IJCAI 2021 论文列表

Topic models are characterized by a latent class variable that represents the different topics. Traditionally, their observable variables are modeled as discrete variables like, for instance, in the prototypical latent Dirichlet allocation (LDA) topic model. In LDA, words in text documents are encoded by discrete count vectors with respect to some dictionary. The classical approach for learning topic models optimizes a likelihood function that is non-concave due to the presence of the latent variable. Hence, this approach mostly boils down to using search heuristics like the EM algorithm for parameter estimation. Recently, it was shown that topic models can be learned with strong algorithmic and statistical guarantees through Pearson's method of moments. Here, we extend this line of work to topic models that feature discrete as well as continuous observable variables (features). Moving beyond discrete variables as in LDA allows for more sophisticated features and a natural extension of topic models to other modalities than text, like, for instance, images. We provide algorithmic and statistical guarantees for the method of moments applied to the extended topic model that we corroborate experimentally on synthetic data. We also demonstrate the applicability of our model on real-world document data with embedded images that we preprocess into continuous state-of-the-art feature vectors.

Many real-world domains are subject to a structured non-stationarity which affects the agent's goals and the environmental dynamics. Meta-reinforcement learning (RL) has been shown successful for training agents that quickly adapt to related tasks. However, most of the existing meta-RL algorithms for non-stationary domains either make strong assumptions on the task generation process or require sampling from it at training time. In this paper, we propose a novel algorithm (TRIO) that optimizes for the future by explicitly tracking the task evolution through time. At training time, TRIO learns a variational module to quickly identify latent parameters from experience samples. This module is learned jointly with an optimal exploration policy that takes task uncertainty into account. At test time, TRIO tracks the evolution of the latent parameters online, hence reducing the uncertainty over future tasks and obtaining fast adaptation through the meta-learned policy. Unlike most existing methods, TRIO does not assume Markovian task-evolution processes, it does not require information about the non-stationarity at training time, and it captures complex changes undergoing in the environment. We evaluate our algorithm on different simulated problems and show it outperforms competitive baselines.

As more and more forms of AI become prevalent, it becomes increasingly important to understand how people develop mental models of these systems. In this work we study people's mental models of an AI agent in a cooperative word guessing game. We run a study in which people play the game with an AI agent while ``thinking out loud''; through thematic analysis we identify features of the mental models developed by participants. In a large-scale study we have participants play the game with the AI agent online and use a post-game survey to probe their mental model. We find that those who win more often have better estimates of the AI agent's abilities. We present three components---global knowledge, local knowledge, and knowledge distribution---for modeling AI systems and propose that understanding the underlying technology is insufficient for developing appropriate conceptual models---analysis of behavior is also necessary.

Medical image segmentation is important for computer-aided diagnosis. Good segmentation demands the model to see the big picture and fine details simultaneously, i.e., to learn image features that incorporate large context while keep high spatial resolutions. To approach this goal, the most widely used methods -- U-Net and variants, extract and fuse multi-scale features. However, the fused features still have small "effective receptive fields" with a focus on local image cues, limiting their performance. In this work, we propose Segtran, an alternative segmentation framework based on transformers, which have unlimited "effective receptive fields" even at high feature resolutions. The core of Segtran is a novel Squeeze-and-Expansion transformer: a squeezed attention block regularizes the self attention of transformers, and an expansion block learns diversified representations. Additionally, we propose a new positional encoding scheme for transformers, imposing a continuity inductive bias for images. Experiments were performed on 2D and 3D medical image segmentation tasks: optic disc/cup segmentation in fundus images (REFUGE'20 challenge), polyp segmentation in colonoscopy images, and brain tumor segmentation in MRI scans (BraTS'19 challenge). Compared with representative existing methods, Segtran consistently achieved the highest segmentation accuracy, and exhibited good cross-domain generalization capabilities.

The study of approximate mechanism design for facility location has been in the center of research at the intersection of artificial intelligence and economics for the last decade, largely due to its practical importance in various domains, such as social planning and clustering. At a high level, the goal is to select a number of locations on which to build a set of facilities, aiming to optimize some social objective based on the preferences of strategic agents, who might have incentives to misreport their private information. This paper presents a comprehensive survey of the significant progress that has been made since the introduction of the problem, highlighting all the different variants and methodologies, as well as the most interesting directions for future research.

We present a method enabling a large number of agents to learn how to flock. This problem has drawn a lot of interest but requires many structural assumptions and is tractable only in small dimensions. We phrase this problem as a Mean Field Game (MFG), where each individual chooses its own acceleration depending on the population behavior. Combining Deep Reinforcement Learning (RL) and Normalizing Flows (NF), we obtain a tractable solution requiring only very weak assumptions. Our algorithm finds a Nash Equilibrium and the agents adapt their velocity to match the neighboring flock’s average one. We use Fictitious Play and alternate: (1) computing an approximate best response with Deep RL, and (2) estimating the next population distribution with NF. We show numerically that our algorithm can learn multi-group or high-dimensional flocking with obstacles.

Mean field game facilitates analyzing multi-armed bandit (MAB) for a large number of agents by approximating their interactions with an average effect. Existing mean field models for multi-agent MAB mostly assume a binary reward function, which leads to tractable analysis but is usually not applicable in practical scenarios. In this paper, we study the mean field bandit game with a continuous reward function. Specifically, we focus on deriving the existence and uniqueness of mean field equilibrium (MFE), thereby guaranteeing the asymptotic stability of the multi-agent system. To accommodate the continuous reward function, we encode the learned reward into an agent state, which is in turn mapped to its stochastic arm playing policy and updated using realized observations. We show that the state evolution is upper semi-continuous, based on which the existence of MFE is obtained. As the Markov analysis is mainly for the case of discrete state, we transform the stochastic continuous state evolution into a deterministic ordinary differential equation (ODE). On this basis, we can characterize a contraction mapping for the ODE to ensure a unique MFE for the bandit game. Extensive evaluations validate our MFE characterization, and exhibit tight empirical regret of the MAB problem.

We consider the problem of assigning items to platforms in the presence of group fairness constraints. In the input, each item belongs to certain categories, called classes in this paper. Each platform specifies the group fairness constraints through an upper bound on the number of items it can serve from each class. Additionally, each platform also has an upper bound on the total number of items it can serve. The goal is to assign items to platforms so as to maximize the number of items assigned while satisfying the upper bounds of each class. This problem models several important real-world problems like ad-auctions, scheduling, resource allocations, school choice etc. We show that if the classes are arbitrary, then the problem is NP-hard and has a strong inapproximability. We consider the problem in both online and offline settings under natural restrictions on the classes. Under these restrictions, the problem continues to remain NP-hard but admits approximation algorithms with small approximation factors. We also implement some of the algorithms. Our experiments show that the algorithms work well in practice both in terms of efficiency and the number of items that get assigned to some platform.

Visual Information Extraction (VIE) task aims to extract key information from multifarious document images (e.g., invoices and purchase receipts). Most previous methods treat the VIE task simply as a sequence labeling problem or classification problem, which requires models to carefully identify each kind of semantics by introducing multimodal features, such as font, color, layout. But simply introducing multimodal features can't work well when faced with numeric semantic categories or some ambiguous texts. To address this issue, in this paper we propose a novel key-value matching model based on a graph neural network for VIE (MatchVIE). Through key-value matching based on relevancy evaluation, the proposed MatchVIE can bypass the recognitions to various semantics, and simply focuses on the strong relevancy between entities. Besides, we introduce a simple but effective operation, Num2Vec, to tackle the instability of encoded values, which helps model converge more smoothly. Comprehensive experiments demonstrate that the proposed MatchVIE can significantly outperform previous methods. Notably, to the best of our knowledge, MatchVIE may be the first attempt to tackle the VIE task by modeling the relevancy between keys and values and it is a good complement to the existing methods.

Graph neural network (GNN) and label propagation algorithm (LPA) are both message passing algorithms, which have achieved superior performance in semi-supervised classification. GNN performs feature propagation by a neural network to make predictions, while LPA uses label propagation across graph adjacency matrix to get results. However, there is still no effective way to directly combine these two kinds of algorithms. To address this issue, we propose a novel Unified Message Passaging Model (UniMP) that can incorporate feature and label propagation at both training and inference time. First, UniMP adopts a Graph Transformer network, taking feature embedding and label embedding as input information for propagation. Second, to train the network without overfitting in self-loop input label information, UniMP introduces a masked label prediction strategy, in which some percentage of input label information are masked at random, and then predicted. UniMP conceptually unifies feature propagation and label propagation and is empirically powerful. It obtains new state-of-the-art semi-supervised classification results in Open Graph Benchmark (OGB).

Detecting anomalies is one fundamental aspect of a safety-critical software system, however, it remains a long-standing problem. Numerous branches of works have been proposed to alleviate the complication and have shown promising results. In particular, self-supervised learning based methods are spurring interest due to their capability of learning diverse representations without additional labels. Among self-supervised learning tactics, contrastive learning is one specific framework showing pronounced results in various fields including anomaly detection. However, the primary objective of contrastive learning is to learn task-agnostic features without any labels, which is not entirely suited to discern anomalies. In this paper, we propose a task-specific variant of contrastive learning named masked contrastive learning, which is more befitted for anomaly detection. Moreover, we propose a new inference method dubbed self-ensemble inference that further boosts performance by leveraging the ability learned through auxiliary self-supervision tasks. By combining our models, we can outperform previous state-of-the-art methods by a significant margin on various benchmark datasets.

In Goal-oriented Reinforcement learning, relabeling the raw goals in past experience to provide agents with hindsight ability is a major solution to the reward sparsity problem. In this paper, to enhance the diversity of relabeled goals, we develop FGI (Foresight Goal Inference), a new relabeling strategy that relabels the goals by looking into the future with a learned dynamics model. Besides, to improve sample efficiency, we propose to use the dynamics model to generate simulated trajectories for policy training. By integrating these two improvements, we introduce the MapGo framework (Model-Assisted Policy optimization for Goal-oriented tasks). In our experiments, we first show the effectiveness of the FGI strategy compared with the hindsight one, and then show that the MapGo framework achieves higher sample efficiency when compared to model-free baselines on a set of complicated tasks.

In many coalition formation games the utility of the agents depends on a social network. In such scenarios there might be a manipulative agent that would like to manipulate his connections in the social network in order to increase his utility. We study a model of coalition formation in which a central organizer, who needs to form k coalitions, obtains information about the social network from the agents. The central organizer has her own objective: she might want to maximize the utilitarian social welfare, maximize the egalitarian social welfare, or only guarantee that every agent will have at least one connection within her coalition. In this paper we study the susceptibility for manipulation of these objectives, given the abilities and information that the manipulator has. Specifically, we show that if the manipulator has very limited information, namely he is only familiar with his immediate neighbours in the network, then a manipulation is almost always impossible. Moreover, if the manipulator is only able to add connections to the social network, then a manipulation is still impossible for some objectives, even if the manipulator has full information on the structure of the network. On the other hand, if the manipulator is able to hide some of his connections, then all objectives are susceptible to manipulation, even if the manipulator has limited information, i.e., when he is familiar with his immediate neighbours and with their neighbours.

Consider a graph G, representing a social network. Assume that initially each node is colored either black or white, which corresponds to a positive or negative opinion regarding a consumer product or a technological innovation. In the majority model, in each round all nodes simultaneously update their color to the most frequent color among their connections. Experiments on the graph data from the real world social networks (SNs) suggest that if all nodes in an extremely small set of high-degree nodes, often referred to as the elites, agree on a color, that color becomes the dominant color at the end of the process. We propose two countermeasures that can be adopted by individual nodes relatively easily and guarantee that the elites will not have this disproportionate power to engineer the dominant output color. The first countermeasure essentially requires each node to make some new connections at random while the second one demands the nodes to be more reluctant towards changing their color (opinion). We verify their effectiveness and correctness both theoretically and experimentally. We also investigate the majority model and a variant of it when the initial coloring is random on the real world SNs and several random graph models. In particular, our results on the Erdős-Rényi, and regular random graphs confirm or support several theoretical findings or conjectures by the prior work regarding the threshold behavior of the process. Finally, we provide theoretical and experimental evidence for the existence of a poly-logarithmic bound on the expected stabilization time of the majority model.

Spoken question answering (SQA) has recently drawn considerable attention in the speech community. It requires systems to find correct answers from the given spoken passages simultaneously. The common SQA systems consist of the automatic speech recognition (ASR) module and text-based question answering module. However, previous methods suffer from severe performance degradation due to ASR errors. To alleviate this problem, this work proposes a novel multi-modal residual knowledge distillation method (MRD-Net), which further distills knowledge at the acoustic level from the audio-assistant (Audio-A). Specifically, we utilize the teacher (T) trained on manual transcriptions to guide the training of the student (S) on ASR transcriptions. We also show that introducing an Audio-A helps this procedure by learning residual errors between T and S. Moreover, we propose a simple yet effective attention mechanism to adaptively leverage audio-text features as the new deep attention knowledge to boost the network performance. Extensive experiments demonstrate that the proposed MRD-Net achieves superior results compared with state-of-the-art methods on three spoken question answering benchmark datasets.

Detecting the newly emerging malware variants in real time is crucial for mitigating cyber risks and proactively blocking intrusions. In this paper, we propose MG-DVD, a novel detection framework based on dynamic heterogeneous graph learning, to detect malware variants in real time. Particularly, MG-DVD first models the fine-grained execution event streams of malware variants into dynamic heterogeneous graphs and investigates real-world meta-graphs between malware objects, which can effectively characterize more discriminative malicious evolutionary patterns between malware and their variants. Then, MG-DVD presents two dynamic walk-based heterogeneous graph learning methods to learn more comprehensive representations of malware variants, which significantly reduces the cost of the entire graph retraining. As a result, MG-DVD is equipped with the ability to detect malware variants in real time, and it presents better interpretability by introducing meaningful meta-graphs. Comprehensive experiments on large-scale samples prove that our proposed MG-DVD outperforms state-of-the-art methods in detecting malware variants in terms of effectiveness and efficiency.

Value function decomposition (VFD) methods under the popular paradigm of centralized training and decentralized execution (CTDE) have promoted multi-agent reinforcement learning progress. However, existing VFD methods proceed from a group's value function decomposition to only solve cooperative tasks. With the individual value function decomposition, we propose MFVFD, a novel multi-agent Q-learning approach for solving cooperative and non-cooperative tasks based on mean-field theory. Our analysis on the Hawk-Dove and Nonmonotonic Cooperation matrix games evaluate MFVFD's convergent solution. Empirical studies on the challenging mixed cooperative-competitive tasks where hundreds of agents coexist demonstrate that MFVFD significantly outperforms existing baselines.

Next Point-of-Interest (POI) recommendation is of great value for location-based services. Existing solutions mainly rely on extensive observed data and are brittle to users with few interactions. Unfortunately, the problem of few-shot next POI recommendation has not been well studied yet. In this paper, we propose a novel meta-optimized model MFNP, which can rapidly adapt to users with few check-in records. Towards the cold-start problem, it seamlessly integrates carefully designed user-specific and region-specific tasks in meta-learning, such that region-aware user preferences can be captured via a rational fusion of region-independent personal preferences and region-dependent crowd preferences. In modelling region-dependent crowd preferences, a cluster-based adaptive network is adopted to capture shared preferences from similar users for knowledge transfer. Experimental results on two real-world datasets show that our model outperforms the state-of-the-art methods on next POI recommendation for cold-start users.

Meta-learning has recently emerged as a promising technique to address the challenge of few-shot learning. However, standard meta-learning methods mainly focus on visual tasks, which makes it hard for them to deal with diverse text data directly. In this paper, we introduce a novel framework for few-shot text classification, which is named as MEta-learning with Data Augmentation (MEDA). MEDA is composed of two modules, a ball generator and a meta-learner, which are learned jointly. The ball generator is to increase the number of shots per class by generating more samples, so that meta-learner can be trained with both original and augmented samples. It is worth noting that ball generator is agnostic to the choice of the meta-learning methods. Experiment results show that on both datasets, MEDA outperforms existing state-of-the-art methods and significantly improves the performance of meta-learning on few-shot text classification.