In industry, feature selection is a standard but necessary step to search for an optimal set of informative feature fields for efficient and effective training of deep Click-Through Rate (CTR) models. Most previous works measure the importance of feature fields by using their corresponding continuous weights from the model, then remove the feature fields with small weight values. However, removing many features that correspond to small but not exact zero weights will inevitably hurt model performance and not be friendly to hot-start model training. There is also no theoretical guarantee that the magnitude of weights can represent the importance, thus possibly leading to sub-optimal results if using these methods. To tackle this problem, we propose a novel Learnable Polarizing Feature Selection (LPFS) method using a smoothed-$\ell^0$ function in literature. Furthermore, we extend LPFS to LPFS++ by our newly designed smoothed-$\ell^0$-liked function to select a more informative subset of features. LPFS and LPFS++ can be used as gates inserted at the input of the deep network to control the active and inactive state of each feature. When training is finished, some gates are exact zero, while others are around one, which is particularly favored by the practical hot-start training in the industry, due to no damage to the model performance before and after removing the features corresponding to exact-zero gates. Experiments show that our methods outperform others by a clear margin, and have achieved great A/B test results in KuaiShou Technology.
Recently, self-attention mechanisms have shown impressive performance in various NLP and CV tasks, which can help capture sequential characteristics and derive global information. In this work, we explore how to extend self-attention modules to better learn subtle feature embeddings for recognizing fine-grained objects, e.g., different bird species or person identities. To this end, we propose a dual cross-attention learning (DCAL) algorithm to coordinate with self-attention learning. First, we propose global-local cross-attention (GLCA) to enhance the interactions between global images and local high-response regions, which can help reinforce the spatial-wise discriminative clues for recognition. Second, we propose pair-wise cross-attention (PWCA) to establish the interactions between image pairs. PWCA can regularize the attention learning of an image by treating another image as distractor and will be removed during inference. We observe that DCAL can reduce misleading attentions and diffuse the attention response to discover more complementary parts for recognition. We conduct extensive evaluations on fine-grained visual categorization and object re-identification. Experiments demonstrate that DCAL performs on par with state-of-the-art methods and consistently improves multiple self-attention baselines, e.g., surpassing DeiT-Tiny and ViT-Base by 2.8% and 2.4% mAP on MSMT17, respectively.
Sparse R-CNN is a recent strong object detection baseline by set prediction on sparse, learnable proposal boxes and proposal features. In this work, we propose to improve Sparse R-CNN with two dynamic designs. First, Sparse R-CNN adopts a one-to-one label assignment scheme, where the Hungarian algorithm is applied to match only one positive sample for each ground truth. Such one-to-one assignment may not be optimal for the matching between the learned proposal boxes and ground truths. To address this problem, we propose dynamic label assignment (DLA) based on the optimal transport algorithm to assign increasing positive samples in the iterative training stages of Sparse R-CNN. We constrain the matching to be gradually looser in the sequential stages as the later stage produces the refined proposals with improved precision. Second, the learned proposal boxes and features remain fixed for different images in the inference process of Sparse R-CNN. Motivated by dynamic convolution, we propose dynamic proposal generation (DPG) to assemble multiple proposal experts dynamically for providing better initial proposal boxes and features for the consecutive training stages. DPG thereby can derive sample-dependent proposal boxes and features for inference. Experiments demonstrate that our method, named Dynamic Sparse R-CNN, can boost the strong Sparse R-CNN baseline with different backbones for object detection. Particularly, Dynamic Sparse R-CNN reaches the state-of-the-art 47.2% AP on the COCO 2017 validation set, surpassing Sparse R-CNN by 2.2% AP with the same ResNet-50 backbone.
Despite achieving remarkable performance, Federated Learning (FL) suffers from two critical challenges, i.e., limited computational resources and low training efficiency. In this paper, we propose a novel FL framework, i.e., FedDUAP, with two original contributions, to exploit the insensitive data on the server and the decentralized data in edge devices to further improve the training efficiency. First, a dynamic server update algorithm is designed to exploit the insensitive data on the server, in order to dynamically determine the optimal steps of the server update for improving the convergence and accuracy of the global model. Second, a layer-adaptive model pruning method is developed to perform unique pruning operations adapted to the different dimensions and importance of multiple layers, to achieve a good balance between efficiency and effectiveness. By integrating the two original techniques together, our proposed FL model, FedDUAP, significantly outperforms baseline approaches in terms of accuracy (up to 4.8% higher), efficiency (up to 2.8 times faster), and computational cost (up to 61.9% smaller).
Vision transformers (ViTs) have gained popularity recently. Even without customized image operators such as convolutions, ViTs can yield competitive performance when properly trained on massive data. However, the computational overhead of ViTs remains prohibitive, due to stacking multi-head self-attention modules and else. Compared to the vast literature and prevailing success in compressing convolutional neural networks, the study of Vision Transformer compression has also just emerged, and existing works focused on one or two aspects of compression. This paper proposes a unified ViT compression framework that seamlessly assembles three effective techniques: pruning, layer skipping, and knowledge distillation. We formulate a budget-constrained, end-to-end optimization framework, targeting jointly learning model weights, layer-wise pruning ratios/masks, and skip configurations, under a distillation loss. The optimization problem is then solved using the primal-dual algorithm. Experiments are conducted with several ViT variants, e.g. DeiT and T2T-ViT backbones on the ImageNet dataset, and our approach consistently outperforms recent competitors. For example, DeiT-Tiny can be trimmed down to 50\% of the original FLOPs almost without losing accuracy. Codes are available online:~\url{https://github.com/VITA-Group/UVC}.
Recently, learning from vast unlabeled data, especially self-supervised learning, has been emerging and attracted widespread attention. Self-supervised learning followed by the supervised fine-tuning on a few labeled examples can significantly improve label efficiency and outperform standard supervised training using fully annotated data. In this work, we present a novel self-supervised deep learning paradigm based on online hard negative pair mining. Specifically, we design a student-teacher network to generate multi-view of the data for self-supervised learning and integrate hard negative pair mining into the training. Then we derive a new triplet-like loss considering both positive sample pairs and mined hard negative sample pairs. Extensive experiments demonstrate the effectiveness of the proposed method and its components on ILSVRC-2012.
Bankruptcy risk prediction for Small and Medium-sized Enterprises (SMEs) is a crucial step for financial institutions to make the loan decision and identify region economics's early warning. However, previous studies in both finance and AI research fields only consider either the intra-risk or the spillover-risk, ignoring their interactions and their combinatorial effect for simplicity. This paper for the first time considers both risks simultaneously and their joint effect in bankruptcy prediction. Specifically, we first propose an enterprise intra-risk encoder with LSTM based on enterprise risk statistical significance indicators from its basic business information and litigation information for its intra-risk learning. Afterward, we propose an enterprise spillover-risk encoder based on enterprise relational information from the enterprise knowledge graph for its spillover-risk embedding. In particular, the spillover-risk encoder is equipped with both the newly proposed Hyper-Graph Neural Networks (Hyper-GNNs) and Heterogeneous Graph Neural Networks (Heter-GNNs), which is able to model spillover risk from two different aspects, i.e. common risk factors based on hyperedges and direct diffusion risk from the neighbors, respectively. With the two kinds of encoders, a unified framework is designed to simultaneously capture intra-risk and spillover-risk for bankruptcy prediction. To evaluate our model, we collect multi-sources SMEs real-world data and build a novel benchmark dataset SMEsD. We provide open access to the dataset, which is expected to promote the financial risk analysis research further. Experiments on SMEsD against nine SOTA baselines demonstrate the effectiveness of the proposed model for bankruptcy prediction.
Bi-type multi-relational heterogeneous graph (BMHG) is one of the most common graphs in practice, for example, academic networks, e-commerce user behavior graph and enterprise knowledge graph. It is a critical and challenge problem on how to learn the numerical representation for each node to characterize subtle structures. However, most previous studies treat all node relations in BMHG as the same class of relation without distinguishing the different characteristics between the intra-class relations and inter-class relations of the bi-typed nodes, causing the loss of significant structure information. To address this issue, we propose a novel Dual Hierarchical Attention Networks (DHAN) based on the bi-typed multi-relational heterogeneous graphs to learn comprehensive node representations with the intra-class and inter-class attention-based encoder under a hierarchical mechanism. Specifically, the former encoder aggregates information from the same type of nodes, while the latter aggregates node representations from its different types of neighbors. Moreover, to sufficiently model node multi-relational information in BMHG, we adopt a newly proposed hierarchical mechanism. By doing so, the proposed dual hierarchical attention operations enable our model to fully capture the complex structures of the bi-typed multi-relational heterogeneous graphs. Experimental results on various tasks against the state-of-the-arts sufficiently confirm the capability of DHAN in learning node representations on the BMHGs.
Stock Movement Prediction (SMP) aims at predicting listed companies' stock future price trend, which is a challenging task due to the volatile nature of financial markets. Recent financial studies show that the momentum spillover effect plays a significant role in stock fluctuation. However, previous studies typically only learn the simple connection information among related companies, which inevitably fail to model complex relations of listed companies in the real financial market. To address this issue, we first construct a more comprehensive Market Knowledge Graph (MKG) which contains bi-typed entities including listed companies and their associated executives, and hybrid-relations including the explicit relations and implicit relations. Afterward, we propose DanSmp, a novel Dual Attention Networks to learn the momentum spillover signals based upon the constructed MKG for stock prediction. The empirical experiments on our constructed datasets against nine SOTA baselines demonstrate that the proposed DanSmp is capable of improving stock prediction with the constructed MKG.
We develop a new measure of the exploration/exploitation trade-off in infinite-horizon reinforcement learning problems called the occupancy information ratio (OIR), which is comprised of a ratio between the infinite-horizon average cost of a policy and the entropy of its long-term state occupancy measure. The OIR ensures that no matter how many trajectories an RL agent traverses or how well it learns to minimize cost, it maintains a healthy skepticism about its environment, in that it defines an optimal policy which induces a high-entropy occupancy measure. Different from earlier information ratio notions, OIR is amenable to direct policy search over parameterized families, and exhibits hidden quasiconcavity through invocation of the perspective transformation. This feature ensures that under appropriate policy parameterizations, the OIR optimization problem has no spurious stationary points, despite the overall problem's nonconvexity. We develop for the first time policy gradient and actor-critic algorithms for OIR optimization based upon a new entropy gradient theorem, and establish both asymptotic and non-asymptotic convergence results with global optimality guarantees. In experiments, these methodologies outperform several deep RL baselines in problems with sparse rewards, where many trajectories may be uninformative and skepticism about the environment is crucial to success.