Search

Maximum a posteriori (MAP) inference is an important task for graphical models. Due to complex dependencies among variables in realistic models, finding an exact solution for MAP inference is often intractable. Thus, many approximation methods have been developed, among which the linear programming (LP) relaxation based methods show promising performance. However, one major drawback of LP relaxation is that it is possible to give fractional solutions. Instead of presenting a tighter relaxation, in this work we propose a continuous but equivalent reformulation of the original MAP inference problem, called LS–LP. We add the $$\ell _2$$-sphere constraint onto the original LP relaxation, leading to an intersected space with the local marginal polytope that is equivalent to the space of all valid integer label configurations. Thus, LS–LP is equivalent to the original MAP inference problem. We propose a perturbed alternating direction method of multipliers (ADMM) algorithm to optimize the LS–LP problem, by adding a sufficiently small perturbation $$\epsilon $$ onto the objective function and constraints. We prove that the perturbed ADMM algorithm globally converges to the $$\epsilon $$-Karush–Kuhn–Tucker ($$\epsilon $$-KKT) point of the LS–LP problem. The convergence rate will also be analyzed. Experiments on several benchmark datasets from Probabilistic Inference Challenge (PIC 2011) and OpenGM 2 show competitive performance of our proposed method against state-of-the-art MAP inference methods.

All-sky meteor radars are primarily used for meteor observations. This paper reports the first observations of ionospheric E-region field-aligned irregularities (FAIs) from a conventional all-sky meteor radar located at Wuhan (31°N, 114°E) for the period of March–June 2018. E-region FAI echoes evident in range-time intensity (RTI) maps show quasiperiodic striations with positive and negative slopes, which are consistent with multiple FAI structures moving across the wide beam of the meteor radar. A statistical analysis shows that out of a total of 111 d, there are 73 d with E-region FAI echoes detected by the meteor radar. The FAI events correspond well with the presence of sporadic-E layers which provide the necessary plasma density gradient for the development of gradient drift instability producing FAIs. The results demonstrate the capability of conventional meteor radars to make simultaneous routine observations of meteors and ionospheric E-region FAIs through incorporating RTI and spectral analysis into the online realtime data processing. Meteor radar observations could potentially address the limitations of ionospheric radars, which cannot provide simultaneous measurements of neutral winds and irregularity structures, and thereby contribute to better understanding of the dynamical processes producing E-region irregularities.

Facial action unit (AU) recognition has attracted great attention because of the applications in a wide range of fields. Missing labels and class imbalance (CIB) are both challenges for facial action unit recognition. Missing labels means that there are only apart label assignments for training samples. CIB is observed from two perspectives: firstly, the number of positive AUs is much smaller than that of negative AUs for each expressional image; secondly, the rate of positive samples of different AUs are significantly different. Both missing labels and CIB lead to performance degradation in AU recognition. In this work, we propose to handle these two challenges in AU recognition simultaneously. Specifically, we formulate AU recognition with missing labels as a multi label learning with missing labels (MLML) problem, which handles the missing label challenge naturally. However, different from most existing MLML approaches which usually employ same features from whole image for all classes, we select the most related features for each AU. To handle the CIB challenge, we further introduce class cardinality bounds which constrain the number of positive AUs for each data instance, as well as the number of positive labels for each AU in the overall dataset. The class cardinality bounds serve as linear constraints for the objective function, which turns the optimization NP-hard. Thus we present convex approximation based on the Lovasz extension, which leads to a linear program that can be efficiently solved by the alternative direction method of multipliers (ADMM). Experimental results on both posed and spontaneous facial expression datasets demonstrate the superiority of the proposed method compared to state-of-the-art.

This work focuses on the problem of multi-label learning with missing labels (MLML), which aims to label each test instance with multiple class labels given training instances that have an incomplete/partial set of these labels. The key point to handle missing labels is propagating the label information from provided labels to missing labels, through a dependency graph that each label of each instance is treated as a node. We build this graph by utilizing different types of label dependencies. Specifically, the instance-level similarity is served as undirected edges to connect the label nodes across different instances and the semantic label hierarchy is used as directed edges to connect different classes. This base graph is referred to as the mixed dependency graph, as it includes both undirected and directed edges. Furthermore, we present another two types of label dependencies to connect the label nodes across different classes. One is the class co-occurrence, which is also encoded as undirected edges. Combining with the base graph, we obtain a new mixed graph, called MG-CO (mixed graph with co-occurrence). The other is the sparse and low rank decomposition of the whole label matrix, to embed high-order dependencies over all labels. Combining with the base graph, the new mixed graph is called as MG-SL (mixed graph with sparse and low rank decomposition). Based on MG-CO and MG-SL, we propose two convex transductive formulations of the MLML problem, denoted as MLMG-CO and MLMG-SL, respectively. Two important applications, including image annotation and tag based image retrieval, can be jointly handled using our proposed methods. Experiments on benchmark datasets show that our methods give significant improvements in performance and robustness to missing labels over the state-of-the-art methods., Comment: Published in International Journal of Computer Vision, 2018

EAST is being developed by ASIPP for establishing the technology foundation of full superconducting tokamaks, which permits the operator to study the critical technology of future fusion energy power. Considering some in-vessel components were damaged during past experiments, it’s important to acquire the information of in-vessel components in EAST timely. However, it’s a challenge problem to complete the inspection task on account of the geometric parameters of the EAST vacuum vessel (VV) and the harsh working condition: radiation, ultra-high vacuum and high temperature. With the objective of inspecting the whole volume of vacuum vessel and in-vessel components, a in-vessel inspection robot system, named long reach robot system, has been designed, manufactured and integrated, which can be penetrated into the EAST VV through a limited size connection port. It’s a serial-chain long reach robot with many actuated joints characterized by high flexibility, large workspace and sufficient degrees of freedoms, which consists of a deploy system for support and guiding, a 8 m long manipulator for carrying the process tool to the desired location. It is specially designed and optimized for EAST without reducing the potential to be used for other future fusion devices. The kinematic has been studied for the purpose of investigating the motion of the long reach robot. The dynamic performance is studied by the simulation for giving reference to the selection of the components. The results show that the robot with appropriate kinematic and dynamic performance can achieve the design objective and satisfy the requirements of the inspection.

Experimental advanced superconducting tokamak (EAST) is a fusion engineering test reactor, it can provide help for acquiring the key technology of fusion energy power. It’s important to obtain the information of the in-vessel components and detect abnormal events timely in EAST. However, it’s difficult to inspect the whole vacuum vessel and plasma facing components under radiation, ultra-high vacuum and temperature constraints. A long reach robot system developed for EAST inspection is introduced. The long reach robot penetrate into the EAST chamber using the size-limited hole with process tool installed on it. The robot, which is designed with the consideration of remote handling, small weight and easy control, has a long reach multi-link structure with the characteristics of numerous degrees of freedoms, high flexibility and large workspace. The robot could be part of the remote handling system for EAST and has the potential ability to be used for other fusion devices. The conceptual design and mechanical structure of the robot are described. The kinematic analyses, including kinematic model and workspace investigation, are discussed. The results indicate that the robot shows appropriate kinematic performance and can meet the requirements of the inspection process. The prototype of the robot has been built and the further validation will be carry on in the future.

Six-axis force/torque sensor (F/T sensor) that can detect three orthogonal forces and torques has been extensively adopted in humanoid robot foot for stable control. Due to the F/T sensors being seriously influenced by the electromagnetic interference from ankle generator, signal stability of F/T sensor is still one of the main influencing factors for accurately dynamical and stable control. Though the problem with influencing electromagnetic noise can be solved generally and partially by shielding of the sensor and wires, this traditional method will to some extend face implementation difficulty induced by special mounting space and application environment especially in humanoid robot ankle. Therefore, the research on the output signal stability and precision of the F/T sensors has still been one of the most essential subjects for humanoid robot dynamic equilibrium control. In this paper, some various signal processing methods have been adopted and analyzed comparatively with the aim at the output signal anti-interference processing of F/T sensors. And the filtering effect and its feasibility were verified experimentally in the dynamic walking motion of humanoid robot platform BHR-2.

The data observed by a spectral airglow temperature imager (SATI) at Beijing National Observatory of Space Environment from July 23, 2008 to July 31, 2009 are used to study night mesopause temperature in Beijing. From variations of temperature at 87 and 94 km obtained from OH (6-2) and O2 (0–1) airglow spectra, temperature at night is shown lowest in the summer and highest in the winter. In summer, average temperature at 87 km is 173.9 K, lower than average temperature 180.1 K at 94 km. But in winter, average temperature at 87 km is 201.2 K, higher than average temperature 194.8 K at 94 km. The altitude of mesopause in Beijing is below 87 km in summer and above 94 km in winter. There are about 120–150 days when the mesopause locates below 87 km, which is in agreement with the results of SABER/TIMED. Variations of temperatures at 87 and 94 km are analyzed by harmonic method. Our results show that amplitudes of annual oscillation of temperature at 87 and 94 km are 17.5 and 7.8 K respectively. Amplitudes of semi-annual oscillation at 87 and 94 km are 1.6 and 5.3 K, which are smaller than those of annual oscillation. Although there are differences among different observations because of different locations and different instruments, our results are in general agreement with observation at similar latitude as Beijing.