Class-Incremental Learning (CIL) aims to train a reliable model with the streaming data, which emerges unknown classes sequentially. Different from traditional closed set learning, CIL has two main challenges: (1) Novel class detection. The initial training data only contains incomplete classes, and streaming test data will accept unknown classes. Therefore, the model needs to not only accurately classify known classes, but also effectively detect unknown classes; (2) Model expansion. After the novel classes are detected, the model needs to be updated without re-training using the entire previous data. However, traditional CIL methods have not fully considered these two challenges. To this end, we propose a Class-Incremental Learning without Forgetting (CILF) framework. In detail, CILF designs to regularize classification with decoupled prototype based loss, which can improve the intra-class and inter-class structure significantly, and acquire a compact embedding representation for novel class detection in result. Then, CILF employs a learnable curriculum clustering operator to estimate the number of semantic clusters via fine-tuning the learned network, in which curriculum operator can adaptively learn the embedding in self-taught form. Last, with the labeled streaming test data, CILF can update the network with robust regularization to mitigate the catastrophic forgetting.

Corporate relative valuation (CRV) refers to the process of comparing a company's value from company products, core staff and other related information, so that we can assess the company's market value, which is critical for venture capital firms. Traditionally, relative valuation methods heavily rely on tedious and expensive human efforts, especially for non-publicly listed companies. However, the availability of information about company's invisible assets, such as patents, talent, and investors, enables a new paradigm for learning and evaluating corporate relative values automatically. Indeed, in this paper, we reveal that, if the companies and their core members are formed as a heterogeneous graph and the attributes of different nodes include semantically-rich multi-modal data, it is able to extract a latent embedding for each company. Along this line, we develop an end-to-end heterogeneous multi-modal graph neural network method, named HM$^2$. Specifically, HM$^2$firstly perform the representation learning for heterogeneous neighbors of input company by taking relationships among nodes into consideration, which aggregates node attributes via linkage-aware multi-head attention mechanism, rather than multi-instance based methods. Then, HM$^2$adopts the self-attention network to aggregate different modal embeddings for final prediction, and employs dynamic triplet loss with embeddings of competitors as the constraint.

Most existing approaches often utilize the pre-fixed structure and large number of labeled data for training complex deep models, which are difficult to implement on incremental scenarios. As a matter of fact, real-world data is always in stream form. Thereby, there exits two challenges for building incremental deep models: a)Capacity Scalability. The entire training data is not available before learning the task. It is a challenge to make the deep model structure scale with streaming data for flexible model evolution and faster convergence. b)Capacity Sustainability. The distribution streaming data usually changes in nature (concept drift), thus it is necessary to update the model while preserving previous knowledge for overcoming the catastrophic forgetting. To this end, we develop an incremental deep model (IDM), which expands the network structure according to streaming data and slows down forgetting with the adaptive fisher regularization. However, IDM ignores another significant challenge with streaming data: c)Capacity Demand. Training a deep model always needs a large amount of labeled data, whereas it is almost impossible to label all unlabeled instances in real time. The core problem is to select a small number of the most discriminative instances to label while keeping the predictive accuracy of the model. Thereby, we focus on the online semi-supervised learning scenario with abrupt changes in data distribution, and further improve IDM to a cost-effective incremental deep model (CE-IDM), which can adaptively select the most discriminative newly coming instances for query to reduce the manual labeling costs. Specifically, CE-IDM adopts a novel extensible deep network structure by using an extra attention model for hidden layers. Based on the adaptive attention weights, CE-IDM develops a novel instance selection criterion by jointly estimating unlabeled instances’ representative and informative degree to satisfy the capacity demand. With the newly labeled instances, CE-IDM can quickly update the model with adaptive depth from streaming data and enable capacity scalability. Also, we address capacity sustainability by exploiting the attention based fisher information matrix, which can slow down the forgetting in consequence. Finally, CE-IDM can deal with the three capacity challenges methioned above in a unified framework. We conduct extensive experiments on real-world data and show that CE-IDM outperforms the state-of-the-art methods with a substantial margin.

Open set classification (OSC) tackles the problem of determining whether the data are in-class or out-of-class during inference, when only provided with a set of in-class examples at training time. Traditional OSC methods usually train discriminative or generative models with the owned in-class data, and then utilize the pre-trained models to classify test data directly. However, these methods always suffer from the embedding confusion problem, i.e., partial out-of-class instances are mixed with in-class ones of similar semantics, making it difficult to classify. To solve this problem, we unify semi-supervised learning to develop a novel OSC algorithm, S2OSC, which incorporates out-of-class instances filtering and model re-training in a transductive manner. In detail, given a pool of newly coming test data, S2OSC firstly filters the mostly distinct out-of-class instances using the pre-trained model, and annotates super-class for them. Then, S2OSC trains a holistic classification model by combing in-class and out-of-class labeled data with the remaining unlabeled test data in a semi-supervised paradigm. Furthermore, considering that data are usually in the streaming form in real applications, we extend S2OSC into an incremental update framework (I-S2OSC), and adopt a knowledge memory regularization to mitigate the catastrophic forgetting problem in incremental update. Despite the simplicity of proposed models, the experimental results show that S2OSC achieves state-of-the-art performance across a variety of OSC tasks, including 85.4% of F1 on CIFAR-10 with only 300 pseudo-labels. We also demonstrate how S2OSC can be expanded to incremental OSC setting effectively with streaming data.

Estimated time of arrival (ETA) is a very important factor in the transportation system. It has attracted increasing attentions and has been widely used as a basic service in navigation systems and intelligent transportation systems. In this paper, we propose a novel solution to the ETA estimation problem, which is an ensemble on tree models and neural networks. We proved the accuracy and robustness of the solution on the A/B list and finally won first place in the SIGSPATIAL 2021 GISCUP competition.

In this work, a new e-commerce search service named text-picture shopping guide (TPSG) is investigated and deployed to one of the most popular shopping platforms called Taobao. Different from traditional services that only contain text options, the TPSG provides pairs of text terms and user-friendly pictures for shopping guide, named text-picture options (TPOs). Instead of manually labeling pictures, we aim to automatically recommend personalized pictures in TPOs. To this end, we build a large-scale graph model on a great amount of data about users, pictures, and terms. Accordingly, a cross-graph convolution learning (CGCL) method is proposed to facilitate the accurate and efficient inference on the constructed graph. To separate the cue of personalized preferences of users to commodities, we factorize the entire mixture-relation graph involving attributes/relations of users and commodities into the user graph, the commodity graph, and the cross user-commodity graph which just characterizes the preferences. Further, we introduce powerful graph convolution to learn more effective representation of these graphs. To reduce the computation burden, specifically, we generalize graph convolution and propose a tensor graph convolution method to learn representation on cross graphs. We conduct extensive offline and online experiments on the large-scale datasets. The results show that the proposed CGCL is very effective and the TPOs recommendation method outperforms manual/advanced selection methods.

Abnormal action detection is the most noteworthy part of anomaly detection, which tries to identify unusual human behaviors in videos. Previous methods typically utilize future frame prediction to detect frames deviating from the normal scenario. While this strategy enjoys success in the accuracy of anomaly detection, critical information such as the cause and location of the abnormality is unable to be acquired. This paper proposes human motion prediction for abnormal action detection. We employ sequence of human poses to represent human motion, and detect irregular behavior by comparing the predicted pose with the actual pose detected in the frame. Hence the proposed method is able to explain why the action is regarded as irregularity and locate where the anomaly happens. Moreover, pose sequence is robust to noise, complex background and small targets in videos. Since posture information is non-Euclidean data, graph convolutional network is adopted for future pose prediction, which not only leads to greater expressive power but also stronger generalization capability.

Convolution on graphs has aroused great interest in AI due to its potential applications to non-gridded data. To bypass the influence of ordering and different node degrees, the summation/average diffusion/aggregation is often imposed on local receptive field in most prior works. However, the collapsing into one node in this way tends to cause signal entanglements of nodes, which would result in a sub-optimal feature and decrease the discriminability of nodes. To address this problem, in this paper, we propose a simple but effective Hashing Graph Convolution (HGC) method by using global-hashing and local-projection on node aggregation for the task of node classification. In contrast to the conventional aggregation with a full collision, the hash-projection can greatly reduce the collision probability during gathering neighbor nodes. Another incidental effect of hash-projection is that the receptive field of each node is normalized into a common-size bucket space, which not only staves off the trouble of different-size neighbors and their order but also makes a graph convolution run like the standard shape-gridded convolution. Considering the few training samples, also, we introduce a prediction-consistent regularization term into HGC to constrain the score consistency of unlabeled nodes in the graph. HGC is evaluated on both transductive and inductive experimental settings and achieves new state-of-the-art results on all datasets for node classification task. The extensive experiments demonstrate the effectiveness of hash-projection.

Curriculum Learning (CL) is a recently proposed learning paradigm that aims to achieve satisfactory performance by properly organizing the learning sequence from simple curriculum examples to more difficult ones. Up to now, few works have been done to explore CL for the data with graph structure. Therefore, this article proposes a novel CL algorithm that can be utilized to guide the Label Propagation (LP) over graphs, of which the target is to “learn” the labels of unlabeled examples on the graphs. Specifically, we assume that different unlabeled examples have different levels of difficulty for propagation, and their label learning should follow a simple-to-difficult sequence with the updated curricula. Furthermore, considering that the practical data are often characterized by multiple modalities, every modality in our method is associated with a “teacher” that not only evaluates the difficulties of examples from its own viewpoint, but also cooperates with other teachers to generate the overall simplest curriculum examples for propagation. By taking the curriculums suggested by the teachers as a whole, the common preference (i.e., commonality) of teachers on selecting the simplest examples can be discovered by a row-sparse matrix, and their distinct opinions (i.e., individuality) are captured by a sparse noise matrix. As a result, an accurate curriculum sequence can be established and the propagation quality can thus be improved. Theoretically, we prove that the propagation risk bound is closely related to the examples’ difficulty information, and empirically, we show that our method can generate higher accuracy than the state-of-the-art CL approach and LP algorithms on various multi-modal tasks.

Practically, we are often in the dilemma that the labeled data at hand are inadequate to train a reliable classifier, and more seriously, some of these labeled data may be mistakenly labeled due to the various human factors. Therefore, this paper proposes a novel semi-supervised learning paradigm that can handle both label insufficiency and label inaccuracy. To address label insufficiency, we use a graph to bridge the data points so that the label information can be propagated from the scarce labeled examples to unlabeled examples along the graph edges. To address label inaccuracy, Graph Trend Filtering (GTF) and Smooth Eigenbase Pursuit (SEP) are adopted to filter out the initial noisy labels. GTF penalizes the l_0 norm of label difference between connected examples in the graph and exhibits better local adaptivity than the traditional l_2 norm-based Laplacian smoother. SEP reconstructs the correct labels by emphasizing the leading eigenvectors of Laplacian matrix associated with small eigenvalues, as these eigenvectors reflect real label smoothness and carry rich class separation cues. We term our algorithm as "Semi-supervised learning under Inadequate and Incorrect Supervision" (SIIS). Thorough experimental results on image classification, text categorization, and speech recognition demonstrate that our SIIS is effective in label error correction, leading to superior performance to the state-of-the-art methods in the presence of label noise and label scarcity.

Regression based method is a hot topic in the face recognition community and has achieved interesting results when dealing with well-aligned frontal face images. However, most of the existing regression analysis based methods are sensitive to pose variations. In this paper, we firstly introduce the orthogonal Procrustes problem (OPP), which is simple but effective, as a model to handle pose variations in two-dimensional face images. OPP seeks an optimal transformation between two images to correct the pose from one to the other. We integrate OPP into the regression model and propose the structural orthogonal Procrustes regression (SOPR) using the nuclear norm constraint on the error term to keep image's structural information. Moreover, a subject-wise strategy is adopted to address the problem that the gallery images may span over different poses. The proposed model is solved by an efficient iteratively reweighted algorithm and experimental results on popular face databases demonstrate the effectiveness of our method.

Feature selection, selecting the most informative subset of features, is an important research direction in dimension reduction. The combinatorial search in feature selection is essentially a binary optimization problem, known as NP hard, which can be alleviated by learning feature weights. Traditional feature weights algorithms rely on heuristic search path. These approaches neglect the interaction and dependency between different features, and thus provide no guarantee for optimality. In this paper, we propose a novel joint feature weights learning framework, which imposes both nonnegative and$\ell _{2,1}$-norm constraints on the feature weights matrix. The nonnegative property ensures the physical significance of learned feature weights. Meanwhile,$\ell _{2,1}$-norm minimization achieves joint selection of the most relevant features by exploiting the whole feature space. More importantly, an efficient iterative algorithm with proved convergence is designed to optimize a convex objective function. Using this framework as a platform, we propose new supervised and unsupervised joint feature selection methods. Particularly, in the proposed unsupervised method, nonnegative graph embedding is developed to exploit intrinsic structure in the weighted space. Comparative experiments on seven real-world data sets indicate that our framework is both effective and efficient.

Hierarchical time series (HTS) is a special class of multivariate time series where many related time series are organized in a hierarchical tree structure and they are consistent across hierarchy levels. HTS modeling is crucial and serves as the basis for business planning and management in many areas such as manufacturing inventory, energy and traffic management. However, due to machine failures, network disturbances or human maloperation, HTS data suffer from missing values across different hierarchical levels. In this paper, we study the missing value estimation problem under hierarchical web traffic settings, where the user-visit traffic are organized in various hierarchical structures, such as geographical structure and website structure. We develop an efficient algorithm, HTSImpute, to accurately estimate the missing value in multivariate noisy web traffic time series with specific hierarchical consistency in HTS settings. Our HTSImpute is able to (1) utilize the temporal dependence information within each individual time series, (2) exploit the intra-relations between time series through hierarchy, (3) guarantee the satisfaction of hierarchical consistency constraints. Results on three synthetic HTS datasets and three real-world hierarchical web traffic datasets demonstrate that our approach is able to provide more accurate and hierarchically consistent estimations than other baselines.