Learning Rotations

Many problems in computer vision today are solved via deep learning. Tasks like pose estimation from images, pose estimation from point clouds or structure from motion can all be formulated as a regression on rotations. However, there is no unique way of parametrizing rotations mathematically: matrices, quaternions, axis-angle representation or Euler angles are all commonly used in the field. Some of them, however, present intrinsic limitations, including discontinuities, gimbal lock or antipodal symmetry. These limitations may make the learning of rotations via neural networks a challenging problem, potentially introducing large errors. Following recent literature, we propose three case studies: a sanity check, a pose estimation from 3D point clouds and an inverse kinematic problem. We do so by employing a full geometric algebra (GA) description of rotations. We compare the GA formulation with a 6D continuous representation previously presented in the literature in terms of regression error and reconstruction accuracy. We empirically demonstrate that parametrizing rotations as bivectors outperforms the 6D representation. The GA approach overcomes the continuity issue of representations as the 6D representation does, but it also needs fewer parameters to be learned and offers an enhanced robustness to noise. GA hence provides a broader framework for describing rotations in a simple and compact way that is suitable for regression tasks via deep learning, showing high regression accuracy and good generalizability in realistic high-noise scenarios.

Development of a Kriging Assessment Framework and Impact Pathway for Researching Contamination With Potentially Toxic Elements in Urban Landfill Soil

Landfill post-closure with contaminant concentration in soil below permissible limit assessed at limited spot does not represent the contamination issue. Assessment limit to professionals also does not gives a potential of change to practice constant assessment to a wider context of assessor - citizen living nearby - as a collaborative effort to sustain a safe environment. Therefore sizeable, qualitative, and cost-effective analysis of the concentrations of contaminants is needed and this work recommends kriging assessment and the logical impact pathway framework as factors of change in landfill aftercare management. The kriging framework is developed utilising lead (Pb) and chromium (Cr) data from inductively coupled plasma mass spectrometry (ICP-MS) analysis. The development of the kriging framework is conducted based on the observation of censored data from ICP-MS analysis. The estimation analysis involves the analysis of ordinary kriging with regression analysis, showing the interpolation of spatial correlation and regression error. Hence, ordinary kriging with regression of the variable of interest, i.e., Pb, using the data of the explanatory variable, i.e., Cr, is inappropriate. Further investigation with the utilisation of guess-field kriging analysis hypothetically exposed a potential contaminated area using an existing but limited number of explanatory variables; although, guess-field kriging may possibly result immense uncertainty at the area where the explanatory variable does not exist. Besides, this work anticipated outcomes in societal impact and sustainability practices from the proposed kriging framework by recommending a logical impact pathway. The development of the kriging framework and impact pathway reassure the necessary actions to be executed by responsible parties and act as the stimulus of a wider spectrum of improvement initiatives to oversee real issues, such as the time of occurrence, and to prevent negative impacts on the environment and humans.

Research on the Calculation Method of Harmonic Contribution in the Case of Dominant System-side Harmonics

Existing methods for quantifying the responsibility of harmonic sources assume a dominant user side and use a harmonic source equivalence circuit to calculate the equivalent system impedance and background harmonic voltage, which in turn assesses the harmonic contribution of that source to the bus of concern. For users who actively participate in harmonic governance, it is very important to evaluate the responsibility of injecting harmonics into users. This paper assumes system-side is dominant, constructs a partial linear regression model and a constant impedance model, and tracks the regression error. The equivalent fundamental impedance is doubly screened to calculate the harmonic impedance for the corresponding number of times, which in turn quantifies the harmonic voltage duty. The results of simulation and the analysis of measured data show that this method has simple calculation model, small regression error (0.0037), high accuracy and practical engineering significance.

MuscleNET: Mapping Electromyography to kinematic and dynamic biomechanical variables by machine learning

Objective: This paper proposes machine learning models for mapping surface electromyography (sEMG) signals to regression of joint angle, joint velocity, joint acceleration, joint torque, and activation torque. Approach: The regression models, collectively known as MuscleNET, take one of four forms: ANN (Forward Artificial Neural Network), RNN (Recurrent Neural Network), CNN (Convolutional Neural Network), and RCNN (Recurrent Convolutional Neural Network). Inspired by conventional biomechanical muscle models, delayed kinematic signals were used along with sEMG signals as the machine learning model's input; specifically, the CNN and RCNN were modeled with novel configurations for these input conditions. The models' inputs contain either raw or filtered sEMG signals, which allowed evaluation of the filtering capabilities of the models. The models were trained using human experimental data and evaluated with different individual data. Main results: Results were compared in terms of regression error (using the root-mean-square) and model computation delay. The results indicate that the RNN (with filtered sEMG signals) and RCNN (with raw sEMG signals) models, both with delayed kinematic data, can extract underlying motor control information (such as joint activation torque or joint angle) from sEMG signals in pick-and-place tasks. The CNNs and RCNNs were able to filter raw sEMG signals. Significance: All forms of MuscleNET were found to map sEMG signals within 2 ms, fast enough for real-time applications such as the control of exoskeletons or active prostheses. The RNN model with filtered sEMG and delayed kinematic signals is particularly appropriate for applications in musculoskeletal simulation and biomechatronic device control.

A Stochastic Estimation Framework for Yearly Evolution of Worldwide Electricity Consumption

The determination of electric energy consumption is remarked as one of the most vital objectives for electrical engineers as it is highly essential in determining the actual energy demand made on the existing electricity supply. Therefore, it is important to find out about the increasing trend in electric energy demands and use all over the world. In this work, we present a prediction scheme for the progression of worldwide aggregates of cumulative electricity consumption using the time series of the records released annually for the net electricity use throughout the world. Consequently, we make use of an autoregressive (AR) model by retaining the best possible autoregression order recording the highest regression accuracy and the lowest standardized regression error. The resultant regression scheme was proficiently employed to regress and forecast the evolution of next-decade data for the net consumption of electricity worldwide from 1980 to 2019 (in billion kilowatt-hours). The experimental outcomes exhibited that the highest accuracy in regressing and forecasting the global consumption of electricity is 95.7%. The prediction results disclose a linearly growing trend in the amount of electricity issued annually over the past four decades’ observation for the global net electricity consumption dataset.

Model-size reduction for reservoir computing by concatenating internal states through time

Reservoir computing (RC) is a machine learning algorithm that can learn complex time series from data very rapidly based on the use of high-dimensional dynamical systems, such as random networks of neurons, called “reservoirs.” To implement RC in edge computing, it is highly important to reduce the amount of computational resources that RC requires. In this study, we propose methods that reduce the size of the reservoir by inputting the past or drifting states of the reservoir to the output layer at the current time step. To elucidate the mechanism of model-size reduction, the proposed methods are analyzed based on information processing capacity proposed by Dambre et al. (Sci Rep 2:514, 2012). In addition, we evaluate the effectiveness of the proposed methods on time-series prediction tasks: the generalized Hénon-map and NARMA. On these tasks, we found that the proposed methods were able to reduce the size of the reservoir up to one tenth without a substantial increase in regression error.

Model-Size Reduction for Reservoir Computing by Concatenating Internal States Through Time

Reservoir computing (RC) is a machine learning algorithm that can learn complex time series from data very rapidly based on the use of high-dimensional dynamical systems, such as random networks of neurons, called "reservoirs." To implement RC in edge computing, it is highly important to reduce the amount of computational resources that RC requires. In this study, we propose methods that reduce the size of the reservoir by inputting the past or drifting states of the reservoir to the output layer at the current time step. To elucidate the mechanism of model-size reduction, the proposed methods are analyzed based on information processing capacity proposed by Dambre et al. (2012). In addition, we evaluate the effectiveness of the proposed methods on time-series prediction tasks: the generalized Hénon-map and NARMA. On these tasks, we found that the proposed methods were able to reduce the size of the reservoir up to one tenth without a substantial increase in regression error.

ROBUSTIFIED EXPECTED MAXIMUM PRODUCTION FRONTIERS

The aim of this paper is to construct a robust nonparametric estimator for the production frontier. We study this problem under a regression model with one-sided errors, where the regression function defines the achievable maximum output, for a given level of inputs-usage, and the regression error defines the inefficiency term. The main tool is a concept of partial regression boundary defined as a special probability-weighted moment. This concept motivates a robustified unconditional alternative to the pioneering class of nonparametric conditional expected maximum production functions. We prove that both the resulting benchmark partial frontier and its estimator share the desirable monotonicity of the true full frontier. We derive the asymptotic properties of the partial and full frontier estimators, and unravel their behavior from a robustness theory point of view. We provide numerical illustrations and Monte Carlo evidence that the presented concept of unconditional expected maximum production functions is more efficient and reliable in filtering out noise than the original conditional version. The methodology is very easy and fast to implement. Its usefulness is discussed through two concrete datasets from the sector of Delivery Services, where outliers are likely to affect the traditional conditional approach.