IJCAI 2022 论文列表

Movement and pose assessment of newborns lets experienced pediatricians predict neurodevelopmental disorders, allowing early intervention for related diseases. However, most of the newest AI approaches for human pose estimation methods focus on adults, lacking publicly benchmark for infant pose estimation. In this paper, we fill this gap by proposing infant pose dataset and Deep Aggregation Vision Transformer for human pose estimation, which introduces a fast trained full transformer framework without using convolution operations to extract features in the early stages. It generalizes Transformer + MLP to high-resolution deep layer aggregation within feature maps, thus enabling information fusion between different vision levels. We pre-train AggPose on COCO pose dataset and apply it on our newly released large-scale infant pose estimation dataset. The results show that AggPose could effectively learn the multi-scale features among different resolutions and significantly improve the performance of infant pose estimation. We show that AggPose outperforms hybrid model HRFormer and TokenPose in the infant pose estimation dataset. Moreover, our AggPose outperforms HRFormer by 0.8 AP on COCO val pose estimation on average. Our code is available at github.com/SZAR-LAB/AggPose.

We propose a novel black-box approach for performing adversarial attacks against knowledge graph embedding models. An adversarial attack is a small perturbation of the data at training time to cause model failure at test time. We make use of an efficient rule learning approach and use abductive reasoning to identify triples which are logical explanations for a particular prediction. The proposed attack is then based on the simple idea to suppress or modify one of the triples in the most confident explanation. Although our attack scheme is model independent and only needs access to the training data, we report results on par with state-of-the-art white-box attack methods that additionally require full access to the model architecture, the learned embeddings, and the loss functions. This is a surprising result which indicates that knowledge graph embedding models can partly be explained post hoc with the help of symbolic methods.

Domain adaptation (DA) aims to transfer the knowledge of a well-labeled source domain to facilitate unlabeled target learning. When turning to specific tasks such as indoor (Wi-Fi) localization, it is essential to learn a cross-domain regressor to mitigate the domain shift. This paper proposes a novel method Adversarial Bi-Regressor Network (ABRNet) to seek more effective cross- domain regression model. Specifically, a discrepant bi-regressor architecture is developed to maximize the difference of bi-regressor to discover uncertain target instances far from the source distribution, and then an adversarial training mechanism is adopted between feature extractor and dual regressors to produce domain-invariant representations. To further bridge the large domain gap, a domain- specific augmentation module is designed to synthesize two source-similar and target-similar inter- mediate domains to gradually eliminate the original domain mismatch. The empirical studies on two cross-domain regressive benchmarks illustrate the power of our method on solving the domain adaptive regression (DAR) problem.

Stock investment selection is a hard issue in the Fintech field due to non-stationary dynamics and complex market interdependencies. Existing studies are mostly based on RNNs, which struggle to capture interactive information among fine granular volatility patterns. Besides, they either treat stocks as isolated, or presuppose a fixed graph structure heavily relying on prior domain knowledge. In this paper, we propose a novel Adaptive Long-Short Pattern Transformer (ALSP-TF) for stock ranking in terms of expected returns. Specifically, we overcome the limitations of canonical self-attention including context and position agnostic, with two additional capacities: (i) fine-grained pattern distiller to contextualize queries and keys based on localized feature scales, and (ii) time-adaptive modulator to let the dependency modeling among pattern pairs sensitive to different time intervals. Attention heads in stacked layers gradually harvest short- and long-term transition traits, spontaneously boosting the diversity of representations. Moreover, we devise a graph self-supervised regularization, which helps automatically assimilate the collective synergy of stocks and improve the generalization ability of overall model. Experiments on three exchange market datasets show ALSP-TF’s superiority over state-of-the-art stock forecast methods.

Reasoning about uncertainty is vital in many real-life autonomous systems. However, current state-of-the-art planning algorithms either cannot reason about uncertainty explicitly, or do so with high computational burden. Here, we focus on making informed decisions efficiently, using reward functions that explicitly deal with uncertainty. We formulate an approximation, namely an abstract observation model, that uses an aggregation scheme to alleviate computational costs. We derive bounds on the expected information-theoretic reward function and, as a consequence, on the value function. We then propose a method to refine aggregation to achieve identical action selection in a fraction of the computational time.

Inspired by the great success of deep neural networks, learning-based methods have gained promising performances for metal artifact reduction (MAR) in computed tomography (CT) images. However, most of the existing approaches put less emphasis on modelling and embedding the intrinsic prior knowledge underlying this specific MAR task into their network designs. Against this issue, we propose an adaptive convolutional dictionary network (ACDNet), which leverages both model-based and learning-based methods. Specifically, we explore the prior structures of metal artifacts, e.g., non-local repetitive streaking patterns, and encode them as an explicit weighted convolutional dictionary model. Then, a simple-yet-effective algorithm is carefully designed to solve the model. By unfolding every iterative substep of the proposed algorithm into a network module, we explicitly embed the prior structure into a deep network , i.e., a clear interpretability for the MAR task. Furthermore, our ACDNet can automatically learn the prior for artifact-free CT images via training data and adaptively adjust the representation kernels for each input CT image based on its content. Hence, our method inherits the clear interpretability of model-based methods and maintains the powerful representation ability of learning-based methods. Comprehensive experiments executed on synthetic and clinical datasets show the superiority of our ACDNet in terms of effectiveness and model generalization. Code and supplementary material are available at https://github.com/hongwang01/ACDNet.

Traditional scheduling techniques suffer from a lack of flexibility. The problem's instances need to be deterministic, and results on datasets with small benchmark instances do usually not transfer to large-scale instances. We propose to develop adaptive algorithms that can leverage the similarities between instances of industrial scheduling problems. In particular, we focus on applications of modern machine learning techniques to combinatorial optimization problems, an emerging and promising research area. Traditional scheduling techniques such as constraint, mixed-integer, or answer set programming are highly generic, domain-independent, and, therefore, do not explicitly exploit the specificities of a problem domain. However, in a production facility, the settings between two consecutive schedules are often very similar. The machines, workers, production capacity, etc., usually stay the same or do not change significantly. Traditional scheduling techniques do not take advantage of such similarities, while machine learning, especially deep learning, can discover and exploit relationships in the data. Therefore, our research aims to incorporate machine learning into combinatorial optimization.

Conventional federated learning (FL) trains one global model for a federation of clients with decentralized data, reducing the privacy risk of centralized training. However, the distribution shift across non-IID datasets, often poses a challenge to this one-model-fits-all solution. Personalized FL aims to mitigate this issue systematically. In this work, we propose APPLE, a personalized cross-silo FL framework that adaptively learns how much each client can benefit from other clients’ models. We also introduce a method to flexibly control the focus of training APPLE between global and local objectives. We empirically evaluate our method's convergence and generalization behaviors, and perform extensive experiments on two benchmark datasets and two medical imaging datasets under two non-IID settings. The results show that the proposed personalized FL framework, APPLE, achieves state-of-the-art performance compared to several other personalized FL approaches in the literature. The code is publicly available at https://github.com/ljaiverson/pFL-APPLE.

In Neural Machine Translation (NMT), data augmentation methods such as back-translation have proven their effectiveness in improving translation performance. In this paper, we propose a novel data augmentation approach for NMT, which is independent of any additional training data. Our approach, AdMix, consists of two parts: 1) introduce faint discrete noise (word replacement, word dropping, word swapping) into the original sentence pairs to form augmented samples; 2) generate new synthetic training data by softly mixing the augmented samples with their original samples in training corpus. Experiments on three translation datasets of different scales show that AdMix achieves significant improvements (1.0 to 2.7 BLEU points) over strong Transformer baseline. When combined with other data augmentation techniques (e.g., back-translation), our approach can obtain further improvements.

The recent success of audio-visual representation learning can be largely attributed to their pervasive property of audio-visual synchronization, which can be used as self-annotated supervision. As a state-of-the-art solution, Audio-Visual Instance Discrimination (AVID) extends instance discrimination to the audio-visual realm. Existing AVID methods construct the contrastive set by random sampling based on the assumption that the audio and visual clips from all other videos are not semantically related. We argue that this assumption is rough, since the resulting contrastive sets have a large number of faulty negatives. In this paper, we overcome this limitation by proposing a novel Active Contrastive Set Mining (ACSM) that aims to mine the contrastive sets with informative and diverse negatives for robust AVID. Moreover, we also integrate a semantically-aware hard-sample mining strategy into our ACSM. The proposed ACSM is implemented into two most recent state-of-the-art AVID methods and significantly improves their performance. Extensive experiments conducted on both action and sound recognition on multiple datasets show the remarkably improved performance of our method.

We study the problem of allocating indivisible goods among agents in a fair manner. While envy-free allocations of indivisible goods are not guaranteed to exist, envy-freeness can be achieved by additionally providing some subsidy to the agents. These subsidies can be alternatively viewed as a divisible good (money) that is fractionally assigned among the agents to realize an envy-free outcome. In this setup, we bound the subsidy required to attain envy-freeness among agents with dichotomous valuations, i.e., among agents whose marginal value for any good is either zero or one. We prove that, under dichotomous valuations, there exists an allocation that achieves envy-freeness with a per-agent subsidy of either 0 or 1. Furthermore, such an envy-free solution can be computed efficiently in the standard value-oracle model. Notably, our results hold for general dichotomous valuations and, in particular, do not require the (dichotomous) valuations to be additive, submodular, or even subadditive. Also, our subsidy bounds are tight and provide a linear (in the number of agents) factor improvement over the bounds known for general monotone valuations.

We consider nonconvex optimization problem with constraint that is a product of simplices. A commonly used algorithm in solving this type of problem is the Multiplicative Weights Update (MWU), an algorithm that is widely used in game theory, machine learning and multi agent systems. Despite it has been known that MWU avoids saddle points, there is a question that remains unaddressed: ``Is there an accelerated version of MWU that avoids saddle points provably?'' In this paper we provide a positive answer to above question. We provide an accelerated MWU based on Riemannian Accelerated Gradient Descent, and prove that the Riemannian Accelerated Gradient Descent, thus the accelerated MWU, avoid saddle points.

When a robot adapts a learned task for a novel environment, any changes to objects in the novel environment have an unknown effect on its task execution. For example, replacing an object in a pick-and-place task affects where the robot should target its actions, but does not necessarily affect the underlying action model. In contrast, replacing a tool that the robot will use to complete a task will effectively alter its end-effector pose with respect to the robot's base coordinate system, and thus the robot's motion must be replanned accordingly. These examples highlight the relationship among (i) differences between the source and target environments, (ii) the level of abstraction at which a robot's task model should be represented to enable transfer to the target environment, and (iii) the information needed to ground the abstracted task representation in the target environment. In this abstract, summarizing our full article [Fitzgerald et al., 2021], we present our taxonomy of transfer problems based on this relationship. We also describe a knowledge representation called the Tiered Task Abstraction (TTA) and demonstrate its applicability to a variety of transfer problems in the taxonomy. Our experimental results indicate a trade-off between the generality and data requirements of a task representation, and reinforce the need for multiple transfer methods that operate at different levels of abstraction.

Abstraction is a powerful technique that has not been considered much for nonmonotonic reasoning formalisms including Answer Set Programming (ASP), apart from related simplification methods. We introduce a notion for abstracting from the domain of an ASP program that shrinks the domain size and over-approximates the set of answer sets, as well as an abstraction-&-refinement methodology that, starting from an initial abstraction, automatically yields an abstraction with an associated answer set matching an answer set of the original program if one exists. Experiments reveal the potential of the approach, by its ability to focus on the program parts that cause unsatisfiability and by achieving concrete abstract answer sets that merely reflect relevant details.

We characterise the problem of abstraction in the context of deep reinforcement learning. Various well established approaches to analogical reasoning and associative memory might be brought to bear on this issue, but they present difficulties because of the need for end-to-end differentiability. We review developments in AI and machine learning that could facilitate their adoption.

In early years, paraphrase generation typically adopts rule-based methods, which are interpretable and able to make global transformations to the original sentence. But they struggle to produce fluent paraphrases. Recently, deep neural networks have shown impressive performances in generating paraphrases. However, the current neural models are black boxes and are prone to make local modifications to the inputs. In this work, we combine these two approaches into RULER, a novel approach that performs abstract rule learning for paraphrasing. The key idea is to explicitly learn generalizable rules that could enhance the paraphrase generation process of neural networks. In RULER, we first propose a rule generalizability metric to guide the model to generate rules underlying the paraphrasing. Then, we leverage neural networks to generate paraphrases by refining the sentences transformed by the learned rules. Extensive experimental results demonstrate the superiority of RULER over previous state-of-the-art methods in terms of paraphrase quality, generalization ability and interpretability.

In the context of probabilistic AAFs, we intro- duce AAFs with marginal probabilities (mAAFs) requiring only marginal probabilities of argu- ments/attacks to be specified and not relying on the independence assumption. Reasoning over mAAFs requires taking into account multiple probability distributions over the possible worlds, so that the probability of extensions is not determined by a unique value, but by an interval. We focus on the problems of computing the max and min probabil- ities of extensions over mAAFs under Dung’s se- mantics, characterize their complexity, and provide closed formulas for polynomial cases.

Weakly supervised object detection (WSOD) is a challenging task, in which image-level labels (e.g., categories of the instances in the whole image) are used to train an object detector. Many existing methods follow the standard multiple instance learning (MIL) paradigm and have achieved promising performance. However, the lack of deterministic information leads to part domination and missing instances. To address these issues, this paper focuses on identifying and fully exploiting the deterministic information in WSOD. We discover that negative instances (i.e. absolutely wrong instances), ignored in most of the previous studies, normally contain valuable deterministic information. Based on this observation, we here propose a negative deterministic information (NDI) based method for improving WSOD, namely NDI-WSOD. Specifically, our method consists of two stages: NDI collecting and exploiting. In the collecting stage, we design several processes to identify and distill the NDI from negative instances online. In the exploiting stage, we utilize the extracted NDI to construct a novel negative contrastive learning mechanism and a negative guided instance selection strategy for dealing with the issues of part domination and missing instances, respectively. Experimental results on several public benchmarks including VOC 2007, VOC 2012 and MS COCO show that our method achieves satisfactory performance.

Recently users’ right-to-be-forgotten is stipulated by many laws and regulations. However, only removing the data from the dataset is not enough, as machine learning models would memorize the training data once the data is involved in model training, increasing the risk of exposing users’ privacy. To solve this problem, currently, the straightforward method, naive retraining, is to discard these data and retrain the model from scratch, which is reliable but brings much computational and time overhead. In this paper, we propose an exact unlearning architecture called ARCANE. Based on ensemble learning, we transform the naive retraining into multiple one-class classification tasks to reduce retraining cost while ensuring model performance, especially in the case of a large number of unlearning requests not considered by previous works. Then we further introduce data preprocessing methods to reduce the retraining overhead and speed up the unlearning, which includes representative data selection for redundancy removal, training state saving to reuse previous calculation results, and sorting to cope with unlearning requests of different distributions. We extensively evaluate ARCANE on three typical datasets with three common model architectures. Experiment results show the effectiveness and superiority of ARCANE over both the naive retraining and the state-of-the-art method in terms of model performance and unlearning speed.