IJCAI 2021 Papers

Despite the sublime success in recent years, the underlying mechanisms powering the advances of reinforcement learning are yet poorly understood. In this paper, we identify these mechanisms - which we call ingredients - in on-policy policy gradient methods and empirically determine their impact on the learning. To allow an equitable assessment, we conduct our experiments based on a unified and modular implementation. Our results underline the significance of recent algorithmic advances and demonstrate that reaching state-of-the-art performance may not need sophisticated algorithms but can also be accomplished by the combination of a few simple ingredients.

In recent years, algorithms and neural architectures based on the Weisfeiler-Leman algorithm, a well-known heuristic for the graph isomorphism problem, emerged as a powerful tool for (supervised) machine learning with graphs and relational data. Here, we give a comprehensive overview of the algorithm's use in a machine learning setting. We discuss the theoretical background, show how to use it for supervised graph- and node classification, discuss recent extensions, and its connection to neural architectures. Moreover, we give an overview of current applications and future directions to stimulate research.

We study the Connected Fair Division problem (CFD), which generalizes the fundamental problem of fairly allocating resources to agents by requiring that the items allocated to each agent form a connected subgraph in a provided item graph G. We expand on previous results by providing a comprehensive complexity-theoretic understanding of CFD based on several new algorithms and lower bounds while taking into account several well-established notions of fairness: proportionality, envy-freeness, EF1 and EFX. In particular, we show that to achieve tractability, one needs to restrict both the agents and the item graph in a meaningful way. We design (XP)-algorithms for the problem parameterized by (1) clique-width of G plus the number of agents and (2) treewidth of G plus the number of agent types, along with corresponding lower bounds. Finally, we show that to achieve fixed-parameter tractability, one needs to not only use a more restrictive parameterization of G, but also include the maximum item valuation as an additional parameter.

Teachers’ spatial behaviours in the classroom can strongly influence students’ engagement, motivation and other behaviours that shape their learning. However, classroom teaching behav-iour is ephemeral, and has largely remained opaque to computational analysis. This paper presents a library called ‘Moodoo’ that can serve to automatically model how teachers make use of the classroom space by analysing indoor positioning traces. The system automatically ex-tracts spatial metrics (e.g. teacher-student ratios, frequency of visits to students’ personal spaces, presence in classroom spaces of interest, index of dispersion and entropy), mapping from the teachers’ low-level positioning data to higher-order spatial constructs.

Heuristic search is among the best performing approaches to classical satisficing planning, with its performance heavily relying on informative and fast heuristics, as well as search-boosting and pruning techniques. While both heuristics and pruning techniques have gained much attention recently, search-boosting techniques in general, and preferred operators in particular have received less attention in the last decade. Our work aims at bringing the light back to preferred operators research, with the introduction of preferred operators pruning technique, based on the concept of novelty. Continuing the research on novelty with respect to an underlying heuristic, we present the definition of preferred operators for such novelty heuristics. For that, we extend the previously defined concepts to operators, allowing us to reason about the novelty of the preferred operators. Our experimental evaluation shows the practical benefit of our suggested approach, compared to the currently used methods.

Compared with traditional sequential learning models, graph-based neural networks exhibit excellent properties when encoding text, such as the capacity of capturing global and local information simultaneously. Especially in the semi-supervised scenario, propagating information along the edge can effectively alleviate the sparsity of labeled data. In this paper, beyond the existing architecture of heterogeneous word-document graphs, for the first time, we investigate how to construct lightweight non-heterogeneous graphs based on different linguistic information to better serve free text representation learning. Then, a novel semi-supervised framework for text classification that refines graph topology under theoretical guidance and shares information across different text graphs, namely Text-oriented Graph-based Transductive Learning (TextGTL), is proposed. TextGTL also performs attribute space interpolation based on dense substructure in graphs to predict low-entropy labels with high-quality feature nodes for data augmentation. To verify the effectiveness of TextGTL, we conduct extensive experiments on various benchmark datasets, observing significant performance gains over conventional heterogeneous graphs. In addition, we also design ablation studies to dive deep into the validity of components in TextTGL.

Most existing text-based person search methods highly depend on exploring the corresponding relations between the regions of the image and the words in the sentence. However, these methods correlated image regions and words in the same semantic granularity. It 1) results in irrelevant corresponding relations between image and text, 2) causes an ambiguity embedding problem. In this study, we propose a novel multi-granularity embedding learning model for text-based person search. It generates multi-granularity embeddings of partial person bodies in a coarse-to-fine manner by revisiting the person image at different spatial scales. Specifically, we distill the partial knowledge from image scrips to guide the model to select the semantically relevant words from the text description. It can learn discriminative and modality-invariant visual-textual embeddings. In addition, we integrate the partial embeddings at each granularity and perform multi-granularity image-text matching. Extensive experiments validate the effectiveness of our method, which can achieve new state-of-the-art performance by the learned discriminative partial embeddings.

The intelligent manipulation of symbolic knowledge has been a long-sought goal of AI. However, when it comes to Natural Language Processing (NLP), symbols have to be mapped to words and phrases, which are not only ambiguous but also language-specific: multilinguality is indeed a desirable property for NLP systems, and one which enables the generalization of tasks where multiple languages need to be dealt with, without translating text. In this paper we survey BabelNet, a popular wide-coverage lexical-semantic knowledge resource obtained by merging heterogeneous sources into a unified semantic network that helps to scale tasks and applications to hundreds of languages. Over its ten years of existence, thanks to its promise to interconnect languages and resources in structured form, BabelNet has been employed in countless ways and directions. We first introduce the BabelNet model, its components and statistics, and then overview its successful use in a wide range of tasks in NLP as well as in other fields of AI.

We present Temporal and Object Quantification Networks (TOQ-Nets), a new class of neuro-symbolic networks with a structural bias that enables them to learn to recognize complex relational-temporal events. This is done by including reasoning layers that implement finite-domain quantification over objects and time. The structure allows them to generalize directly to input instances with varying numbers of objects in temporal sequences of varying lengths. We evaluate TOQ-Nets on input domains that require recognizing event-types in terms of complex temporal relational patterns. We demonstrate that TOQ-Nets can generalize from small amounts of data to scenarios containing more objects than were present during training and to temporal warpings of input sequences.

One practical requirement in solving dynamic games is to ensure that the players play well from any decision point onward. To satisfy this requirement, existing efforts focus on equilibrium refinement, but the scalability and applicability of existing techniques are limited. In this paper, we propose Temporal-Induced Self-Play (TISP), a novel reinforcement learning-based framework to find strategies with decent performances from any decision point onward. TISP uses belief-space representation, backward induction, policy learning, and non-parametric approximation. Building upon TISP, we design a policy-gradient-based algorithm TISP-PG. We prove that TISP-based algorithms can find approximate Perfect Bayesian Equilibrium in zero-sum one-sided stochastic Bayesian games with finite horizon. We test TISP-based algorithms in various games, including finitely repeated security games and a grid-world game. The results show that TISP-PG is more scalable than existing mathematical programming-based methods and significantly outperforms other learning-based methods.

Heterogeneous information network (HIN) embedding, learning the low-dimensional representation of multi-type nodes, has been applied widely and achieved excellent performance. However, most of the previous works focus more on static heterogeneous networks or learning node embedding within specific snapshots, and seldom attention has been paid to the whole evolution process and capturing all temporal dynamics. In order to fill the gap of obtaining multi-type node embeddings by considering all temporal dynamics during the evolution, we propose a novel temporal HIN embedding method (THINE). THINE not only uses attention mechanism and meta-path to preserve structures and semantics in HIN but also combines the Hawkes process to simulate the evolution of the temporal network. Our extensive evaluations with various real-world temporal HINs demonstrate that THINE achieves state-of-the-art performance in both static and dynamic tasks, including node classification, link prediction, and temporal link recommendation.

A flexible communication protocol is necessary to build a decentralized multiagent system whose member agents are not coupled to each other's decision making. Information-based protocol languages capture a protocol in terms of causality and integrity constraints based on the information exchanged by the agents. Thus, they enable highly flexible enactments in which the agents proceed asynchronously and messages may be arbitrarily reordered. However, the existing semantics for such languages can produce a large number of protocol enactments, which makes verification of a protocol property intractable. This paper formulates a protocol semantics declaratively via inference rules that determine when a message emission or reception becomes enabled during an enactment, and its effect on the local state of an agent. The semantics enables heuristics for determining when alternative extensions of a current enactment would be equivalent, thereby helping produce parsimonious models and yielding improved protocol verification methods.

Visual information extraction (VIE) has attracted increasing attention in recent years. The existing methods usually first organized optical character recognition (OCR) results in plain texts and then utilized token-level category annotations as supervision to train a sequence tagging model. However, it expends great annotation costs and may be exposed to label confusion, the OCR errors will also significantly affect the final performance. In this paper, we propose a unified weakly-supervised learning framework called TCPNet (Tag, Copy or Predict Network), which introduces 1) an efficient encoder to simultaneously model the semantic and layout information in 2D OCR results, 2) a weakly-supervised training method that utilizes only sequence-level supervision; and 3) a flexible and switchable decoder which contains two inference modes: one (Copy or Predict Mode) is to output key information sequences of different categories by copying a token from the input or predicting one in each time step, and the other (Tag Mode) is to directly tag the input sequence in a single forward pass. Our method shows new state-of-the-art performance on several public benchmarks, which fully proves its effectiveness.

Using the principle of imitation learning and the theory of optimal transport we propose in this paper a novel model for unsupervised domain adaptation named Teacher Imitation Domain Adaptation with Optimal Transport (TIDOT). Our model includes two cooperative agents: a teacher and a student. The former agent is trained to be an expert on labeled data in the source domain, whilst the latter one aims to work with unlabeled data in the target domain. More specifically, optimal transport is applied to quantify the total of the distance between embedded distributions of the source and target data in the joint space, and the distance between predictive distributions of both agents, thus by minimizing this quantity TIDOT could mitigate not only the data shift but also the label shift. Comprehensive empirical studies show that TIDOT outperforms existing state-of-the-art performance on benchmark datasets.

Political discourses provide a forum for representatives to express their opinions and contribute towards policy making. Analyzing these discussions is crucial for recognizing possible delegates and making better voting choices in an independent nation. A politician's vote on a proposition is usually associated with their past discourses and impacted by cohesion forces in political parties. We focus on predicting a speaker's vote on a bill by augmenting linguistic models with temporal and cohesion contexts. We propose TEC, a time evolving graph based model that jointly employs links between motions, speakers, and temporal politician states. TEC outperforms competitive models, illustrating the benefit of temporal and contextual signals for predicting a politician's stance.

Prediction based on Irregularly Sampled Time Series (ISTS) is of wide concern in real-world applications. For more accurate prediction, methods had better grasp more data characteristics. Different from ordinary time series, ISTS is characterized by irregular time intervals of intra-series and different sampling rates of inter-series. However, existing methods have suboptimal predictions due to artificially introducing new dependencies in a time series and biasedly learning relations among time series when modeling these two characteristics. In this work, we propose a novel Time Encoding (TE) mechanism. TE can embed the time information as time vectors in the complex domain. It has the properties of absolute distance and relative distance under different sampling rates, which helps to represent two irregularities. Meanwhile, we create a new model named Time Encoding Echo State Network (TE-ESN). It is the first ESNs-based model that can process ISTS data. Besides, TE-ESN incorporates long short-term memories and series fusion to grasp horizontal and vertical relations. Experiments on one chaos system and three real-world datasets show that TE-ESN performs better than all baselines and has better reservoir property.

Existing research for image captioning usually represents an image using a scene graph with low-level facts (objects and relations) and fails to capture the high-level semantics. In this paper, we propose a Theme Concepts extended Image Captioning (TCIC) framework that incorporates theme concepts to represent high-level cross-modality semantics. In practice, we model theme concepts as memory vectors and propose Transformer with Theme Nodes (TTN) to incorporate those vectors for image captioning. Considering that theme concepts can be learned from both images and captions, we propose two settings for their representations learning based on TTN. On the vision side, TTN is configured to take both scene graph based features and theme concepts as input for visual representation learning. On the language side, TTN is configured to take both captions and theme concepts as input for text representation re-construction. Both settings aim to generate target captions with the same transformer-based decoder. During the training, we further align representations of theme concepts learned from images and corresponding captions to enforce the cross-modality learning. Experimental results on MS COCO show the effectiveness of our approach compared to some state-of-the-art models.

Network representation learning aims at transferring node proximity in networks into distributed vectors, which can be leveraged in various downstream applications. Recent research has shown that nodes in a network can often be organized in latent hierarchical structures, but without a particular underlying taxonomy, the learned node embedding is less useful nor interpretable. In this work, we aim to improve network embedding by modeling the conditional node proximity in networks indicated by node labels residing in real taxonomies. In the meantime, we also aim to model the hierarchical label proximity in the given taxonomies, which is too coarse by solely looking at the hierarchical topologies. Comprehensive experiments and case studies demonstrate the utility of TAXOGAN.

Robots assisting us in factories or homes must learn to make use of objects as tools to perform tasks, e.g., a tray for carrying objects. We consider the problem of learning commonsense knowledge of when a tool may be useful and how its use may be composed with other tools to accomplish a high-level task instructed by a human. We introduce TANGO, a novel neural model for predicting task-specific tool interactions. TANGO is trained using demonstrations obtained from human teachers instructing a virtual robot in a physics simulator. TANGO encodes the world state consisting of objects and symbolic relationships between them using a graph neural network. The model learns to attend over the scene using knowledge of the goal and the action history, finally decoding the symbolic action to execute. Crucially, we address generalization to unseen environments where some known tools are missing, but alternative unseen tools are present. We show that by augmenting the representation of the environment with pre-trained embeddings derived from a knowledge-base, the model can generalize effectively to novel environments. Experimental results show a 60.5-78.9% improvement over the baseline in predicting successful symbolic plans in unseen settings for a simulated mobile manipulator.