IJCAI 2020 Papers

As a matter of fact, it is usually taken for granted that the occurrence of unauthorized behaviors is necessary for the fraud detection in online payment services. However, we seek to break this stereotype in this work. We strive to design an ex-ante anti-fraud method that can work before unauthorized behaviors occur. The feasibility of our solution is supported by the cooperation of a characteristic and a finding in online payment fraud scenarios: The well-recognized characteristic is that online payment frauds are mostly caused by account compromise. Our finding is that account theft is indeed predictable based on users' high-risk behaviors, without relying on the behaviors of thieves. Accordingly, we propose an account risk prediction scheme to realize the ex-ante fraud detection. It takes in an account's historical transaction sequence, and outputs its risk score. The risk score is then used as an early evidence of whether a new transaction is fraudulent or not, before the occurrence of the new transaction. We examine our method on a real-world B2C transaction dataset from a commercial bank. Experimental results show that the ex-ante detection method can prevent more than 80\% of the fraudulent transactions before they actually occur. When the proposed method is combined with an interim detection to form a real-time anti-fraud system, it can detect more than 94\% of fraudulent transactions while maintaining a very low false alarm rate (less than 0.1\%).

Human labeling of data can be very time-consuming and expensive, yet, in many cases it is critical for the success of the learning process. In order to minimize human labeling efforts, we propose a novel active learning solution that does not rely on existing sources of unlabeled data. It uses a small amount of labeled data as the core set for the synthesis of useful membership queries (MQs) — unlabeled instances generated by an algorithm for human labeling. Our solution uses modification operators, functions that modify instances to some extent. We apply the operators on a small set of instances (core set), creating a set of new membership queries. Using this framework, we look at the instance space as a search space and apply search algorithms in order to generate new examples highly relevant to the learner. We implement this framework in the textual domain and test it on several text classification tasks and show improved classifier performance as more MQs are labeled and incorporated into the training set. To the best of our knowledge, this is the first work on membership queries in the textual domain.

Arbitrary shape text detection in natural scenes is an extremely challenging task. Unlike existing text detection approaches that only perceive texts based on limited feature representations, we propose a novel framework, namely TextFuseNet, to exploit the use of richer features fused for text detection. More specifically, we propose to perceive texts from three levels of feature representations, i.e., character-, word- and global-level, and then introduce a novel text representation fusion technique to help achieve robust arbitrary text detection. The multi-level feature representation can adequately describe texts by dissecting them into individual characters while still maintaining their general semantics. TextFuseNet then collects and merges the texts’ features from different levels using a multi-path fusion architecture which can effectively align and fuse different representations. In practice, our proposed TextFuseNet can learn a more adequate description of arbitrary shapes texts, suppressing false positives and producing more accurate detection results. Our proposed framework can also be trained with weak supervision for those datasets that lack character-level annotations. Experiments on several datasets show that the proposed TextFuseNet achieves state-of-the-art performance. Specifically, we achieve an F-measure of 94.3% on ICDAR2013, 92.1% on ICDAR2015, 87.1% on Total-Text and 86.6% on CTW-1500, respectively.

Text style transfer pursues altering the style of a sentence while remaining its main content unchanged. Due to the lack of parallel corpora, most recent work focuses on unsupervised methods and has achieved noticeable progress. Nonetheless, the intractability of completely disentangling content from style for text leads to a contradiction of content preservation and style transfer accuracy. To address this problem, we propose a style instance supported method, StyIns. Instead of representing styles with embeddings or latent variables learned from single sentences, our model leverages the generative flow technique to extract underlying stylistic properties from multiple instances of each style, which form a more discriminative and expressive latent style space. By combining such a space with the attention-based structure, our model can better maintain the content and simultaneously achieve high transfer accuracy. Furthermore, the proposed method can be flexibly extended to semi-supervised learning so as to utilize available limited paired data. Experiments on three transfer tasks, sentiment modification, formality rephrasing, and poeticness generation, show that StyIns obtains a better balance between content and style, outperforming several recent baselines.

Label enhancement (LE) is a procedure of recovering the label distributions from the logical labels in the multi-label data, the purpose of which is to better represent and mine the label ambiguity problem through the form of label distribution. Existing LE work mainly concentrates on how to leverage the topological information of the feature space and the correlation among the labels, and all are based on single view data. In view of the fact that there are many multi-view data in real-world applications, which can provide richer semantic information from different perspectives, this paper first presents a multi-view label enhancement problem and proposes a tensor-based multi-view label enhancement method, named TMV-LE. Firstly, we introduce the tensor factorization to get the common subspace which contains the high-order relationships among different views. Secondly, we use the common representation and multiple views to jointly mine a more comprehensive topological structure in the dataset. Finally, the topological structure of the feature space is migrated to the label space to get the label distributions. Extensive comparative studies validate that the performance of multi-view multi-label learning can be improved significantly with TMV-LE.

Time series prediction is an important problem in machine learning. Previous methods for time series prediction did not involve additional information. With a lot of dynamic knowledge graphs available, we can use this additional information to predict the time series better. Recently, there has been a focus on the application of deep representation learning on dynamic graphs. These methods predict the structure of the graph by reasoning over the interactions in the graph at previous time steps. In this paper, we propose a new framework to incorporate the information from dynamic knowledge graphs for time series prediction. We show that if the information contained in the graph and the time series data are closely related, then this inter-dependence can be used to predict the time series with improved accuracy. Our framework, DArtNet, learns a static embedding for every node in the graph as well as a dynamic embedding which is dependent on the dynamic attribute value (time-series). Then it captures the information from the neighborhood by taking a relation specific mean and encodes the history information using RNN. We jointly train the model link prediction and attribute prediction. We evaluate our method on five specially curated datasets for this problem and show a consistent improvement in time series prediction results. We release the data and code of model DArtNet for future research.

Video inpainting aims to synthesize visually pleasant and temporally consistent content in missing regions of video. Due to a variety of motions across different frames, it is highly challenging to utilize effective temporal information to recover videos. Existing deep learning based methods usually estimate optical flow to align frames and thereby exploit useful information between frames. However, these methods tend to generate artifacts once the estimated optical flow is inaccurate. To alleviate above problem, we propose a novel end-to-end Temporal Adaptive Alignment Network(TAAN) for video inpainting. The TAAN aligns reference frames with target frame via implicit motion estimation at a feature level and then reconstruct target frame by taking the aggregated aligned reference frame features as input. In the proposed network, a Temporal Adaptive Alignment (TAA) module based on deformable convolutions is designed to perform temporal alignment in a local, dense and adaptive manner. Both quantitative and qualitative evaluation results show that our method significantly outperforms existing deep learning based methods.

Math word problem (MWP) is challenging due to the limitation in training data where only one “standard” solution is available. MWP models often simply fit this solution rather than truly understand or solve the problem. The generalization of models (to diverse word scenarios) is thus limited. To address this problem, this paper proposes a novel approach, TSN-MD, by leveraging the teacher network to integrate the knowledge of equivalent solution expressions and then to regularize the learning behavior of the student network. In addition, we introduce the multiple-decoder student network to generate multiple candidate solution expressions by which the final answer is voted. In experiments, we conduct extensive comparisons and ablative studies on two large-scale MWP benchmarks, and show that using TSN-MD can surpass the state-of-the-art works by a large margin. More intriguingly, the visualization results demonstrate that TSN-MD not only produces correct final answers but also generates diverse equivalent expressions of the solution.

Non-autoregressive translation (NAT) achieves faster inference speed but at the cost of worse accuracy compared with autoregressive translation (AT). Since AT and NAT can share model structure and AT is an easier task than NAT due to the explicit dependency on previous target-side tokens, a natural idea is to gradually shift the model training from the easier AT task to the harder NAT task. To smooth the shift from AT training to NAT training, in this paper, we introduce semi-autoregressive translation (SAT) as intermediate tasks. SAT contains a hyperparameter k, and each k value defines a SAT task with different degrees of parallelism. Specially, SAT covers AT and NAT as its special cases: it reduces to AT when k=1 and to NAT when k=N (N is the length of target sentence). We design curriculum schedules to gradually shift k from 1 to N, with different pacing functions and number of tasks trained at the same time. We called our method as task-level curriculum learning for NAT (TCL-NAT). Experiments on IWSLT14 De-En, IWSLT16 En-De, WMT14 En-De and De-En datasets show that TCL-NAT achieves significant accuracy improvements over previous NAT baselines and reduces the performance gap between NAT and AT models to 1-2 BLEU points, demonstrating the effectiveness of our proposed method.

Real-world applications often involve domain-specific and task-based performance objectives that are not captured by the standard machine learning losses, but are critical for decision making. A key challenge for direct integration of more meaningful domain and task-based evaluation criteria into an end-to-end gradient-based training process is the fact that often such performance objectives are not necessarily differentiable and may even require additional decision-making optimization processing. We propose the Task-Oriented Prediction Network (TOPNet), an end-to-end learning scheme that automatically integrates task-based evaluation criteria into the learning process via a learnable surrogate loss function, which directly guides the model towards the task-based goal. A major benefit of the proposed TOPNet learning scheme lies in its capability of automatically integrating non-differentiable evaluation criteria, which makes it particularly suitable for diversified and customized task-based evaluation criteria in real-world tasks. We validate the performance of TOPNet on two real-world financial prediction tasks, revenue surprise forecasting and credit risk modeling. The experimental results demonstrate that TOPNet significantly outperforms both traditional modeling with standard losses and modeling with hand-crafted heuristic differentiable surrogate losses.

To enable DNNs on edge devices like mobile phones, low-rank approximation has been widely adopted because of its solid theoretical rationale and efficient implementations. Several previous works attempted to directly approximate a pre-trained model by low-rank decomposition; however, small approximation errors in parameters can ripple over a large prediction loss. As a result, performance usually drops significantly and a sophisticated effort on fine-tuning is required to recover accuracy. Apparently, it is not optimal to separate low-rank approximation from training. Unlike previous works, this paper integrates low rank approximation and regularization into the training process. We propose Trained Rank Pruning (TRP), which alternates between low rank approximation and training. TRP maintains the capacity of the original network while imposing low-rank constraints during training. A nuclear regularization optimized by stochastic sub-gradient descent is utilized to further promote low rank in TRP. The TRP trained network inherently has a low-rank structure, and is approximated with negligible performance loss, thus eliminating the fine-tuning process after low rank decomposition. The proposed method is comprehensively evaluated on CIFAR-10 and ImageNet, outperforming previous compression methods using low rank approximation.

Target localization and proposal generation are two essential subtasks in generic visual tracking, and it is a challenge to address both the two efficiently. In this paper, we propose an efficient two-stage architecture which makes full use of the complementarity of two subtasks to achieve robust localization and high-quality proposals generation of the target jointly. Specifically, our model performs a novel deformable central correlation operation by an online learning model in both two stages to locate new target centers while generating target proposals in the vicinity of these centers. The proposals are refined in the refinement stage to further improve accuracy and robustness. Moreover, the model benefits from multi-level features aggregation in a neck module and a feature enhancement module. We conduct extensive ablation studies to demonstrate the effectiveness of our proposed methods. Our tracker runs at over 30 FPS and sets a new state-of-the-art on five tracking benchmarks, including LaSOT, VOT2018, TrackingNet, GOT10k, OTB2015.

We consider an agent that operates with two models of the environment: one that captures expected behaviors and one that captures additional exceptional behaviors. We study the problem of synthesizing agent strategies that enforce a goal against environments operating as expected while also making a best effort against exceptional environment behaviors. We formalize these concepts in the context of linear-temporal logic, and give an algorithm for solving this problem. We also show that there is no trade-off between enforcing the goal under the expected environment specification and making a best-effort for it under the exceptional one.

Explanations help users make sense of recommendations, increasing the likelihood of adoption. Existing approaches to explainable recommendations tend to rely on rigidly standardized templates, only allowing fill-in-the-blank aspect-level sentiments. For more flexible, literate, and varied explanations that cover various aspects of interest, we propose to synthesize an explanation by selecting snippets from reviews to optimize representativeness and coherence. To fit the target user's aspect preferences, we contextualize the opinions based on a compatible explainable recommendation model. Experiments on datasets of varying product categories showcase the efficacies of our method as compared to baselines based on templates, review summarization, selection, and text generation.

We consider a class of two-player turn-based zero-sum games on graphs with reachability objectives, known as reachability games, where the objective of Player 1 (P1) is to reach a set of goal states, and that of Player 2 (P2) is to prevent this. In particular, we consider the case where the players have asymmetric information about each other's action capabilities: P2 starts with an incomplete information (misperception) about P1's action set, and updates the misperception when P1 uses an action previously unknown to P2. When P1 is made aware of P2's misperception, the key question is whether P1 can control P2's perception so as to deceive P2 into selecting actions to P1's advantage? To answer this question, we introduce a dynamic hypergame model to capture the reachability game with evolving misperception of P2. Then, we present a fixed-point algorithm to compute the deceptive winning region and strategy for P1 under almost-sure winning condition. Finally, we show that the synthesized deceptive winning strategy is at least as powerful as the (non-deceptive) winning strategy in the game in which P1 does not account for P2's misperception. We illustrate our algorithm using a robot motion planning in an adversarial environment.

We study a question answering problem on a social network, where a requester is seeking an answer from the agents on the network. The goal is to design reward mechanisms to incentivize the agents to propagate the requester's query to their neighbours if they don't have the answer. Existing mechanisms are vulnerable to Sybil-attacks, i.e., an agent may get more reward by creating fake identities. Hence, we combat this problem by first proving some impossibility results to resolve Sybil-attacks and then characterizing a class of mechanisms which satisfy Sybil-proofness (prevents Sybil-attacks) as well as other desirable properties. Except for Sybil-proofness, we also consider cost minimization for the requester and agents' collusions.

We propose Switching Poisson gamma dynamical systems (SPGDS) to model sequentially observed multivariate count data. Different from previous models, SPGDS assigns its latent variables into mixture of gamma distributed parameters to model complex sequences and describe the nonlinear dynamics, meanwhile, capture various temporal dependencies. For efficient inference, we develop a scalable hybrid stochastic gradient-MCMC and switching recurrent autoencoding variational inference, which is scalable to large scale sequences and fast in out-of-sample prediction. Experiments on both unsupervised and supervised tasks demonstrate that the proposed model not only has excellent fitting and prediction performance on complex dynamic sequences, but also separates different dynamical patterns within them.

In this paper we focus on a less usual way to represent Boolean functions, namely on representations by switch-lists. Given a truth table representation of a Boolean function f the switch-list representation (SLR) of f is a list of Boolean vectors from the truth table which have a different function value than the preceding Boolean vector in the truth table. The main aim of this paper is to include the language SL of all SLR in the Knowledge Compilation Map [Darwiche and Marquis, 2002] and to argue, that SL may in certain situations constitute a reasonable choice for a target language in knowledge compilation. First we compare SL with a number of standard representation languages (such as CNF, DNF, and OBDD) with respect to their relative succinctness. As a by-product of this analysis we also give a short proof of a long standing open question from [Darwiche and Marquis, 2002], namely the incomparability of MODS (models) and PI (prime implicates) languages. Next we analyze which standard transformations and queries (those considered in [Darwiche and Marquis, 2002] can be performed in poly-time with respect to the size of the input SLR. We show that this collection is quite broad and the combination of poly-time transformations and queries is quite unique.

The Databionic swarm (DBS) is a flexible and robust clustering framework that consists of three independent modules: swarm based projection, high-dimensional data visualization and representation guided clustering. The first module is the parameter-free projection method Pswarm, which exploits concepts of self-organization and emergence, game theory, and swarm intelligence. The second module is a parameter-free high-dimensional data visualization technique called topographic map. It uses the generalized U-matrix, which enables to estimate first, if any cluster tendency exists and second, the estimation of the number of clusters. The third module offers a clustering method which can be verified by the visualization and vice versa. Benchmarking w.r.t. conventional algorithms demonstrated that DBS can outperform them. Several applications showed that cluster structures provided by DBS are meaningful. Exemplary, a clustering of worldwide country-related data w.r.t the COVID-19 pandemic is presented here. Code and data is made available via open source.