Search

There is an urgent need for an effective video classification method by means of a small number of samples. The deficiency of samples could be effectively alleviated by generating samples through Generative Adversarial Networks (GAN), but the generation of videos on a typical category remains to be underexplored since the complex actions and the changeable viewpoints are difficult to simulate. In this paper, we propose a generative data augmentation method for temporal stream of the Temporal Segment Networks with the dynamic image. The dynamic image compresses the motion information of video into a still image, removing the interference factors such as the background. Thus it is easier to generate images with categorical motion information using GAN. We use the generated dynamic images to enhance the features, with regularization achieved as well, thereby to achieve the effect of video augmentation. In order to deal with the uneven quality of generated images, we propose a Self-Paced Selection (SPS) method, which automatically selects the high-quality generated samples to be added to the network training. Our method is verified on two benchmark datasets, HMDB51 and UCF101. The experimental results show that the method can improve the accuracy of video classification under the circumstance of sample insufficiency and sample imbalance.

Insufficient labeled training datasets is one of the bottlenecks of 3D hand pose estimation from monocular RGB images. Synthetic datasets have a large number of images with precise annotations, but the obvious difference with real-world datasets impacts the generalization. Little work has been done to bridge the gap between two domains over their wide difference. In this paper, we propose a domain adaptation method called Adaptive Wasserstein Hourglass (AW Hourglass) for weakly-supervised 3D hand pose estimation, which aims to distinguish the difference and explore the common characteristics (e.g. hand structure) of synthetic and real-world datasets. Learning the common characteristics helps the network focus on pose-related information. The similarity of the characteristics makes it easier to enforce domain-invariant constraints. During training, based on the relation between these common characteristics and 3D pose learned from fully-annotated synthetic datasets, it is beneficial for the network to restore the 3D pose of weakly labeled real-world datasets with the aid of 2D annotations and depth images. While in testing, the network predicts the 3D pose with the input of RGB.

As a green and secure wireless transmission way, secure spatial modulation (SM) is becoming a hot research area. Its basic idea is to exploit both the index of activated transmit antenna and amplitude phase modulation (APM) signal to carry messages, improve security, and save energy. In this paper, we reviewed its crucial techniques: transmit antenna selection (TAS), artificial noise (AN) projection, power allocation (PA), and joint detection at desired receiver. To achieve the optimal performance of maximum likelihood (ML) detector, a deep-neural-network (DNN) joint detector is proposed to jointly infer the index of transmit antenna and signal constellation point with a lower-complexity. Here, each layer of DNN is redesigned to optimize the joint inference performance of two distinct types of information: transmit antenna index and signal constellation point. Simulation results show that the proposed DNN method performs 3dB better than the conventional DNN structure and is close to ML detection in the low and medium signal-to-noise ratio regions in terms of the bit error rate (BER) performance, but its complexity is far lower-complexity compared to ML. Finally, three key techniques TAS, PA, and AN projection at transmitter can be combined to make SM a true secure modulation.

Ability to quantify and predict progression of a disease is fundamental for selecting an appropriate treatment. Many clinical metrics cannot be acquired frequently either because of their cost (e.g. MRI, gait analysis) or because they are inconvenient or harmful to a patient (e.g. biopsy, x-ray). In such scenarios, in order to estimate individual trajectories of disease progression, it is advantageous to leverage similarities between patients, i.e. the covariance of trajectories, and find a latent representation of progression. Most of existing methods for estimating trajectories do not account for events in-between observations, what dramatically decreases their adequacy for clinical practice. In this study, we develop a machine learning framework named Coordinatewise-Soft-Impute (CSI) for analyzing disease progression from sparse observations in the presence of confounding events. CSI is guaranteed to converge to the global minimum of the corresponding optimization problem. Experimental results also demonstrates the effectiveness of CSI using both simulated and real dataset.

For several models of random constraint satisfaction problems, it was conjectured by physicists and later proved that a sharp satisfiability transition occurs. For random $k$-SAT and related models it happens at clause density $\alpha$ around $2^k$. Just below the threshold, further results suggest that the solution space has a "1RSB" structure of a large bounded number of near-orthogonal clusters inside the space of variable assignments $\{0,1\}^N$. In the unsatisfiable regime, it is natural to consider max-satisfiability: violating the least number of constraints. For a simplified variant, the strong refutation problem, there is strong evidence that an algorithmic transition occurs around $\alpha = N^{k/2-1}$. For $\alpha$ bounded in $N$, a very precise estimate of the max-sat value was obtained by Achlioptas, Naor, and Peres (2007), but it is not sharp enough to indicate the nature of the energy landscape. Later work (Sen, 2016; Panchenko, 2016) shows that for $\alpha$ very large (roughly, above $64^k$) the max-sat value approaches the mean-field (complete graph) limit: this is conjectured to have an "FRSB" structure where near-optimal configurations form clusters within clusters, in an ultrametric hierarchy of infinite depth inside $\{0,1\}^N$. A stronger form of FRSB was shown in several recent works to have algorithmic implications (again, in complete graphs). Consequently we find it of interest to understand how the model transitions from 1RSB near the satisfiability threshold, to (conjecturally) FRSB for large $\alpha$. In this paper we show that in the random regular $k$-NAE-SAT model, the 1RSB description breaks down already above $\alpha \asymp 4^k/k^3$. This is proved by an explicit perturbation in the 2RSB parameter space, inspired by the "bug proliferation" mechanism proposed by physicists (Montanari and Ricci-Tersenghi, 2003; Krzakala, Pagnani, and Weigt, 2004).