IJCAI 2020 论文列表

With the explosive growth of video data, video summarization which converts long-time videos to key frame sequences has become an important task in information retrieval and machine learning. Determinantal point processes (DPPs) which are elegant probabilistic models have been successfully applied to video summarization. However, existing DPP-based video summarization methods suffer from poor efficiency of outputting a specified size summary or neglecting inherent sequential nature of videos. In this paper, we propose a new model in the DPP lineage named k-SDPP in vein of sequential determinantal point processes but with fixed user specified size k. Our k-SDPP partitions sampled frames of a video into segments where each segment is with constant number of video frames. Moreover, an efficient branch and bound method (BB) considering sequential nature of the frames is provided to optimally select k frames delegating the summary from the divided segments. Experimental results show that our proposed BB method outperforms not only k-DPP and sequential DPP (seqDPP) but also the partition and Markovian assumption based methods.

Multimodal analysis that incorporates time series and textual corpora as input data sources is becoming a promising approach, especially in the financial industry. However, the main focus of such analysis has been on achieving high prediction accuracy rather than on understanding the association between the two data modalities. In this work, we address the important problem of automatically dis- covering a small set of top news articles associated with a given time series. Towards this goal, we pro- pose a novel multi-modal neural model called MSIN that jointly learns both the numerical time series and the categorical text articles in order to unearth the correlation between them. Through multiple steps of data interrelation between the two data modalities, MSIN learns to focus on a small subset of text articles that best align with the current performance in the time series. This succinct set is timely discovered and presented as recommended documents for the given time series, offering MSIN as an automated information filtering system. We empirically evaluate its performance on discovering daily top relevant news articles collected from Thomson Reuters for two given stock time series, AAPL and GOOG, over a period of seven consecutive years. The experimental results demonstrate MSIN achieves up to 84.9% and 87.2% respectively in recalling the ground truth articles, superior to SOTA algorithms that rely on conventional attention mechanisms in deep learning.

Zero-shot object detection (ZSD) has received considerable attention from the community of computer vision in recent years. It aims to simultaneously locate and categorize previously unseen objects during inference. One crucial problem of ZSD is how to accurately predict the label of each object proposal, i.e. categorizing object proposals, when conducting ZSD for unseen categories. Previous ZSD models generally relied on learning an embedding from visual space to semantic space or learning a joint embedding between semantic description and visual representation. As the features in the learned semantic space or the joint projected space tend to suffer from the hubness problem, namely the feature vectors are likely embedded to an area of incorrect labels, and thus it will lead to lower detection precision. In this paper, instead, we propose to learn a deep embedding from the semantic space to the visual space, which enables to well alleviate the hubness problem, because, compared with semantic space or joint embedding space, the distribution in visual space has smaller variance. After learning a deep embedding model, we perform $k$ nearest neighbor search in the visual space of unseen categories to determine the category of each semantic description. Extensive experiments on two public datasets show that our approach significantly outperforms the existing methods.

Obstructive sleep apnea is a serious sleep disorder that affects an estimated one billion adults worldwide. It causes breathing to repeatedly stop and start during sleep which over years increases the risk of hypertension, heart disease, stroke, Alzheimer's, and cancer. In this demo, we present Yolo4Apnea a deep learning system extending You Only Look Once (Yolo) system to detect sleep apnea events from abdominal breathing patterns in real-time enabling immediate awareness and action. Abdominal breathing is measured using a respiratory inductance plethysmography sensor worn around the stomach. The source code is available at https://github.com/simula-vias/Yolo4Apnea

We present improvements in maximum a-posteriori inference for Markov Logic, a widely used SRL formalism. Several approaches, including Cutting Plane Aggregation (CPA), perform inference through translation to Integer Linear Programs. Aggregation exploits context-specific symmetries independently of evidence and reduces the size of the program. We illustrate much more symmetries occurring in long ground clauses that are ignored by CPA and can be exploited by higher-order aggregations. We propose Full-Constraint-Aggregation, a superior algorithm to CPA which exploits the ignored symmetries via a lifted translation method and some constraint relaxations. RDBMS and heuristic techniques are involved to improve the overall performance. We introduce Xeggora as an evolutionary extension of RockIt, the query engine that uses CPA. Xeggora evaluation on real-world benchmarks shows progress in efficiency compared to RockIt especially for models with long formulas.

While Point-of-Interest (POI) recommendation has been a popular topic of study for some time, little progress has been made for understanding why and how people make their decisions for the selection of POIs. To this end, in this paper, we propose a user decision profiling framework, named PROUD, which can identify the key factors in people's decisions on choosing POIs. Specifically, we treat each user decision as a set of factors and provide a method for learning factor embeddings. A unique perspective of our approach is to identify key factors, while preserving decision structures seamlessly, via a novel scalar projection maximization objective. Exactly solving the objective is non-trivial due to a sparsity constraint. To address this, our PROUD adopts a self projection attention and an L2 regularized sparse activation to directly estimate the likelihood of each factor to be a key factor. Finally, extensive experiments on real-world data validate the advantage of PROUD in preserving user decision structures. Also, our case study indicates that the identified key decision factors can help us to provide more interpretable recommendations and analyses.

Most of today's AI technologies are geared towards mastering specific tasks performance through learning from a huge volume of data. However, less attention has still been given to make the AI understand its own purposes or be responsible socially. In this paper, a new model of agent is presented with the capacity to represent itself as a distinct individual with identity, a mind of its own, unique experiences, and social lives. In this way, the agent can interact with its surroundings and other agents seamlessly and meaningfully. A practical framework for developing an agent architecture with this model of self and self-awareness is proposed allowing self to be ascribed to an existing intelligent agent architecture in general to enable its social ability, interactivity, and co-presence with others. Possible applications are discussed with some exemplifying cases based on an implementation of a conversational agent.

Although security games have attracted intensive research attention over the past years, few existing works consider how information from local communities would affect the game. In this paper, we introduce a new player -- a strategic informant, who can observe and report upcoming attacks -- to the defender-attacker security game setting. Characterized by a private type, the informant has his utility structure that leads to his strategic behaviors. We model the game as a 3-player extensive-form game and propose a novel solution concept of Strong Stackelberg-perfect Bayesian equilibrium. To compute the optimal defender strategy, we first show that although the informant can have infinitely many types in general, the optimal defense plan can only include a finite (exponential) number of different patrol strategies. We then prove that there exists a defense plan with only a linear number of patrol strategies that achieve the optimal defender's utility, which significantly reduces the computational burden and allows us to solve the game in polynomial time using linear programming. Finally, we conduct extensive experiments to show the effect of the strategic informant and demonstrate the effectiveness of our algorithm.

Pedestrian detection at nighttime is a crucial and frontier problem in surveillance, but has not been well explored by the computer vision and artificial intelligence communities. Most of existing methods detect pedestrians under favorable lighting conditions (e.g. daytime) and achieve promising performances. In contrast, they often fail under unstable lighting conditions (e.g. nighttime). Night is a critical time for criminal suspects to act in the field of security. The existing nighttime pedestrian detection dataset is captured by a car camera, specially designed for autonomous driving scenarios. The dataset for nighttime surveillance scenario is still vacant. There are vast differences between autonomous driving and surveillance, including viewpoint and illumination. In this paper, we build a novel pedestrian detection dataset from the nighttime surveillance aspect: NightSurveillance1. As a benchmark dataset for pedestrian detection at nighttime, we compare the performances of state-of-the-art pedestrian detectors and the results reveal that the methods cannot solve all the challenging problems of NightSurveillance. We believe that NightSurveillance can further advance the research of pedestrian detection, especially in the field of surveillance security at nighttime.

Graph Neural Networks (GNNs) have been shown to be powerful in a wide range of graph-related tasks. While there exists various GNN models, a critical common ingredient is neighborhood aggregation, where the embedding of each node is updated by referring to the embedding of its neighbors. This paper aims to provide a better understanding of this mechanisms by asking the following question: Is neighborhood aggregation always necessary and beneficial? In short, the answer is no. We carve out two conditions under which neighborhood aggregation is not helpful: (1) when a node's neighbors are highly dissimilar and (2) when a node's embedding is already similar with that of its neighbors. We propose novel metrics that quantitatively measure these two circumstances and integrate them into an Adaptive-layer module. Our experiments show that allowing for node-specific aggregation degrees have significant advantage over current GNNs.

In the standard model of committee selection, we are given a set of ordinal votes over a set of candidates and a desired committee size, and the task is to select a committee that relates to the given votes. Motivated by possible interactions and dependencies between candidates, we study a generalization of committee selection in which the candidates are connected via a network and the task is to select a committee that relates to the given votes while also satisfy certain properties with respect to this candidate network. To accommodate certain correspondences to the voter preferences, we consider three standard voting rules (in particular, $k$-Borda, Chamberlin-Courant, and Gehrlein stability); to model different aspects of interactions and dependencies between candidates, we consider two graph properties (in particular, Independent Set and Connectivity). We study the parameterized complexity of the corresponding combinatorial problems and discuss certain implications of our algorithmic results.

Multi-view, Multi-instance, and Multi-label Learning (M3L) can model complex objects (bags), which are represented with different feature views, made of diverse instances, and annotated with discrete non-exclusive labels. Existing M3L approaches assume a complete correspondence between bags and views, and also assume a complete annotation for training. However, in practice, neither the correspondence between bags, nor the bags' annotations are complete. To tackle such a weakly-supervised M3L task, a solution called WSM3L is introduced. WSM3L adapts multimodal dictionary learning to learn a shared dictionary (representational space) across views and individual encoding vectors of bags for each view. The label similarity and feature similarity of encoded bags are jointly used to match bags across views. In addition, it replenishes the annotations of a bag based on the annotations of its neighborhood bags, and introduces a dispatch and aggregation term to dispatch bag-level annotations to instances and to reversely aggregate instance-level annotations to bags. WSM3L unifies these objectives and processes in a joint objective function to predict the instance-level and bag-level annotations in a coordinated fashion, and it further introduces an alternative solution for the objective function optimization. Extensive experimental results show the effectiveness of WSM3L on benchmark datasets.

To extract crucial local features and enhance the complementary relation between local and global features, this paper proposes a Weakly Supervised Local-Global Relation Network (WS-LGRN), which uses the attention mechanism to deal with part location and feature fusion problems. Firstly, the Attention Map Generator quickly finds the local regions-of-interest under the supervision of image-level labels. Secondly, bilinear attention pooling is employed to generate and refine local features. Thirdly, Relational Reasoning Unit is designed to model the relation among all features before making classification. The weighted fusion mechanism in the Relational Reasoning Unit makes the model benefit from the complementary advantages between different features. In addition, contrastive losses are introduced for local and global features to increase the inter-class dispersion and intra-class compactness at different granularities. Experiments on lab-controlled and real-world facial expression dataset show that WS-LGRN achieves state-of-the-art performance, which demonstrates its superiority in FER.

Significant progress has been made recently in developing few-shot object segmentation methods. Learning is shown to be successful in few-shot segmentation settings, using pixel-level, scribbles and bounding box supervision. This paper takes another approach, i.e., only requiring image-level label for few-shot object segmentation. We propose a novel multi-modal interaction module for few-shot object segmentation that utilizes a co-attention mechanism using both visual and word embedding. Our model using image-level labels achieves 4.8% improvement over previously proposed image-level few-shot object segmentation. It also outperforms state-of-the-art methods that use weak bounding box supervision on PASCAL-5^i. Our results show that few-shot segmentation benefits from utilizing word embeddings, and that we are able to perform few-shot segmentation using stacked joint visual semantic processing with weak image-level labels. We further propose a novel setup, Temporal Object Segmentation for Few-shot Learning (TOSFL) for videos. TOSFL can be used on a variety of public video data such as Youtube-VOS, as demonstrated in both instance-level and category-level TOSFL experiments.

Word embeddings are known to exhibit stereotypical biases towards gender, race, religion, among other criteria. Severa fairness metrics have been proposed in order to automatically quantify these biases. Although all metrics have a similar objective, the relationship between them is by no means clear. Two issues that prevent a clean comparison is that they operate with different inputs, and that their outputs are incompatible with each other. In this paper we propose WEFE, the word embeddings fairness evaluation framework, to encapsulate, evaluate and compare fairness metrics. Our framework needs a list of pre-trained embeddings and a set of fairness criteria, and it is based on checking correlations between fairness rankings induced by these criteria. We conduct a case study showing that rankings produced by existing fairness methods tend to correlate when measuring gender bias. This correlation is considerably less for other biases like race or religion. We also compare the fairness rankings with an embedding benchmark showing that there is no clear correlation between fairness and good performance in downstream tasks.

Finance is a particularly challenging application area for deep learning models due to low noise-to-signal ratio, non-stationarity, and partial observability. Non-deliverable-forwards (NDF), a derivatives contract used in foreign exchange (FX) trading, presents additional difficulty in the form of long-term planning required for an effective selection of start and end date of the contract. In this work, we focus on tackling the problem of NDF position length selection by leveraging high-dimensional sequential data consisting of spot rates, technical indicators and expert tenor patterns. To this end, we curate, analyze and release a dataset from the Depository Trust & Clearing Corporation (DTCC) NDF data that includes a comprehensive list of NDF volumes and daily spot rates for 64 FX pairs. We introduce WaveATTentionNet (WATTNet), a novel temporal convolution (TCN) model for spatio-temporal modeling of highly multivariate time series, and validate it across NDF markets with varying degrees of dissimilarity between the training and test periods in terms of volatility and general market regimes. The proposed method achieves a significant positive return on investment (ROI) in all NDF markets under analysis, outperforming recurrent and classical baselines by a wide margin. Finally, we propose two orthogonal interpretability approaches to verify noise robustness and detect the driving factors of the learned tenor selection strategy.

Using a convolutional neural network to build visual encoding and decoding models of the human brain is a good starting point for the study on relationship between deep learning and human visual cognitive mechanism. However, related studies have not fully considered their differences. In this paper, we assume that only a portion of neural network features is directly related to human brain signals, which we call shared features. In the encoding process, we extract shared features from the lower and higher layers of the neural network, and then build a non-negative sparse map to predict brain activities. In the decoding process, we use back-propagation to reconstruct visual stimuli, and use dictionary learning and a deep image prior to improve the robustness and accuracy of the algorithm. Experiments on a public fMRI dataset confirm the rationality of the encoding models, and comparing with a recently proposed method, our reconstruction results obtain significantly higher accuracy.

This paper presents a new video question answering task on screencast tutorials. We introduce a dataset including question, answer and context triples from the tutorial videos for a software. Unlike other video question answering works, all the answers in our dataset are grounded to the domain knowledge base. An one-shot recognition algorithm is designed to extract the visual cues, which helps enhance the performance of video question answering. We also propose several baseline neural network architectures based on various aspects of video contexts from the dataset. The experimental results demonstrate that our proposed models significantly improve the question answering performances by incorporating multi-modal contexts and domain knowledge.

The p-center problem consists of choosing p centers from a set of candidates to minimize the maximum cost between any client and its assigned facility. In this paper, we transform the p-center problem into a series of set covering subproblems, and propose a vertex weighting-based tabu search (VWTS) algorithm to solve them. The proposed VWTS algorithm integrates distinguishing features such as a vertex weighting technique and a tabu search strategy to help the search to jump out of the local optima. Computational experiments on 138 most commonly used benchmark instances show that VWTS is highly competitive comparing to the state-of-the-art methods in spite of its simplicity. As a well-known NP-hard problem which has already been studied for over half a century, it is a challenging task to break the records on these classic datasets. Yet VWTS improves the best known results for 14 out of 54 large instances, and matches the optimal results for all remaining 84 ones. In addition, the computational time taken by VWTS is much shorter than other algorithms in the literature.