Motivated by applications to online learning in sparse estimation and Bayesian optimization, we consider the problem of online unconstrained nonsubmodular minimization with delayed costs in both full information and bandit feedback settings. In contrast to previous works on online unconstrained submodular minimization, we focus on a class of nonsubmodular functions with special structure, and prove regret guarantees for several variants of the online and approximate online bandit gradient descent algorithms in static and delayed scenarios. We derive bounds for the agent's regret in the full information and bandit feedback setting, even if the delay between choosing a decision and receiving the incurred cost is unbounded. Key to our approach is the notion of $(\alpha, \beta)$-regret and the extension of the generic convex relaxation model from~\citet{El-2020-Optimal}, the analysis of which is of independent interest. We conduct and showcase several simulation studies to demonstrate the efficacy of our algorithms.
Mainstream image caption models are usually two-stage captioners, i.e., calculating object features by pre-trained detector, and feeding them into a language model to generate text descriptions. However, such an operation will cause a task-based information gap to decrease the performance, since the object features in detection task are suboptimal representation and cannot provide all necessary information for subsequent text generation. Besides, object features are usually represented by the last layer features that lose the local details of input images. In this paper, we propose a novel One-Stage Image Captioner (OSIC) with dynamic multi-sight learning, which directly transforms input image into descriptive sentences in one stage. As a result, the task-based information gap can be greatly reduced. To obtain rich features, we use the Swin Transformer to calculate multi-level features, and then feed them into a novel dynamic multi-sight embedding module to exploit both global structure and local texture of input images. To enhance the global modeling of encoder for caption, we propose a new dual-dimensional refining module to non-locally model the interaction of the embedded features. Finally, OSIC can obtain rich and useful information to improve the image caption task. Extensive comparisons on benchmark MS-COCO dataset verified the superior performance of our method.
We present a new distribution-free conformal prediction algorithm for sequential data (e.g., time series), called the \textit{sequential predictive conformal inference} (\texttt{SPCI}). We specifically account for the nature that the time series data are non-exchangeable, and thus many existing conformal prediction algorithms based on temporal residuals are not applicable. The main idea is to exploit the temporal dependence of conformity scores; thus, the past conformity scores contain information about future ones. Then we cast the problem of conformal prediction interval as predicting the quantile of a future residual, given a prediction algorithm. Theoretically, we establish asymptotic valid conditional coverage upon extending consistency analyses in quantile regression. Using simulation and real-data experiments, we demonstrate a significant reduction in interval width of \texttt{SPCI} compared to other existing methods under the desired empirical coverage.
This paper presents a Temporal Graph Neural Network (TGNN) framework for detection and localization of false data injection and ramp attacks on the system state in smart grids. Capturing the topological information of the system through the GNN framework along with the state measurements can improve the performance of the detection mechanism. The problem is formulated as a classification problem through a GNN with message passing mechanism to identify abnormal measurements. The residual block used in the aggregation process of message passing and the gated recurrent unit can lead to improved computational time and performance. The performance of the proposed model has been evaluated through extensive simulations of power system states and attack scenarios showing promising performance. The sensitivity of the model to intensity and location of the attacks and model's detection delay versus detection accuracy have also been evaluated.
When searching for policies, reward-sparse environments often lack sufficient information about which behaviors to improve upon or avoid. In such environments, the policy search process is bound to blindly search for reward-yielding transitions and no early reward can bias this search in one direction or another. A way to overcome this is to use intrinsic motivation in order to explore new transitions until a reward is found. In this work, we use a recently proposed definition of intrinsic motivation, Curiosity, in an evolutionary policy search method. We propose Curiosity-ES, an evolutionary strategy adapted to use Curiosity as a fitness metric. We compare Curiosity with Novelty, a commonly used diversity metric, and find that Curiosity can generate higher diversity over full episodes without the need for an explicit diversity criterion and lead to multiple policies which find reward.
The lack of large-scale datasets has been impeding the advance of deep learning approaches to the problem of F-formation detection. Moreover, most research works on this problem rely on input sensor signals of object location and orientation rather than image signals. To address this, we develop a new, large-scale dataset of simulated images for F-formation detection, called F-formation Simulation Dataset (F2SD). F2SD contains nearly 60,000 images simulated from GTA-5, with bounding boxes and orientation information on images, making it useful for a wide variety of modelling approaches. It is also closer to practical scenarios, where three-dimensional location and orientation information are costly to record. It is challenging to construct such a large-scale simulated dataset while keeping it realistic. Furthermore, the available research utilizes conventional methods to detect groups. They do not detect groups directly from the image. In this work, we propose (1) a large-scale simulation dataset F2SD and a pipeline for F-formation simulation, (2) a first-ever end-to-end baseline model for the task, and experiments on our simulation dataset.
The appearance of the same object may vary in different scene images due to perspectives and occlusions between objects. Humans can easily identify the same object, even if occlusions exist, by completing the occluded parts based on its canonical image in the memory. Achieving this ability is still a challenge for machine learning, especially under the unsupervised learning setting. Inspired by such an ability of humans, this paper proposes a compositional scene modeling method to infer global representations of canonical images of objects without any supervision. The representation of each object is divided into an intrinsic part, which characterizes globally invariant information (i.e. canonical representation of an object), and an extrinsic part, which characterizes scene-dependent information (e.g., position and size). To infer the intrinsic representation of each object, we employ a patch-matching strategy to align the representation of a potentially occluded object with the canonical representations of objects, and sample the most probable canonical representation based on the category of object determined by amortized variational inference. Extensive experiments are conducted on four object-centric learning benchmarks, and experimental results demonstrate that the proposed method not only outperforms state-of-the-arts in terms of segmentation and reconstruction, but also achieves good global object identification performance.
Self-supervised facial representation has recently attracted increasing attention due to its ability to perform face understanding without relying on large-scale annotated datasets heavily. However, analytically, current contrastive-based self-supervised learning still performs unsatisfactorily for learning facial representation. More specifically, existing contrastive learning (CL) tends to learn pose-invariant features that cannot depict the pose details of faces, compromising the learning performance. To conquer the above limitation of CL, we propose a novel Pose-disentangled Contrastive Learning (PCL) method for general self-supervised facial representation. Our PCL first devises a pose-disentangled decoder (PDD) with a delicately designed orthogonalizing regulation, which disentangles the pose-related features from the face-aware features; therefore, pose-related and other pose-unrelated facial information could be performed in individual subnetworks and do not affect each other's training. Furthermore, we introduce a pose-related contrastive learning scheme that learns pose-related information based on data augmentation of the same image, which would deliver more effective face-aware representation for various downstream tasks. We conducted a comprehensive linear evaluation on three challenging downstream facial understanding tasks, i.e., facial expression recognition, face recognition, and AU detection. Experimental results demonstrate that our method outperforms cutting-edge contrastive and other self-supervised learning methods with a great margin.
3D hand pose estimation from RGB images suffers from the difficulty of obtaining the depth information. Therefore, a great deal of attention has been spent on estimating 3D hand pose from 2D hand joints. In this paper, we leverage the advantage of spatial-temporal Graph Convolutional Neural Networks and propose LG-Hand, a powerful method for 3D hand pose estimation. Our method incorporates both spatial and temporal dependencies into a single process. We argue that kinematic information plays an important role, contributing to the performance of 3D hand pose estimation. We thereby introduce two new objective functions, Angle and Direction loss, to take the hand structure into account. While Angle loss covers locally kinematic information, Direction loss handles globally kinematic one. Our LG-Hand achieves promising results on the First-Person Hand Action Benchmark (FPHAB) dataset. We also perform an ablation study to show the efficacy of the two proposed objective functions.
Multi-view representation learning has developed rapidly over the past decades and has been applied in many fields. However, most previous works assumed that each view is complete and aligned. This leads to an inevitable deterioration in their performance when encountering practical problems such as missing or unaligned views. To address the challenge of representation learning on partially aligned multi-view data, we propose a new cross-view graph contrastive learning framework, which integrates multi-view information to align data and learn latent representations. Compared with current approaches, the proposed method has the following merits: (1) our model is an end-to-end framework that simultaneously performs view-specific representation learning via view-specific autoencoders and cluster-level data aligning by combining multi-view information with the cross-view graph contrastive learning; (2) it is easy to apply our model to explore information from three or more modalities/sources as the cross-view graph contrastive learning is devised. Extensive experiments conducted on several real datasets demonstrate the effectiveness of the proposed method on the clustering and classification tasks.