IJCAI 2023 Papers

Existential rules, also known as tuple-generating dependencies (TGDs) or Datalog+/- rules, are heavily studied in the communities of Knowledge Representation and Reasoning, Semantic Web, and Databases, due to their rich modelling capabilities. In this paper we consider TGDs in the temporal setting, by introducing and studying DatalogMTLE---an extension of metric temporal Datalog (DatalogMTL) obtained by allowing for existential rules in programs. We show that DatalogMTLE is undecidable even in the restricted cases of guarded and weakly-acyclic programs. To address this issue we introduce uniform semantics which, on the one hand, is well-suited for modelling temporal knowledge as it prevents from unintended value invention and, on the other hand, provides decidability of reasoning; in particular, it becomes 2-EXPSPACE-complete for weakly-acyclic programs but remains undecidable for guarded programs. We provide an implementation for the decidable case and demonstrate its practical feasibility. Thus we obtain an expressive, yet decidable, rule-language and a system which is suitable for complex temporal reasoning with existential rules.

Human pose forecasting is a sequential modeling task that aims to predict future poses from historical motions. Most existing approaches focus on the spatial-temporal neural network model design for learning movement patterns to reduce prediction errors. However, they usually do not strictly follow the temporal constraints in the inference stage. Even though a small Mean Per Joint Position Error (MPJPE) is achieved, some of the predicted poses are not temporal feasible solutions, which disobeys the continuity of the body movement. In this paper, we consider the temporal constrained feasible solutions for human pose forecasting, where the predicted poses of input historical poses are guaranteed to obey the temporal constraints strictly in the inference stage. Rather than direct supervision of the prediction in the original pose space, a temporal constrained subspace is explicitly learned and then followed by an inverse transformation to obtain the final predictions. We evaluate the proposed method on large-scale benchmarks, including Human3.6M, AMASS, and 3DPW. State-of-the-art performance has been achieved with the temporal constrained feasible solutions.

In real teaching scenarios, an excellent teacher always teaches what he (or she) is good at but the student is not. This gives the student the best assistance in making up for his (or her) weaknesses and becoming a good one overall. Enlightened by this, we introduce the "Teaching what you Should Teach" strategy into a knowledge distillation framework, and propose a data-based distillation method named "TST" that searches for desirable augmented samples to assist in distilling more efficiently and rationally. To be specific, we design a neural network-based data augmentation module with priori bias to find out what meets the teacher's strengths but the student's weaknesses, by learning magnitudes and probabilities to generate suitable data samples. By training the data augmentation module and the generalized distillation paradigm alternately, a student model is learned with excellent generalization ability. To verify the effectiveness of our method, we conducted extensive comparative experiments on object recognition, detection, and segmentation tasks. The results on the CIFAR-100, ImageNet-1k, MS-COCO, and Cityscapes datasets demonstrate that our method achieves state-of-the-art performance on almost all teacher-student pairs. Furthermore, we conduct visualization studies to explore what magnitudes and probabilities are needed for the distillation process.

Sleep stage classification is of great significance to the diagnosis of sleep disorders. However, existing sleep stage classification models based on deep learning are usually relatively large in size (wider and deeper), which makes them hard to be deployed on wearable devices. Therefore, it is a challenge to lighten the existing sleep stage classification models. In this paper, we propose a novel general knowledge distillation framework for sleep stage classification tasks called SleepKD. Our SleepKD, composed of the multi-level module, teacher assistant module, and other knowledge distillation modules, aims to lighten large-scale sleep stage classification models. Specifically, the multi-level module is able to transfer the multi-level knowledge extracted from sleep signals by the teacher model (large-scale model) to the student model (lightweight model). Moreover, the teacher assistant module bridges the large gap between the teacher and student network, and further improves the distillation. We evaluate our method on two public sleep datasets (Sleep-EDF and ISRUC-III). Compared to the baseline methods, the results show that our knowledge distillation framework achieves state-of-the-art performance. SleepKD can significantly lighten the sleep model while maintaining its classification performance. The source code is available at https://github.com/HychaoWang/SleepKD.

Text-driven 3D style transfer aims at stylizing a scene according to the text and generating arbitrary novel views with consistency. Simply combining image/video style transfer methods and novel view synthesis methods results in flickering when changing viewpoints, while existing 3D style transfer methods learn styles from images instead of texts. To address this problem, we for the first time design an efficient text-driven model for 3D style transfer, named TeSTNeRF, to stylize the scene using texts via cross-modal learning: we leverage an advanced text encoder to embed the texts in order to control 3D style transfer and align the input text and output stylized images in latent space. Furthermore, to obtain better visual results, we introduce style supervision, learning feature statistics from style images and utilizing 2D stylization results to rectify abrupt color spill. Extensive experiments demonstrate that TeSTNeRF significantly outperforms existing methods and provides a new way to guide 3D style transfer.

We study a single task allocation problem where each worker connects to some other workers to form a network and the task requester only connects to some of the workers. The goal is to design an allocation mechanism such that each worker is incentivized to invite her neighbours to join the allocation, although they are competing for the task. Moreover, the performance of each worker is uncertain, which is modelled as the quality level of her task execution. The literature has proposed solutions to tackle the uncertainty problem by paying them after verifying their execution. Here, we extend the problem to the network setting. We propose a new mechanism that guarantees that inviting more workers and reporting/performing according to her true ability is a dominant strategy for each worker. We believe that the new solution can be widely applied in the digital economy powered by social connections such as crowdsourcing.

The computation of minimal rare itemsets is a well known task in data mining, with numerous applications, e.g., drugs effects analysis and network security, among others. This paper presents a novel approach to the computation of minimal rare itemsets. First, we introduce a generalization of the traditional minimal rare itemset model called k-minimal rare itemset. A k-minimal rare itemset is defined as an itemset that becomes frequent or rare based on the removal of at least k or at most (k − 1) items from it. We claim that our work is the first to propose this generalization in the field of data mining. We then present a SAT-based framework for efficiently discovering k-minimal rare itemsets from large transaction databases. Afterwards, by partitioning the k-minimal rare itemset mining problem into smaller sub-problems, we aim to make it more manageable and easier to solve. Finally, to evaluate the effectiveness and efficiency of our approach, we conduct extensive experimental analysis using various popular datasets. We compare our method with existing specialized algorithms and CP-based algorithms commonly used for this task.

Text irregularities pose significant challenges to scene text recognizers. Thin-Plate Spline (TPS)-based rectification is widely regarded as an effective means to deal with them. Currently, the calculation of TPS transformation parameters purely depends on the quality of regressed text borders. It ignores the text content and often leads to unsatisfactory rectified results for severely distorted text. In this work, we introduce TPS++, an attention-enhanced TPS transformation that incorporates the attention mechanism to text rectification for the first time. TPS++ formulates the parameter calculation as a joint process of foreground control point regression and content-based attention score estimation, which is computed by a dedicated designed gated-attention block. TPS++ builds a more flexible content-aware rectifier, generating a natural text correction that is easier to read by the subsequent recognizer. Moreover, TPS++ shares the feature backbone with the recognizer in part and implements the rectification at feature-level rather than image-level, incurring only a small overhead in terms of parameters and inference time. Experiments on public benchmarks show that TPS++ consistently improves the recognition and achieves state-of-the-art accuracy. Meanwhile, it generalizes well on different backbones and recognizers. Code is at https://github.com/simplify23/TPS_PP.

In multi-domain task-oriented dialogue systems, users proactively propose a series of domain-specific requests that can often be under-or over-specified, sometimes with ambiguous and cross-domain demands. System-sided initiative would be necessary to identify certain situations and appropriately interact with users to resolve them. However, most existing task-oriented dialogue systems fail to consider such mixed-initiative interaction strategies, performing low efficiency and poor collaboration ability in human-computer conversation. In this paper, we construct a multi-domain task-oriented dialogue dataset with mixed-initiative strategies named TITAN from the large-scale dialogue corpus MultiWOZ 2.1. It contains a total of 1,800 human-human conversations where the system can either ask clarification questions actively or provides relevant information to address failure situations and implicit user requests. We report the results of several baseline models on system response generation and dialogue act prediction to assess the performance of SOTA methods on TITAN. These models can capture mixed-initiative dialogue acts, while remaining the deficiency to actively generate implicit requests and accurately provide alternative information, suggesting ample room for improvement in future studies.

Video question answering aims at answering a question about the video content by reasoning the alignment semantics within them. However, since relying heavily on human instructions, i.e., annotations or priors, current contrastive learning-based VideoQA methods remains challenging to perform fine-grained visual-linguistic alignments. In this work, we innovatively resort to game theory, which can simulate complicated relationships among multiple players with specific interaction strategies, e.g., video, question, and answer as ternary players, to achieve fine-grained alignment for VideoQA task. Specifically, we carefully design a VideoQA-specific interaction strategy to tailor the characteristics of VideoQA, which can mathematically generate the fine-grained visual-linguistic alignment label without label-intensive efforts. Our TG-VQA outperforms existing state-of-the-art by a large margin (more than 5%) on long-term and short-term VideoQA datasets, verifying its effectiveness and generalization ability. Thanks to the guidance of game-theoretic interaction, our model impressively convergences well on limited data (10^4 videos), surpassing most of those pre-trained on large-scale data (10^7 videos).

In crowdsourcing, existing efforts mainly use real datasets collected from crowdsourcing as test datasets to evaluate the effectiveness of aggregation algorithms. However, these work ignore the fact that the datasets obtained by crowdsourcing are usually sparse and imbalanced due to limited budget. As a result, applying the same aggregation algorithm on different datasets often show contradicting conclusions. For example, on the RTE dataset, Dawid and Skene model performs significantly better than Majority Voting, while on the LableMe dataset, the experiments give the opposite conclusion. It is challenging to obtain comprehensive and balanced datasets at a low cost. To our best knowledge, little effort have been made to the fair evaluation of aggregation algorithms. To fill in this gap, we propose a novel method named TDG4Crowd that can automatically generate comprehensive and balanced datasets. Using Kullback Leibler divergence and Kolmogorov–Smirnov test, the experiment results show the superior of our method compared with others. Aggregation algorithms also perform more consistently on the synthetic datasets generated using our method.

We consider the problem of synthesizing resilient and stochastically stable strategies for systems of cooperating agents striving to minimize the expected time between consecutive visits to selected locations in a known environment. A strategy profile is resilient if it retains its functionality even if some of the agents fail, and stochastically stable if the visiting time variance is small. We design a novel specification language for objectives involving resilience and stochastic stability, and we show how to efficiently compute strategy profiles (for both autonomous and coordinated agents) optimizing these objectives. Our experiments show that our strategy synthesis algorithm can construct highly non-trivial and efficient strategy profiles for environments with general topology.

Stream Learning (SL) attempts to learn from a data stream efficiently. A data stream learning algorithm should adapt to input data distribution shifts without sacrificing accuracy. These distribution shifts are known as ”concept drifts” in the literature. SL provides many supervised, semi-supervised, and unsupervised methods for detecting and adjusting to concept drift. On the other hand, Continual Learning (CL) attempts to preserve previous knowledge while performing well on the current concept when confronted with concept drift. In Online Continual Learning (OCL), this learning happens online. This survey explores the intersection of those two online learning paradigms to find synergies. We identify this intersection as Online Streaming Continual Learning (OSCL). The study starts with a gentle introduction to SL and then explores CL. Next, it explores OSCL from SL and OCL perspectives to point out new research trends and give directions for future research.

We introduce in this survey the major concepts, models, and algorithms proposed so far to infer causal relations from observational time series, a task usually referred to as causal discovery in time series. To do so, after a description of the underlying concepts and modelling assumptions, we present different methods according to the family of approaches they belong to: Granger causality, constraint-based approaches, noise-based approaches, score-based approaches, logic-based approaches, topology-based approaches, and difference-based approaches. We then evaluate several representative methods to illustrate the behaviour of different families of approaches. This illustration is conducted on both artificial and real datasets, with different characteristics. The main conclusions one can draw from this survey is that causal discovery in times series is an active research field in which new methods (in every family of approaches) are regularly proposed, and that no family or method stands out in all situations. Indeed, they all rely on assumptions that may or may not be appropriate for a particular dataset.

Meeting the UN's objective of developing sustainable agriculture requires, in particular, accompanying small farms in their agroecological transition. This transition often requires making the agrosystem more complex and increasing the number of crops to increase biodiversity and ecosystem services. This paper introduces a flexible model based on Constraint Programming (CP) to address the crop allocation problem. This problem takes a cropping calendar as input and aims at allocating crops to respect several constraints. We have shown that it is possible to model both agroecological and operational constraints at the level of a small farm. Experiments on an organic micro-farm have shown that it is possible to combine these constraints to design very different cropping scenarios and that our approach can apply to real situations. Our promising results in this case study also demonstrate the potential of AI-based tools to address small farmers' challenges in the context of the sustainable agriculture transition.

We present a novel recommendation system designed to provide real-time treatment strategies to therapists during psychotherapy sessions. Our system utilizes a turn-level rating mechanism that forecasts the therapeutic outcome by calculating a similarity score between the profound representation of a scoring inventory and the patient's current spoken sentence. By transcribing and segmenting the continuous audio stream into patient and therapist turns, our system conducts immediate evaluation of their therapeutic working alliance. The resulting dialogue pairs, along with their computed working alliance ratings, are then utilized in a deep reinforcement learning recommendation system. In this system, the sessions are treated as users, while the topics are treated as items. To showcase the system's effectiveness, we not only evaluate its performance using an existing dataset of psychotherapy sessions but also demonstrate its practicality through a web app. Through this demo, we aim to provide a tangible and engaging experience of our recommendation system in action.

This article proposes a novel spectral domain based solution to the challenging polyp segmentation. The main contribution is based on an interesting finding of the significant existence of the middle frequency sub-band during the CNN process. Consequently, a Sub-Band based Attention (SBA) module is proposed, which uniformly adopts either the high or middle sub-bands of the encoder features to boost the decoder features and thus concretely improve the feature discrimination. A strong encoder supplying informative sub-bands is also very important, while we highly value the local-and-global information enriched CNN features. Therefore, a Transformer Attended Convolution (TAC) module as the main encoder block is introduced. It takes the Transformer features to boost the CNN features with stronger long-range object contexts. The combination of SBA and TAC leads to a novel polyp segmentation framework, SBA-Net. It adopts TAC to effectively obtain encoded features which also input to SBA, so that efficient sub-bands based attention maps can be generated for progressively decoding the bottleneck features. Consequently, SBA-Net can achieve the robust polyp segmentation, as the experimental results demonstrate.

Learning causal structure among event types from discrete-time event sequences is a particularly important but challenging task. Existing methods, such as the multivariate Hawkes processes based methods, mostly boil down to learning the so-called Granger causality which assumes that the cause event happens strictly prior to its effect event. Such an assumption is often untenable beyond applications, especially when dealing with discrete-time event sequences in low-resolution; and typical discrete Hawkes processes mainly suffer from identifiability issues raised by the instantaneous effect, i.e., the causal relationship that occurred simultaneously due to the low-resolution data will not be captured by Granger causality. In this work, we propose Structure Hawkes Processes (SHPs) that leverage the instantaneous effect for learning the causal structure among events type in discrete-time event sequence. The proposed method is featured with the Expectation-Maximization of the likelihood function and a sparse optimization scheme. Theoretical results show that the instantaneous effect is a blessing rather than a curse, and the causal structure is identifiable under the existence of the instantaneous effect. Experiments on synthetic and real-world data verify the effectiveness of the proposed method.

As a long-standing task, image restoration aims to recover the latent sharp image from its degraded counterpart. In recent years, owing to the strong ability of self-attention in capturing long-range dependencies, Transformer based methods have achieved promising performance on multifarious image restoration tasks. However, the canonical self-attention leads to quadratic complexity with respect to input size, hindering its further applications in image restoration. In this paper, we propose a Strip Attention Network (SANet) for image restoration to integrate information in a more efficient and effective manner. Specifically, a strip attention unit is proposed to harvest the contextual information for each pixel from its adjacent pixels in the same row or column. By employing this operation in different directions, each location can perceive information from an expanded region. Furthermore, we apply various receptive fields in different feature groups to enhance representation learning. Incorporating these designs into a U-shaped backbone, our SANet performs favorably against state-of-the-art algorithms on several image restoration tasks. The code is available at https://github.com/c-yn/SANet.