IJCAI 2022 Papers

Most translation tasks among languages belong to the zero-resource translation problem where parallel corpora are unavailable. Multilingual neural machine translation (MNMT) enables one-pass translation using shared semantic space for all languages compared to the two-pass pivot translation but often underperforms the pivot-based method. In this paper, we propose a novel method, named as Unified Multilingual Multiple teacher-student Model for NMT (UM4). Our method unifies source-teacher, target-teacher, and pivot-teacher models to guide the student model for the zero-resource translation. The source teacher and target teacher force the student to learn the direct source-target translation by the distilled knowledge on both source and target sides. The monolingual corpus is further leveraged by the pivot-teacher model to enhance the student model. Experimental results demonstrate that our model of 72 directions significantly outperforms previous methods on the WMT benchmark.

Multi-hop reasoning over real-life knowledge graphs (KGs) is a highly challenging problem as traditional subgraph matching methods are not capable to deal with noise and missing information. Recently, to address this problem a promising approach based on jointly embedding logical queries and KGs into a low-dimensional space to identify answer entities has emerged. However, existing proposals ignore critical semantic knowledge inherently available in KGs, such as type information. To leverage type information, we propose a novel type-aware model, TypE-aware Message Passing (TEMP), which enhances the entity and relation representation in queries, and simultaneously improves generalization, and deductive and inductive reasoning. Remarkably, TEMP is a plug-and-play model that can be easily incorporated into existing embedding-based models to improve their performance. Extensive experiments on three real-world datasets demonstrate TEMP’s effectiveness.

A new paradigm of mechanism design, called mechanism design over social networks, investigates agents’ incentives to diffuse the information of mechanisms to their followers over social networks. In this paper we consider it for two-sided matching, where the agents on one side, say students, are distributed over social networks and thus are not fully observable to the mechanism designer, while the agents on the other side, say colleges, are known a priori. The main purpose of this paper is to clarify the existence of mechanisms that satisfy several properties that are classified into four criteria: incentive constraints, efficiency constraints, stability constraints, and fairness constraints. We proposed three mechanisms and showed that no mechanism is better than these mechanisms, i.e., they are in the Pareto frontier according to the set of properties defined in this paper.

Strategic behavior in two-sided matching markets has been traditionally studied in a "one-sided" manipulation setting where the agent who misreports is also the intended beneficiary. Our work investigates "two-sided" manipulation of the deferred acceptance algorithm where the misreporting agent and the manipulator (or beneficiary) are on different sides. Specifically, we generalize the recently proposed accomplice manipulation model (where a man misreports on behalf of a woman) along two complementary dimensions: (a) the two for one model, with a pair of misreporting agents (man and woman) and a single beneficiary (the misreporting woman), and (b) the one for all model, with one misreporting agent (man) and a coalition of beneficiaries (all women). Our main contribution is to develop polynomial-time algorithms for finding an optimal manipulation in both settings. We obtain these results despite the fact that an optimal one for all strategy fails to be inconspicuous, while it is unclear whether an optimal two for one strategy satisfies the inconspicuousness property. We also study the conditions under which stability of the resulting matching is preserved. Experimentally, we show that two-sided manipulations are more frequently available and offer better quality matches than their one-sided counterparts.

A variety of real-world applications rely on far future information to make decisions, thus calling for efficient and accurate long sequence multivariate time series forecasting. While recent attention-based forecasting models show strong abilities in capturing long-term dependencies, they still suffer from two key limitations. First, canonical self attention has a quadratic complexity w.r.t. the input time series length, thus falling short in efficiency. Second, different variables’ time series often have distinct temporal dynamics, which existing studies fail to capture, as they use the same model parameter space, e.g., projection matrices, for all variables’ time series, thus falling short in accuracy. To ensure high efficiency and accuracy, we propose Triformer, a triangular, variable-specific attention. (i) Linear complexity: we introduce a novel patch attention with linear complexity. When stacking multiple layers of the patch attentions, a triangular structure is proposed such that the layer sizes shrink exponentially, thus maintaining linear complexity. (ii) Variable-specific parameters: we propose a light-weight method to enable distinct sets of model parameters for different variables’ time series to enhance accuracy without compromising efficiency and memory usage. Strong empirical evidence on four datasets from multiple domains justifies our design choices, and it demonstrates that Triformer outperforms state-of-the-art methods w.r.t. both accuracy and efficiency. Source code is publicly available at https://github.com/razvanc92/triformer.

Given the ubiquity of AI-based decisions that affect individuals’ lives, providing transparent explanations about algorithms is ethically sound and often legally mandatory. How do individuals strategically adapt following explanations? What are the consequences of adaptation for algorithmic accuracy? We simulate the interplay between explanations shared by an Institution (e.g. a bank) and the dynamics of strategic adaptation by Individuals reacting to such feedback. Our model identifies key aspects related to strategic adaptation and the challenges that an institution could face as it attempts to provide explanations. Resorting to an agent-based approach, our model scrutinizes: i) the impact of transparency in explanations, ii) the interaction between faking behavior and detection capacity and iii) the role of behavior imitation. We find that the risks of transparent explanations are alleviated if effective methods to detect faking behaviors are in place. Furthermore, we observe that behavioral imitation --- as often happens across societies --- can alleviate malicious adaptation and contribute to accuracy, even after transparent explanations.

Deep generative models of sequence-structure data have attracted widespread attention in drug discovery. However, such models cannot fully extract the semantic features of molecules from sequential representations. Moreover, mode collapse reduces the diversity of the generated molecules. This paper proposes a transformer-based objective-reinforced generative adversarial network (TransORGAN) to generate molecules. TransORGAN leverages a transformer architecture as a generator and uses a stochastic policy gradient for reinforcement learning to generate plausible molecules with rich semantic features. The discriminator grants rewards that guide the policy update of the generator, while an objective-reinforced penalty encourages the generation of diverse molecules. Experiments were performed using the ZINC chemical dataset, and the results demonstrated the usefulness of TransORGAN in terms of uniqueness, novelty, and diversity of the generated molecules.

We focus on learning with limited labelled data (especially meta-learning) in conjunction with so-far under-researched multilingual textual document classification. The core principle in such learning is to achieve transferability of learned knowledge to new datasets and tasks. Currently, factors influencing the success of transfer remain mostly unclear. Their identification from experiments is challenging due to small amounts of labels making results considerably unstable. When instability of the investigated models is not explicitly taken into consideration (as is common in existing benchmarking studies), it may result in randomness possibly even invalidating the findings. We want to remedy this by in-depth exploration of factors that influence the stability and the transferability of learning with limited labelled data in multilingual textual documents classification, such as misinformation detection.

With the availability of domain specific historical negotiation data, the practical applications of machine learning techniques can prove to be increasingly effective in the field of automated negotiation. Yet a large portion of the literature focuses on domain independent negotiation and thus passes the possibility of leveraging any domain specific insights from historical data. Moreover, during sequential negotiation, utility functions may alter due to various reasons including market demand, partner agreements, weather conditions, etc. This poses a unique set of challenges and one can easily infer that one strategy that fits all is rather impossible in such scenarios. In this work, we present a simple yet effective method of learning an end-to-end negotiation strategy from historical negotiation data. Next, we show that transfer learning based solutions are effective in designing adaptive strategies when underlying utility functions of agents change. Additionally, we also propose an online method of detecting and measuring such changes in the utility functions. Combining all three contributions we propose an adaptive automated negotiating agent framework that enables the automatic creation of transfer learning based negotiating agents capable of adapting to changes in utility functions. Finally, we present the results of an agent generated using our framework in different ANAC domains with 100 different utility functions each and show that our agent outperforms the benchmark score by domain independent agents by 6%.

Semantic parsing is an important NLP problem, particularly for voice assistants such as Alexa and Google Assistant. State-of-the-art (SOTA) semantic parsers are seq2seq architectures based on large language models that have been pretrained on vast amounts of text. To better leverage that pretraining, recent work has explored a reformulation of semantic parsing whereby the output sequences are themselves natural language sentences, but in a controlled fragment of natural language. This approach delivers strong results, particularly for few-shot semantic parsing, which is of key importance in practice and the focus of our paper. We push this line of work forward by introducing an automated methodology that delivers very significant additional improvements by utilizing modest amounts of unannotated data, which is typically easy to obtain. Our method is based on a novel synthesis of four techniques: joint training with auxiliary unsupervised tasks; constrained decoding; self-training; and paraphrasing. We show that this method delivers new SOTA few-shot performance on the Overnight dataset, particularly in very low-resource settings, and very compelling few-shot results on a new semantic parsing dataset.

Collaborative filtering (CF), as a standard method for recommendation with implicit feedback, tackles a semi-supervised learning problem where most interaction data are unobserved. Such a nature makes existing approaches highly rely on mining negatives for providing correct training signals. However, mining proper negatives is not a free lunch, encountering with a tricky trade-off between mining informative hard negatives and avoiding false ones. We devise a new approach named as Hardness-Aware Debiased Contrastive Collaborative Filtering (HDCCF) to resolve the dilemma. It could sufficiently explore hard negatives from two-fold aspects: 1) adaptively sharpening the gradients of harder instances through a set-wise objective, and 2) implicitly leveraging item/user frequency information with a new sampling strategy. To circumvent false negatives, we develop a principled approach to improve the reliability of negative instances and prove that the objective is an unbiased estimation of sampling from the true negative distribution. Extensive experiments demonstrate the superiority of the proposed model over existing CF models and hard negative mining methods.

We explore the transformer neural network architecture for modeling music, specifically Irish and Swedish traditional dance music. Given the repetitive structures of these kinds of music, the transformer should be as successful with fewer parameters and complexity as the hitherto most successful model, a vanilla long short-term memory network. We find that achieving good performance with the transformer is not straightforward, and careful consideration is needed for the sampling strategy, evaluating intermediate outputs in relation to engineering choices, and finally analyzing what the model learns. We discuss these points with several illustrations, providing reusable insights for engineering other music generation systems. We also report the high performance of our final transformer model in a competition of music generation systems focused on a type of Swedish dance.

Evaluating the high-effect factors of citizens' happiness is beneficial to a wide range of policy-making for economics and politics in most countries. Benefiting from the high-efficiency of regression models, previous efforts by sociology scholars have analyzed the effect of happiness factors with high interpretability. However, restricted to their research concerns, they are specifically interested in some subset of factors modeled as linear functions. Recently, deep learning shows promising prediction accuracy while addressing challenges in interpretability. To this end, we introduce Shapley value that is inherent in solid theory for factor contribution interpretability to work with deep learning models by taking into account interactions between multiple factors. The proposed solution computes the Shapley value of a factor, i.e., its average contribution to the prediction in different coalitions based on coalitional game theory. Aiming to evaluate the interpretability quality of our solution, experiments are conducted on a Chinese General Social Survey (CGSS) questionnaire dataset. Through systematic reviews, the experimental results of Shapley value are highly consistent with academic studies in social science, which implies our solution for citizens' happiness prediction has 2-fold implications, theoretically and practically.

Federated learning (FL) is an emerging paradigm of collaborative machine learning that preserves user privacy while building powerful models. Nevertheless, due to the nature of open participation by self-interested entities, it needs to guard against potential misbehaviours by legitimate FL participants. FL verification techniques are promising solutions for this problem. They have been shown to effectively enhance the reliability of FL networks and build trust among participants. Verifiable FL has become an emerging topic of research that has attracted significant interest from the academia and the industry alike. Currently, there is no comprehensive survey on the field of verifiable federated learning, which is interdisciplinary in nature and can be challenging for researchers to enter into. In this paper, we bridge this gap by reviewing works focusing on verifiable FL. We propose a novel taxonomy for verifiable FL covering both centralised and decentralised settings, summarise the commonly adopted performance evaluation approaches, and discuss promising directions towards a versatile verifiable FL framework.

Conventional model upgrades for visual search systems require offline refresh of gallery features by feeding gallery images into new models (dubbed as “backfill”), which is time-consuming and expensive, especially in large-scale applications. The task of backward-compatible representation learning is therefore introduced to support backfill-free model upgrades, where the new query features are interoperable with the old gallery features. Despite the success, previous works only investigated a close-set training scenario (i.e., the new training set shares the same classes as the old one), and are limited by more realistic and challenging open-set scenarios. To this end, we first introduce a new problem of universal backward-compatible representation learning, covering all possible data split in model upgrades. We further propose a simple yet effective method, dubbed as Universal Backward-Compatible Training (UniBCT) with a novel structural prototype refinement algorithm, to learn compatible representations in all kinds of model upgrading benchmarks in a unified manner. Comprehensive experiments on the large-scale face recognition datasets MS1Mv3 and IJB-C fully demonstrate the effectiveness of our method. Source code is available at https://github.com/TencentARC/OpenCompatible.

Machine learning tasks are often formulated as complex optimization problems, where the objective function can be non-differentiable, non-continuous, non-unique, inaccurate, dynamic, and have many local optima, making conventional optimization algorithms fail. Evolutionary Algorithms (EAs), inspired by Darwin's theory of evolution, are general-purpose randomized heuristic optimization algorithms, mimicking variational reproduction and natural selection. EAs have yielded encouraging outcomes for solving complex optimization problems (e.g., neural architecture search) in machine learning. However, due to the heuristic nature of EAs, most outcomes to date have been empirical and lack theoretical support, encumbering their acceptance to the general machine learning community. In this paper, I will review the progress towards theoretically grounded evolutionary learning, from the aspects of analysis methodology, theoretical perspectives and learning algorithms. Due to space limit, I will include a few representative examples and highlight our contributions. I will also discuss some future challenges.

Though deep reinforcement learning (DRL) has obtained substantial success, it may encounter catastrophic failures due to the intrinsic uncertainty of both transition and observation. Most of the existing methods for safe reinforcement learning can only handle transition disturbance or observation disturbance since these two kinds of disturbance affect different parts of the agent; besides, the popular worst-case return may lead to overly pessimistic policies. To address these issues, we first theoretically prove that the performance degradation under transition disturbance and observation disturbance depends on a novel metric of Value Function Range (VFR), which corresponds to the gap in the value function between the best state and the worst state. Based on the analysis, we adopt conditional value-at-risk (CVaR) as an assessment of risk and propose a novel reinforcement learning algorithm of CVaR-Proximal-Policy-Optimization (CPPO) which formalizes the risk-sensitive constrained optimization problem by keeping its CVaR under a given threshold. Experimental results show that CPPO achieves a higher cumulative reward and is more robust against both observation and transition disturbances on a series of continuous control tasks in MuJoCo.

Disentangled sequential autoencoders (DSAEs) represent a class of probabilistic graphical models that describes an observed sequence with dynamic latent variables and a static latent variable. The former encode information at a frame rate identical to the observation, while the latter globally governs the entire sequence. This introduces an inductive bias and facilitates unsupervised disentanglement of the underlying local and global factors. In this paper, we show that the vanilla DSAE suffers from being sensitive to the choice of model architecture and capacity of the dynamic latent variables, and is prone to collapse the static latent variable. As a countermeasure, we propose TS-DSAE, a two-stage training framework that first learns sequence-level prior distributions, which are subsequently employed to regularise the model and facilitate auxiliary objectives to promote disentanglement. The proposed framework is fully unsupervised and robust against the global factor collapse problem across a wide range of model configurations. It also avoids typical solutions such as adversarial training which usually involves laborious parameter tuning, and domain-specific data augmentation. We conduct quantitative and qualitative evaluations to demonstrate its robustness in terms of disentanglement on both artificial and real-world music audio datasets.

Dense retrieval (DR) has extended the employment of pre-trained language models, like BERT, for text ranking. However, recent studies have raised the robustness issue of DR model against query variations, like query with typos, along with non-trivial performance losses. Herein, we argue that it would be beneficial to allow the DR model to learn to align the relative positions of query-passage pairs in the representation space, as query variations cause the query vector to drift away from its original position, affecting the subsequent DR effectiveness. To this end, we propose RoDR, a novel robust DR model that learns to calibrate the in-batch local ranking of query variation to that of original query for the DR space alignment. Extensive experiments on MS MARCO and ANTIQUE datasets show that RoDR significantly improves the retrieval results on both the original queries and different types of query variations. Meanwhile, RoDR provides a general query noise-tolerate learning framework that boosts the robustness and effectiveness of various existing DR models. Our code and models are openly available at https://github.com/cxa-unique/RoDR.