Causal inference for out-of-distribution recognition via sample balancing

Image classification algorithms are commonly based on the Independent and Identically Distribution (i.i.d.) assumption, but in practice, the Out-Of-Distribution (OOD) problem widely exists, that is, the contexts of images in the model predicting are usually unseen during training. In this case, existing models trained under the i.i.d. assumption are limiting generalisation. Causal inference is an important method to learn the causal associations which are invariant across different environments, thus improving the generalisation ability of the model. However, existing methods usually require partitioning of the environment to learn invariant features, which mostly have imbalance problems due to the lack of constraints. In this paper, we propose a balanced causal learning framework (BCL), starting from how to divide the dataset in a balanced way and the balance of training after the division, which automatically generates fine-grained balanced data partitions in an unsupervised manner and balances the training difficulty of different classes, thereby enhancing the generalisation ability of models in different environments. Experiments on the OOD datasets NICO and NICO++ demonstrate that BCL achieves stable predictions on OOD data, and we also find that models using BCL focus more accurately on the foreground of images compared with the existing causal inference method, which effectively improves the generalisation ability.

Rapid epitaxy of 2-inch and high-quality α-Ga_(2)O_(3) films by mist-CVD method

High thickness uniformity and large-scale films of α-Ga_(2)O_(3) are crucial factors for the development of power devices.In this work, a high-quality 2-inch α-Ga_(2)O_(3) epitaxial film on c-plane sapphire substrates was prepared by the mist-CVD method.The growth rate and phase control mechanisms were systematically investigated. The growth rate of the α-Ga_(2)O_(3) films was limited by the evaporation of the microdroplets containing gallium acetylacetonate. By adjusting the substrate position(z) from 80 to 50 mm, the growth rate was increased from 307 nm/h to 1.45 μm/h when the growth temperature was fixed at 520 °C.When the growth temperature exceeded 560 °C, ε-Ga_(2)O_(3) was observed to form at the edges of 2-inch sapphire substrate.Phase control was achieved by adjusting the growth temperature. When the growth temperature was 540 °C and the substrate position was 50 mm, the full-width at half maximum(FWHM) of the rocking curves for the(0006) and(10-14) planes were 0.023° and 1.17°. The screw and edge dislocations were 2.3 × 10~6 and 3.9 × 10~(10)cm~(-2), respectively. Furthermore, the bandgaps and optical transmittance of α-Ga_(2)O_(3) films grown under different conditions were characterized utilizing UV-visible and near-IR scanning spectra.

Extracting useful high-frequency information from wide-field electromagnetic data using time-domain signal reconstruction

The wide-field electromagnetic method is widely used in hydrocarbon exploration,mineral deposit detection,and geological disaster prediction.However,apparent resistivity and normalized field amplitude exceeding 2048 Hz often exhibit upward warping in data,making geophysical inversion and interpretation challenging.The cumulative error of the crystal oscillator in signal transmission and acquisition contributes to an upturned apparent resistivity curve.To address this,a high-frequency information extraction method is proposed based on time-domain signal reconstruction,which helps to record a complete current data sequence;moreover,it helps estimate the crystal oscillator error for the transmitted signal.Considering the recorded error,a received signal was corrected using a set of reconstruction algorithms.After processing,the high-frequency component of the wide-field electromagnetic data was not upturned,while accurate high-frequency information was extracted from the signal.Therefore,the proposed method helped effectively extract high-frequency components of all wide-field electromagnetic data.

High quality crystal growth and characterization of ferroelectric Ba_(0.77)Ca_(0.23)TiO_(3) single crystal

Ba_(0.77)Ca_(0.23)TiO_(3)(BCT)single crystal has been widely studied as a promising lead-free ferroelectric material.In this work,high-quality BCT crystal was successfully grown by the Czochralski(CZ)method.The as-grown crystal is crack-free and shows black coloration.It possesses a high dielectric stability over a wide temperature range,while the dielectric loss is rather small below 90℃.Furthermore,it possesses excellent ferroelectric properties with residual polarization strength(Pr)and coercive field(Ec)of 17.93μC/cm^(2) and 8.47 kV/cm,respectively.Besides,BCT crystal shows large electromechanical coupling factors,with kt,k31,k33 and k15 of 0.535,0.254,0.714 and 0.721,respectively.The piezoelectric coefficients d31,d33 and d15 are measured to be−36.5,130 and 246 pC/N,respectively.

Design and Experimental Validation of a Worm-Like Tensegrity Robot for In-Pipe Locomotion

Traditional rigid-body in-pipe robots usually have complex and heavy structures with limited flexibility and adaptability.Although soft in-pipe robots have great improvements in flexibility,they still have manufacturing difficulties due to their reliance on high-performance soft materials.Tensegrity structure is a kind of self-stressed spatial structure consisting discrete rigid struts connected by a continuous net of tensional flexible strings,which combines the advantages of both rigid structures and soft structures.By applying tensegrity structures into robotics,this paper proposes a novel worm-like tensegrity robot for moving inside pipes.First,a robot module capable of body deformation is designed based on the concept of tensegrity and its deformation performance is analyzed.Then,the optimal parameters of the module are obtained based on the tensegrity form-finding.The deformation ability of the tensegrity module is tested experimentally.Finally,the worm-like tensegrity robot that can crawl inside pipes is developed by connecting three modules in series.Motion performance and load capacity are tested on the prototype of the worm-like tensegrity robot by experiments of moving in horizontal pipe,vertical pipe,and elbow pipe.Experimental results demonstrate the effectiveness of the proposed design and suggest that the robot has high compliance,mobility,and adaptability although with simple structure and low cost.

A person-following method based on monocular camera for quadruped robots

This paper proposes a person-following method based on monocular vision,which allows quadruped robots to track a target person in both indoor and outdoor environments with different illumination conditions.Our method is composed of a person detector,a Kalman filter(KF)tracker,and a re-identification module.To be more specific,the person detector uses a human pose estimation method to detect persons.The KF is applied to predict the position of the target person and update its state with detection results.A re-identification module is proposed to deal with distractions,where the Convolutional Channel Features(CCF)is used to extract appearance features and Online Boosting is used to distinguish the target person from others.Especially,we design a target recapture mechanism based on the Recurrent Neural Network(RNN).Combining motion information with appearance features,the system can accurately re-identify the target person.Without extra customized markers,our method can track the target person steadily in real-time only using a monocular camera.Experiments results can validate the robustness and effectiveness of the proposed method.