AAAI 2023 论文列表

Skeletal motions have been heavily relied upon for human activity recognition (HAR). Recently, a universal vulnerability of skeleton-based HAR has been identified across a variety of classifiers and data, calling for mitigation. To this end, we propose the first black-box defense method for skeleton-based HAR to our best knowledge. Our method is featured by full Bayesian treatments of the clean data, the adversaries and the classifier, leading to (1) a new Bayesian Energy-based formulation of robust discriminative classifiers, (2) a new adversary sampling scheme based on natural motion manifolds, and (3) a new post-train Bayesian strategy for black-box defense. We name our framework Bayesian Energy-based Adversarial Training or BEAT. BEAT is straightforward but elegant, which turns vulnerable black-box classifiers into robust ones without sacrificing accuracy. It demonstrates surprising and universal effectiveness across a wide range of skeletal HAR classifiers and datasets, under various attacks. Appendix and code are available.

Vision Transformers (ViTs) have a radically different architecture with significantly less inductive bias than Convolutional Neural Networks. Along with the improvement in performance, security and robustness of ViTs are also of great importance to study. In contrast to many recent works that exploit the robustness of ViTs against adversarial examples, this paper investigates a representative causative attack, i.e., backdoor. We first examine the vulnerability of ViTs against various backdoor attacks and find that ViTs are also quite vulnerable to existing attacks. However, we observe that the clean-data accuracy and backdoor attack success rate of ViTs respond distinctively to patch transformations before the positional encoding. Then, based on this finding, we propose an effective method for ViTs to defend both patch-based and blending-based trigger backdoor attacks via patch processing. The performances are evaluated on several benchmark datasets, including CIFAR10, GTSRB, and TinyImageNet, which show the proposedds defense is very successful in mitigating backdoor attacks for ViTs. To the best of our knowledge, this paper presents the first defensive strategy that utilizes a unique characteristic of ViTs against backdoor attacks.

Deepfake video detection has drawn significant attention from researchers due to the security issues induced by deepfake videos. Unfortunately, most of the existing deepfake detection approaches have not competently modeled the natural structures and movements of human faces. In this paper, we formulate the deepfake video detection problem into a graph classification task, and propose a novel paradigm named Facial Action Dependencies Estimation (FADE) for deepfake video detection. We propose a Multi-Dependency Graph Module (MDGM) to capture abundant dependencies among facial action units, and extracts subtle clues in these dependencies. MDGM can be easily integrated into the existing frame-level detection schemes to provide significant performance gains. Extensive experiments demonstrate the superiority of our method against the state-of-the-art methods.

Deep neural networks (DNNs) have been widely adopted in many decision-making industrial applications. Their fairness issues, i.e., whether there exist unintended biases in the DNN, receive much attention and become critical concerns, which can directly cause negative impacts in our daily life and potentially undermine the fairness of our society, especially with their increasing deployment at an unprecedented speed. Recently, some early attempts have been made to provide fairness assurance of DNNs, such as fairness testing, which aims at finding discriminatory samples empirically, and fairness certification, which develops sound but not complete analysis to certify the fairness of DNNs. Nevertheless, how to formally compute discriminatory samples and fairness scores (i.e., the percentage of fair input space), is still largely uninvestigated. In this paper, we propose DeepGemini, a novel fairness formal analysis technique for DNNs, which contains two key components: discriminatory sample discovery and fairness score computation. To uncover discriminatory samples, we encode the fairness of DNNs as safety properties and search for discriminatory samples by means of state-of-the-art verification techniques for DNNs. This reduction enables us to be the first to formally compute discriminatory samples. To compute the fairness score, we develop counterexample guided fairness analysis, which utilizes four heuristics to efficiently approximate a lower bound of fairness score. Extensive experimental evaluations demonstrate the effectiveness and efficiency of DeepGemini on commonly-used benchmarks, and DeepGemini outperforms state-of-the-art DNN fairness certification approaches in terms of both efficiency and scalability.

Forced alignment refers to a technology that time-aligns a given transcription with a corresponding speech. However, as the forced alignment technologies have developed using speech audio, they might fail in alignment when the input speech audio is noise-corrupted or is not accessible. We focus on that there is another component that the speech can be inferred from, the speech video (i.e., talking face video). Since the drawbacks of audio-based forced alignment can be complemented using the visual information when the audio signal is under poor condition, we try to develop a novel video-based forced alignment method. However, different from audio forced alignment, it is challenging to develop a reliable visual forced alignment technology for the following two reasons: 1) Visual Speech Recognition (VSR) has a much lower performance compared to audio-based Automatic Speech Recognition (ASR), and 2) the translation from text to video is not reliable, so the method typically used for building audio forced alignment cannot be utilized in developing visual forced alignment. In order to alleviate these challenges, in this paper, we propose a new method that is appropriate for visual forced alignment, namely Deep Visual Forced Alignment (DVFA). The proposed DVFA can align the input transcription (i.e., sentence) with the talking face video without accessing the speech audio. Moreover, by augmenting the alignment task with anomaly case detection, DVFA can detect mismatches between the input transcription and the input video while performing the alignment. Therefore, we can robustly align the text with the talking face video even if there exist error words in the text. Through extensive experiments, we show the effectiveness of the proposed DVFA not only in the alignment task but also in interpreting the outputs of VSR models.

Deep artificial neural networks (ANNs) play a major role in modeling the visual pathways of primate and rodent. However, they highly simplify the computational properties of neurons compared to their biological counterparts. Instead, Spiking Neural Networks (SNNs) are more biologically plausible models since spiking neurons encode information with time sequences of spikes, just like biological neurons do. However, there is a lack of studies on visual pathways with deep SNNs models. In this study, we model the visual cortex with deep SNNs for the first time, and also with a wide range of state-of-the-art deep CNNs and ViTs for comparison. Using three similarity metrics, we conduct neural representation similarity experiments on three neural datasets collected from two species under three types of stimuli. Based on extensive similarity analyses, we further investigate the functional hierarchy and mechanisms across species. Almost all similarity scores of SNNs are higher than their counterparts of CNNs with an average of 6.6%. Depths of the layers with the highest similarity scores exhibit little differences across mouse cortical regions, but vary significantly across macaque regions, suggesting that the visual processing structure of mice is more regionally homogeneous than that of macaques. Besides, the multi-branch structures observed in some top mouse brain-like neural networks provide computational evidence of parallel processing streams in mice, and the different performance in fitting macaque neural representations under different stimuli exhibits the functional specialization of information processing in macaques. Taken together, our study demonstrates that SNNs could serve as promising candidates to better model and explain the functional hierarchy and mechanisms of the visual system.

Despite the quality improvement brought by the recent methods, video super-resolution (SR) is still very challenging, especially for videos that are low-light and noisy. The current best solution is to subsequently employ best models of video SR, denoising, and illumination enhancement, but doing so often lowers the image quality, due to the inconsistency between the models. This paper presents a new parametric representation called the Deep Parametric 3D Filters (DP3DF), which incorporates local spatiotemporal information to enable simultaneous denoising, illumination enhancement, and SR efficiently in a single encoder-and-decoder network. Also, a dynamic residual frame is jointly learned with the DP3DF via a shared backbone to further boost the SR quality. We performed extensive experiments, including a large-scale user study, to show our method's effectiveness. Our method consistently surpasses the best state-of-the-art methods on all the challenging real datasets with top PSNR and user ratings, yet having a very fast run time. The code is available at https://github.com/xiaogang00/DP3DF.

Adversarial attack on point clouds plays a vital role in evaluating and improving the adversarial robustness of 3D deep learning models. Current attack methods are mainly applied by point perturbation in a non-manifold manner. In this paper, we formulate a novel manifold attack, which deforms the underlying 2-manifold surfaces via parameter plane stretching to generate adversarial point clouds. First, we represent the mapping between the parameter plane and underlying surface using generative-based networks. Second, the stretching is learned in the 2D parameter domain such that the generated 3D point cloud fools a pretrained classifier with minimal geometric distortion. Extensive experiments show that adversarial point clouds generated by manifold attack are smooth, undefendable and transferable, and outperform those samples generated by the state-of-the-art non-manifold ones.

Data-driven predictive solutions predominant in commercial applications tend to suffer from biases and stereotypes, which raises equity concerns. Prediction models may discover, use, or amplify spurious correlations based on gender or other protected personal characteristics, thus discriminating against marginalized groups. Mitigating gender bias has become an important research focus in natural language processing (NLP) and is an area where annotated corpora are available. Data augmentation reduces gender bias by adding counterfactual examples to the training dataset. In this work, we show that some of the examples in the augmented dataset can be not important or even harmful to fairness. We hence propose a general method for pruning both the factual and counterfactual examples to maximize the model’s fairness as measured by the demographic parity, equality of opportunity, and equality of odds. The fairness achieved by our method surpasses that of data augmentation on three text classification datasets, using no more than half of the examples in the augmented dataset. Our experiments are conducted using models of varying sizes and pre-training settings. WARNING: This work uses language that is offensive in nature.

The widespread adoption of electronic health records (EHRs) has opened up new opportunities for using deep neural networks to enhance healthcare. However, modeling EHR data can be challenging due to its complex properties, such as missing values, data scarcity in multi-hospital systems, and multimodal irregularity. How to tackle various issues in EHRs for improving medical prediction is challenging and under exploration. I separately illustrate my works to address these issues in EHRs and discuss potential future directions.

Stochastic partial differential equations (SPDEs) are crucial for modelling dynamics with randomness in many areas including economics, physics, and atmospheric sciences. Recently, using deep learning approaches to learn the PDE solution for accelerating PDE simulation becomes increasingly popular. However, SPDEs have two unique properties that require new design on the models. First, the model to approximate the solution of SPDE should be generalizable over both initial conditions and the random sampled forcing term. Second, the random forcing terms usually have poor regularity whose statistics may diverge (e.g., the space-time white noise). To deal with the problems, in this work, we design a deep neural network called \emph{Deep Latent Regularity Net} (DLR-Net). DLR-Net includes a regularity feature block as the main component, which maps the initial condition and the random forcing term to a set of regularity features. The processing of regularity features is inspired by regularity structure theory and the features provably compose a set of basis to represent the SPDE solution. The regularity features are then fed into a small backbone neural operator to get the output. We conduct experiments on various SPDEs including the dynamic $\Phi^4_1$ model and the stochastic 2D Navier-Stokes equation to predict their solutions, and the results demonstrate that the proposed DLR-Net can achieve SOTA accuracy compared with the baselines. Moreover, the inference time is over 20 times faster than the traditional numerical solver and is comparable with the baseline deep learning models.

In the scene of self-supervised graph learning, Mutual Information (MI) was recently introduced for graph encoding to generate robust node embeddings. A successful representative is Deep Graph Infomax (DGI), which essentially operates on the space of node features but ignores topological structures, and just considers global graph summary. In this paper, we present an effective model called Deep Graph Structural Infomax (DGSI) to learn node representation. We explore to derive the structural mutual information from the perspective of Information Bottleneck (IB), which defines a trade-off between the sufficiency and minimality of representation on the condition of the topological structure preservation. Intuitively, the derived constraints formally maximize the structural mutual information both edge-wise and local neighborhood-wise. Besides, we develop a general framework that incorporates the global representational mutual information, local representational mutual information, and sufficient structural information into the node representation. Essentially, our DGSI extends DGI and could capture more fine-grained semantic information as well as beneficial structural information in a self-supervised manner, thereby improving node representation and further boosting the learning performance. Extensive experiments on different types of datasets demonstrate the effectiveness and superiority of the proposed method.

The ability of snapshot compressive imaging (SCI) systems to efficiently capture high-dimensional (HD) data has led to an inverse problem, which consists of recovering the HD signal from the compressed and noisy measurement. While reconstruction algorithms grow fast to solve it with the recent advances of deep learning, the fundamental issue of accurate and stable recovery remains. To this end, we propose deep equilibrium models (DEQ) for video SCI, fusing data-driven regularization and stable convergence in a theoretically sound manner. Each equilibrium model implicitly learns a nonexpansive operator and analytically computes the fixed point, thus enabling unlimited iterative steps and infinite network depth with only a constant memory requirement in training and testing. Specifically, we demonstrate how DEQ can be applied to two existing models for video SCI reconstruction: recurrent neural networks (RNN) and Plug-and-Play (PnP) algorithms. On a variety of datasets and real data, both quantitative and qualitative evaluations of our results demonstrate the effectiveness and stability of our proposed method. The code and models are available at: https://github.com/IndigoPurple/DEQSCI.

Self-supervised monocular depth estimation has been widely studied recently. Most of the work has focused on improving performance on benchmark datasets, such as KITTI, but has offered a few experiments on generalization performance. In this paper, we investigate the backbone networks (e.g., CNNs, Transformers, and CNN-Transformer hybrid models) toward the generalization of monocular depth estimation. We first evaluate state-of-the-art models on diverse public datasets, which have never been seen during the network training. Next, we investigate the effects of texture-biased and shape-biased representations using the various texture-shifted datasets that we generated. We observe that Transformers exhibit a strong shape bias and CNNs do a strong texture-bias. We also find that shape-biased models show better generalization performance for monocular depth estimation compared to texture-biased models. Based on these observations, we newly design a CNN-Transformer hybrid network with a multi-level adaptive feature fusion module, called MonoFormer. The design intuition behind MonoFormer is to increase shape bias by employing Transformers while compensating for the weak locality bias of Transformers by adaptively fusing multi-level representations. Extensive experiments show that the proposed method achieves state-of-the-art performance with various public datasets. Our method also shows the best generalization ability among the competitive methods.

Knowledge Tracing (KT) is a crucial task in the field of online education, since it aims to predict students' performance on exercises based on their learning history. One typical solution for knowledge tracing is to combine the classic models in educational psychology, such as Item Response Theory (IRT) and Cognitive Diagnosis (CD), with Deep Neural Networks (DNN) technologies. In this solution, a student and related exercises are mapped into feature vectors based on the student's performance at the current time step, however, it does not consider the impact of historical behavior sequences, and the relationships between historical sequences and students. In this paper, we develop DAKTN, a novel model which assimilates the historical sequences to tackle this challenge for better knowledge tracing. To be specific, we apply a pooling layer to incorporate the student behavior sequence in the embedding layer. After that, we further design a local activation unit, which can adaptively calculate the representation vectors by taking the relevance of historical sequences into consideration with respect to candidate student and exercises. Through experimental results on three real-world datasets, DAKTN significantly outperforms state-of-the-art baseline models. We also present the reasonableness of DAKTN by ablation testing.

Semi-supervised anomaly detection is a data mining task which aims at learning features from partially-labeled datasets. We propose Deep Anomaly Detection and Search (DADS) with reinforcement learning. During the training process, the agent searches for possible anomalies in unlabeled dataset to enhance performance. Empirically, we compare DADS with several methods in the settings of leveraging known anomalies to detect both other known and unknown anomalies. Results show that DADS achieves good performance.

Continual Test-Time Adaptation (CTTA) aims to adapt the source model to continually changing unlabeled target domains without access to the source data. Existing methods mainly focus on model-based adaptation in a self-training manner, such as predicting pseudo labels for new domain datasets. Since pseudo labels are noisy and unreliable, these methods suffer from catastrophic forgetting and error accumulation when dealing with dynamic data distributions. Motivated by the prompt learning in NLP, in this paper, we propose to learn an image-layer visual domain prompt for target domains while having the source model parameters frozen. During testing, the changing target datasets can be adapted to the source model by reformulating the input data with the learned visual prompts. Specifically, we devise two types of prompts, i.e., domains-specific prompts and domains-agnostic prompts, to extract current domain knowledge and maintain the domain-shared knowledge in the continual adaptation. Furthermore, we design a homeostasis-based adaptation strategy to suppress domain-sensitive parameters in domain-invariant prompts to learn domain-shared knowledge more effectively. This transition from the model-dependent paradigm to the model-free one enables us to bypass the catastrophic forgetting and error accumulation problems. Experiments show that our proposed method achieves significant performance gains over state-of-the-art methods on four widely-used benchmarks, including CIFAR-10C, CIFAR-100C, ImageNet-C, and VLCS datasets.

Recent research on Generalized Zero-Shot Learning (GZSL) has focused primarily on generation-based methods. However, current literature has overlooked the fundamental principles of these methods and has made limited progress in a complex manner. In this paper, we aim to deconstruct the generator-classifier framework and provide guidance for its improvement and extension. We begin by breaking down the generator-learned unseen class distribution into class-level and instance-level distributions. Through our analysis of the role of these two types of distributions in solving the GZSL problem, we generalize the focus of the generation-based approach, emphasizing the importance of (i) attribute generalization in generator learning and (ii) independent classifier learning with partially biased data. We present a simple method based on this analysis that outperforms SotAs on four public GZSL datasets, demonstrating the validity of our deconstruction. Furthermore, our proposed method remains effective even without a generative model, representing a step towards simplifying the generator-classifier structure. Our code is available at https://github.com/cdb342/DGZ.

Click-through rate (CTR) prediction is crucial in recommendation and online advertising systems. Existing methods usually model user behaviors, while ignoring the informative context which influences the user to make a click decision, e.g., click pages and pre-ranking candidates that inform inferences about user interests, leading to suboptimal performance. In this paper, we propose a Decision-Making Context Interaction Network (DCIN), which deploys a carefully designed Context Interaction Unit (CIU) to learn decision-making contexts and thus benefits CTR prediction. In addition, the relationship between different decision-making context sources is explored by the proposed Adaptive Interest Aggregation Unit (AIAU) to improve CTR prediction further. In the experiments on public and industrial datasets, DCIN significantly outperforms the state-of-the-art methods. Notably, the model has obtained the improvement of CTR+2.9%/CPM+2.1%/GMV+1.5% for online A/B testing and served the main traffic of Meituan Waimai advertising system.