AAAI 2023 Papers

Spiking camera, a novel retina-inspired vision sensor, has shown its great potential for capturing high-speed dynamic scenes with a sampling rate of 40,000 Hz. The spiking camera abandons the concept of exposure window, with each of its photosensitive units continuously capturing photons and firing spikes asynchronously. However, the special sampling mechanism prevents the frame-based algorithm from being used to spiking camera. It remains to be a challenge to reconstruct dynamic scenes and perform common computer vision tasks for spiking camera. In this paper, we propose a self-supervised joint learning framework for optical flow estimation and reconstruction of spiking camera. The framework reconstructs clean frame-based spiking representations in a self-supervised manner, and then uses them to train the optical flow networks. We also propose an optical flow based inverse rendering process to achieve self-supervision by minimizing the difference with respect to the original spiking temporal aggregation image. The experimental results demonstrate that our method bridges the gap between synthetic and real-world scenes and achieves desired results in real-world scenarios. To the best of our knowledge, this is the first attempt to jointly reconstruct dynamic scenes and estimate optical flow for spiking camera from a self-supervised learning perspective.

Cross-domain recommendation has attracted increasing attention from industry and academia recently. However, most existing methods do not exploit the interest invariance between domains, which would yield sub-optimal solutions. In this paper, we propose a cross-domain recommendation method: Self-supervised Interest Transfer Network (SITN), which can effectively transfer invariant knowledge between domains via prototypical contrastive learning. Specifically, we perform two levels of cross-domain contrastive learning: 1) instance-to-instance contrastive learning, 2) instance-to-cluster contrastive learning. Not only that, we also take into account users' multi-granularity and multi-view interests. With this paradigm, SITN can explicitly learn the invariant knowledge of interest clusters between domains and accurately capture users' intents and preferences. We conducted extensive experiments on a public dataset and a large-scale industrial dataset collected from one of the world's leading e-commerce corporations. The experimental results indicate that SITN achieves significant improvements over state-of-the-art recommendation methods. Additionally, SITN has been deployed on a micro-video recommendation platform, and the online A/B testing results further demonstrate its practical value. Supplement is available at: https://github.com/fanqieCoffee/SITN-Supplement.

For image local forgery detection, the existing methods require a large amount of labeled data for training, and most of them cannot detect multiple types of forgery simultaneously. In this paper, we firstly analyzed the JPEG compression traces which are mainly caused by different JPEG compression chains, and designed a trace extractor to learn such traces. Then, we utilized the trace extractor as the backbone and trained self-supervised to strengthen the discrimination ability of learned traces. With its benefits, regions with different JPEG compression chains can easily be distinguished within a forged image. Furthermore, our method does not rely on a large amount of training data, and even does not require any forged images for training. Experiments show that the proposed method can detect image local forgery on different datasets without re-training, and keep stable performance over various types of image local forgery.

We devise a new regularization for denoising with self-supervised learning. The regularization uses a deep image prior learned by the network, rather than a traditional predefined prior. Specifically, we treat the output of the network as a ``prior'' that we again denoise after ``re-noising.'' The network is updated to minimize the discrepancy between the twice-denoised image and its prior. We demonstrate that this regularization enables the network to learn to denoise even if it has not seen any clean images. The effectiveness of our method is based on the fact that CNNs naturally tend to capture low-level image statistics. Since our method utilizes the image prior implicitly captured by the deep denoising CNN to guide denoising, we refer to this training strategy as an Implicit Deep Denoiser Prior (IDDP). IDDP can be seen as a mixture of learning-based methods and traditional model-based denoising methods, in which regularization is adaptively formulated using the output of the network. We apply IDDP to various denoising tasks using only observed corrupted data and show that it achieves better denoising results than other self-supervised denoising methods.

Cognitive diagnosis is a fundamental yet critical research task in the field of intelligent education, which aims to discover the proficiency level of different students on specific knowledge concepts. Despite the effectiveness of existing efforts, previous methods always considered the mastery level on the whole students, so they still suffer from the Long Tail Effect. A large number of students who have sparse interaction records are usually wrongly diagnosed during inference. To relieve the situation, we proposed a Self-supervised Cognitive Diagnosis (SCD) framework which leverages the self-supervised manner to assist the graph-based cognitive diagnosis, then the performance on those students with sparse data can be improved. Specifically, we came up with a graph confusion method that drops edges under some special rules to generate different sparse views of the graph. By maximizing the cross-view consistency of node representations, our model could pay more attention on long-tailed students. Additionally, we proposed an importance-based view generation rule to improve the influence of long-tailed students. Extensive experiments on real-world datasets show the effectiveness of our approach, especially on the students with much sparser interaction records. Our code is available at https://github.com/zeng-zhen/SCD.

As one of the most important research topics in the unsupervised learning field, Multi-View Clustering (MVC) has been widely studied in the past decade and numerous MVC methods have been developed. Among these methods, the recently emerged Graph Neural Networks (GNN) shine a light on modeling both topological structure and node attributes in the form of graphs, to guide unified embedding learning and clustering. However, the effectiveness of existing GNN-based MVC methods is still limited due to the insufficient consideration in utilizing the self-supervised information and graph information, which can be reflected from the following two aspects: 1) most of these models merely use the self-supervised information to guide the feature learning and fail to realize that such information can be also applied in graph learning and sample weighting; 2) the usage of graph information is generally limited to the feature aggregation in these models, yet it also provides valuable evidence in detecting noisy samples. To this end, in this paper we propose Self-Supervised Graph Attention Networks for Deep Weighted Multi-View Clustering (SGDMC), which promotes the performance of GNN-based deep MVC models by making full use of the self-supervised information and graph information. Specifically, a novel attention-allocating approach that considers both the similarity of node attributes and the self-supervised information is developed to comprehensively evaluate the relevance among different nodes. Meanwhile, to alleviate the negative impact caused by noisy samples and the discrepancy of cluster structures, we further design a sample-weighting strategy based on the attention graph as well as the discrepancy between the global pseudo-labels and the local cluster assignment. Experimental results on multiple real-world datasets demonstrate the effectiveness of our method over existing approaches.

Continual graph learning routinely finds its role in a variety of real-world applications where the graph data with different tasks come sequentially. Despite the success of prior works, it still faces great challenges. On the one hand, existing methods work with the zero-curvature Euclidean space, and largely ignore the fact that curvature varies over the com- ing graph sequence. On the other hand, continual learners in the literature rely on abundant labels, but labeling graph in practice is particularly hard especially for the continuously emerging graphs on-the-fly. To address the aforementioned challenges, we propose to explore a challenging yet practical problem, the self-supervised continual graph learning in adaptive Riemannian spaces. In this paper, we propose a novel self-supervised Riemannian Graph Continual Learner (RieGrace). In RieGrace, we first design an Adaptive Riemannian GCN (AdaRGCN), a unified GCN coupled with a neural curvature adapter, so that Riemannian space is shaped by the learnt curvature adaptive to each graph. Then, we present a Label-free Lorentz Distillation approach, in which we create teacher-student AdaRGCN for the graph sequence. The student successively performs intra-distillation from itself and inter-distillation from the teacher so as to consolidate knowledge without catastrophic forgetting. In particular, we propose a theoretically grounded Generalized Lorentz Projection for the contrastive distillation in Riemannian space. Extensive experiments on the benchmark datasets show the superiority of RieGrace, and additionally, we investigate on how curvature changes over the graph sequence.

Deep learning methods have demonstrated promising performance on the NP-hard Graph Matching (GM) problems. However, the state-of-the-art methods usually require the ground-truth labels, which may take extensive human efforts or be impractical to collect. In this paper, we present a robust self-supervised bidirectional learning method (IA-SSGM) to tackle GM in an unsupervised manner. It involves an affinity learning component and a classic GM solver. Specifically, we adopt the Hungarian solver to generate pseudo correspondence labels for the simple probabilistic relaxation of the affinity matrix. In addition, a bidirectional recycling consistency module is proposed to generate pseudo samples by recycling the pseudo correspondence back to permute the input. It imposes a consistency constraint between the pseudo affinity and the original one, which is theoretically supported to help reduce the matching error. Our method further develops a graph contrastive learning jointly with the affinity learning to enhance its robustness against the noise and outliers in real applications. Experiments deliver superior performance over the previous state-of-the-arts on five real-world benchmarks, especially under the more difficult outlier scenarios, demon- strating the effectiveness of our method.

We present CrissCross, a self-supervised framework for learning audio-visual representations. A novel notion is introduced in our framework whereby in addition to learning the intra-modal and standard 'synchronous' cross-modal relations, CrissCross also learns 'asynchronous' cross-modal relationships. We perform in-depth studies showing that by relaxing the temporal synchronicity between the audio and visual modalities, the network learns strong generalized representations useful for a variety of downstream tasks. To pretrain our proposed solution, we use 3 different datasets with varying sizes, Kinetics-Sound, Kinetics400, and AudioSet. The learned representations are evaluated on a number of downstream tasks namely action recognition, sound classification, and action retrieval. Our experiments show that CrissCross either outperforms or achieves performances on par with the current state-of-the-art self-supervised methods on action recognition and action retrieval with UCF101 and HMDB51, as well as sound classification with ESC50 and DCASE. Moreover, CrissCross outperforms fully-supervised pretraining while pretrained on Kinetics-Sound.

Self-supervised learning has demonstrated remarkable capability in representation learning for skeleton-based action recognition. Existing methods mainly focus on applying global data augmentation to generate different views of the skeleton sequence for contrastive learning. However, due to the rich action clues in the skeleton sequences, existing methods may only take a global perspective to learn to discriminate different skeletons without thoroughly leveraging the local relationship between different skeleton joints and video frames, which is essential for real-world applications. In this work, we propose a Partial Spatio-Temporal Learning (PSTL) framework to exploit the local relationship from a partial skeleton sequences built by a unique spatio-temporal masking strategy. Specifically, we construct a negative-sample-free triplet steam structure that is composed of an anchor stream without any masking, a spatial masking stream with Central Spatial Masking (CSM), and a temporal masking stream with Motion Attention Temporal Masking (MATM). The feature cross-correlation matrix is measured between the anchor stream and the other two masking streams, respectively. (1) Central Spatial Masking discards selected joints from the feature calculation process, where the joints with a higher degree of centrality have a higher possibility of being selected. (2) Motion Attention Temporal Masking leverages the motion of action and remove frames that move faster with a higher possibility. Our method achieves state-of-the-art performance on NTURGB+D 60, NTURGB+D 120 and PKU-MMD under various downstream tasks. Furthermore, to simulate the real-world scenarios, a practical evaluation is performed where some skeleton joints are lost in downstream tasks.In contrast to previous methods that suffer from large performance drops, our PSTL can still achieve remarkable results under this challenging setting, validating the robustness of our method.

Graph convolutional neural network (GCN) based methods have achieved noticeable performance in solving mixed integer programming problems (MIPs). However, the generalization of existing work is limited due to the problem structure. This paper proposes a self-paced learning (SPL) based GCN network (SPGCN) with curriculum learning (CL) to make the utmost of samples. SPGCN employs a GCN model to imitate the branching variable selection during the branch and bound process, while the training process is conducted in a self-paced fashion. Specifically, SPGCN contains a loss-based automatic difficulty measurer, where the training loss of the sample represents the difficulty level. In each iteration, a dynamic training dataset is constructed according to the difficulty level for GCN model training. Experiments on four NP-hard datasets verify that CL can lead to generalization improvement and convergence speedup in solving MIPs, where SPL performs better than predefined CL methods.

Most Graph Neural Networks follow the message-passing paradigm, assuming the observed structure depicts the ground-truth node relationships. However, this fundamental assumption cannot always be satisfied, as real-world graphs are always incomplete, noisy, or redundant. How to reveal the inherent graph structure in a unified way remains under-explored. We proposed PRI-GSL, a Graph Structure Learning framework guided by the Principle of Relevant Information, providing a simple and unified framework for identifying the self-organization and revealing the hidden structure. PRI-GSL learns a structure that contains the most relevant yet least redundant information quantified by von Neumann entropy and Quantum Jensen Shannon divergence. PRI-GSL incorporates the evolution of quantum continuous walk with graph wavelets to encode node structural roles, showing in which way the nodes interplay and self-organize with the graph structure. Extensive experiments demonstrate the superior effectiveness and robustness of PRI-GSL.

Hand and face play an important role in expressing sign language. Their features are usually especially leveraged to improve system performance. However, to effectively extract visual representations and capture trajectories for hands and face, previous methods always come at high computations with increased training complexity. They usually employ extra heavy pose-estimation networks to locate human body keypoints or rely on additional pre-extracted heatmaps for supervision. To relieve this problem, we propose a self-emphasizing network (SEN) to emphasize informative spatial regions in a self-motivated way, with few extra computations and without additional expensive supervision. Specifically, SEN first employs a lightweight subnetwork to incorporate local spatial-temporal features to identify informative regions, and then dynamically augment original features via attention maps. It's also observed that not all frames contribute equally to recognition. We present a temporal self-emphasizing module to adaptively emphasize those discriminative frames and suppress redundant ones. A comprehensive comparison with previous methods equipped with hand and face features demonstrates the superiority of our method, even though they always require huge computations and rely on expensive extra supervision. Remarkably, with few extra computations, SEN achieves new state-of-the-art accuracy on four large-scale datasets, PHOENIX14, PHOENIX14-T, CSL-Daily, and CSL. Visualizations verify the effects of SEN on emphasizing informative spatial and temporal features. Code is available at https://github.com/hulianyuyy/SEN_CSLR

Knowledge distillation (KD) is a promising teacher-student learning paradigm that transfers information from a cumbersome teacher to a student network. To avoid the training cost of a large teacher network, the recent studies propose to distill knowledge from the student itself, called Self-KD. However, due to the limitations of the performance and capacity of the student, the soft-labels or features distilled by the student barely provide reliable guidance. Moreover, most of the Self-KD algorithms are specific to classification tasks based on soft-labels, and not suitable for semantic segmentation. To alleviate these contradictions, we revisit the label and feature distillation problem in segmentation, and propose Self-Decoupling and Ensemble Distillation for Efficient Segmentation (SDES). Specifically, we design a decoupled prediction ensemble distillation (DPED) algorithm that generates reliable soft-labels with multiple expert decoders, and a decoupled feature ensemble distillation (DFED) mechanism to utilize more important channel-wise feature maps for encoder learning. The extensive experiments on three public segmentation datasets demonstrate the superiority of our approach and the efficacy of each component in the framework through the ablation study.

Contrastive loss has significantly improved performance in supervised classification tasks by using a multi-viewed framework that leverages augmentation and label information. The augmentation enables contrast with another view of a single image but enlarges training time and memory usage. To exploit the strength of multi-views while avoiding the high computation cost, we introduce a multi-exit architecture that outputs multiple features of a single image in a single-viewed framework. To this end, we propose Self-Contrastive (SelfCon) learning, which self-contrasts within multiple outputs from the different levels of a single network. The multi-exit architecture efficiently replaces multi-augmented images and leverages various information from different layers of a network. We demonstrate that SelfCon learning improves the classification performance of the encoder network, and empirically analyze its advantages in terms of the single-view and the sub-network. Furthermore, we provide theoretical evidence of the performance increase based on the mutual information bound. For ImageNet classification on ResNet-50, SelfCon improves accuracy by +0.6% with 59% memory and 48% time of Supervised Contrastive learning, and a simple ensemble of multi-exit outputs boosts performance up to +1.5%. Our code is available at https://github.com/raymin0223/self-contrastive-learning.

High-resolution (HR) images are usually downscaled to low-resolution (LR) ones for better display and afterward upscaled back to the original size to recover details. Recent work in image rescaling formulates downscaling and upscaling as a unified task and learns a bijective mapping between HR and LR via invertible networks. However, in real-world applications (e.g., social media), most images are compressed for transmission. Lossy compression will lead to irreversible information loss on LR images, hence damaging the inverse upscaling procedure and degrading the reconstruction accuracy. In this paper, we propose the Self-Asymmetric Invertible Network (SAIN) for compression-aware image rescaling. To tackle the distribution shift, we first develop an end-to-end asymmetric framework with two separate bijective mappings for high-quality and compressed LR images, respectively. Then, based on empirical analysis of this framework, we model the distribution of the lost information (including downscaling and compression) using isotropic Gaussian mixtures and propose the Enhanced Invertible Block to derive high-quality/compressed LR images in one forward pass. Besides, we design a set of losses to regularize the learned LR images and enhance the invertibility. Extensive experiments demonstrate the consistent improvements of SAIN across various image rescaling datasets in terms of both quantitative and qualitative evaluation under standard image compression formats (i.e., JPEG and WebP). Code is available at https://github.com/yang-jin-hai/SAIN.

Efficiently training accurate deep models for weakly supervised semantic segmentation (WSSS) with image-level labels is challenging and important. Recently, end-to-end WSSS methods have become the focus of research due to their high training efficiency. However, current methods suffer from insufficient extraction of comprehensive semantic information, resulting in low-quality pseudo-labels and sub-optimal solutions for end-to-end WSSS. To this end, we propose a simple and novel Self Correspondence Distillation (SCD) method to refine pseudo-labels without introducing external supervision. Our SCD enables the network to utilize feature correspondence derived from itself as a distillation target, which can enhance the network's feature learning process by complementing semantic information. In addition, to further improve the segmentation accuracy, we design a Variation-aware Refine Module to enhance the local consistency of pseudo-labels by computing pixel-level variation. Finally, we present an efficient end-to-end Transformer-based framework (TSCD) via SCD and Variation-aware Refine Module for the accurate WSSS task. Extensive experiments on the PASCAL VOC 2012 and MS COCO 2014 datasets demonstrate that our method significantly outperforms other state-of-the-art methods. Our code is available at https://github.com/Rongtao-Xu/RepresentationLearning/tree/main/SCD-AAAI2023.

Attention mechanisms, such as local and non-local attention, play a fundamental role in recent deep learning based speech enhancement (SE) systems. However, a natural speech contains many fast-changing and relatively briefly acoustic events, therefore, capturing the most informative speech features by indiscriminately using local and non-local attention is challenged. We observe that the noise type and speech feature vary within a sequence of speech and the local and non-local can respectively process different types of corrupted speech regions. To leverage this, we propose Selector-Enhancer, a dual-attention based convolution neural network (CNN) with a feature-filter that can dynamically select regions from low-resolution speech features and feed them to local or non-local attention operations. In particular, the proposed feature-filter is trained by using reinforcement learning (RL) with a developed difficulty-regulated reward that related to network performance, model complexity and “the difficulty of the SE task”. The results show that our method achieves comparable or superior performance to existing approaches. In particular, Selector-Enhancer is effective for real-world denoising, where the number and types of noise are varies on a single noisy mixture.

Benefiting from the sequence-level knowledge distillation, the Non-Autoregressive Transformer (NAT) achieves great success in neural machine translation tasks. However, existing knowledge distillation has side effects, such as propagating errors from the teacher to NAT students, which may limit further improvements of NAT models and are rarely discussed in existing research. In this paper, we introduce selective knowledge distillation by introducing an NAT evaluator to select NAT-friendly targets that are of high quality and easy to learn. In addition, we introduce a simple yet effective progressive distillation method to boost NAT performance. Experiment results on multiple WMT language directions and several representative NAT models show that our approach can realize a flexible trade-off between the quality and complexity of training data for NAT models, achieving strong performances. Further analysis shows that distilling only 5% of the raw translations can help an NAT outperform its counterpart trained on raw data by about 2.4 BLEU.