. We achieve a new state-of-the-art performance on the FVQA task and demonstrate the effectiveness and interpretability of our model with extensive experiments.

The knowledge acquisition bottleneck strongly affects the creation of multilingual sense-annotated data, hence limiting the power of supervised systems when applied to multilingual Word Sense Disambiguation. In this paper, we propose a semi-supervised approach based upon a novel label propagation scheme, which, by jointly leveraging contextualized word embeddings and the multilingual information enclosed in a knowledge base, projects sense labels from a high-resource language, i.e., English, to lower-resourced ones. Backed by several experiments, we provide empirical evidence that our automatically created datasets are of a higher quality than those generated by other competitors and lead a supervised model to achieve state-of-the-art performances in all multilingual Word Sense Disambiguation tasks. We make our datasets available for research purposes at https://github.com/SapienzaNLP/mulan.