Robust Tuning Research Articles

Hyper-tune

The ever-growing demand and complexity of machine learning are putting pressure on hyper-parameter tuning systems: while the evaluation cost of models continues to increase, the scalability of state-of-the-arts starts to become a crucial bottleneck. In this paper, inspired by our experience when deploying hyper-parameter tuning in a real-world application in production and the limitations of existing systems, we propose Hyper-Tune, an efficient and robust distributed hyper-parameter tuning framework. Compared with existing systems, Hyper-Tune highlights multiple system optimizations, including (1) automatic resource allocation, (2) asynchronous scheduling, and (3) multi-fidelity optimizer. We conduct extensive evaluations on benchmark datasets and a large-scale real-world dataset in production. Empirically, with the aid of these optimizations, Hyper-Tune outperforms competitive hyper-parameter tuning systems on a wide range of scenarios, including XGBoost, CNN, RNN, and some architectural hyper-parameters for neural networks. Compared with the state-of-the-art BOHB and A-BOHB, Hyper-Tune achieves up to 11.2X and 5.1X speedups, respectively.

Flexible encoding of objects and space in single cells of the dentate gyrus

The hippocampus is involved in the formation of memories that require associations among stimuli to construct representations of space and the items and events within that space. Neurons in the dentate gyrus (DG), an initial input region of the hippocampus, have robust spatial tuning, but it is unclear how nonspatial information may be integrated with spatial activity in this region. We recorded from the DG of 21 adult mice as they foraged for food in an environment that contained discrete objects. We found DG cells with multiple firing fields at a fixed distance and direction from objects (landmark vector cells) and cells that exhibited localized changes in spatial firing when objects in the environment were manipulated. By classifying recorded DG cells into putative dentate granule cells and mossy cells, we examined how the addition or displacement of objects affected the spatial firing of these DG cell types. Object-related activity was detected in a significant proportion of mossy cells. Although few granule cells with responses to object manipulations were recorded, likely because of the sparse nature of granule cell firing, there was generally no significant difference in the proportion of granule cells and mossy cells with object responses. When mice explored a second environment with the same objects, DG spatial maps completely reorganized, and a different subset of cells responded to object manipulations. Together, these data reveal the capacity of DG cells to detect small changes in the environment while preserving a stable spatial representation of the overall context.

Biased weighting of temporally discrete visual stimuli in a continuous report decision-making task: A combined behavioral and electrophysiological study.

Integrating evidence from multiple sources to guide decisions is something humans do on a daily basis. Existing research suggests that not all sources of information are weighted equally in decision-making tasks, and that observers are subject to biases in the face of internal and external noise. Here we describe two experiments that measured observers' ability to integrate successive visual signals. Participants viewed pairs of gratings presented sequentially and reproduced their average orientation. Experiment 1 revealed a recency bias in evidence integration, such that observers' average judgments were closer to the orientation of the second grating than the first. Mixture distribution modeling revealed that this was caused by a recency bias in averaging, as well as a tendency to disregard the first stimulus altogether in some trials. In Experiment 2 we replicated these findings, and quantified orientation-specific patterns of neural activity recorded during the task using population-tuning curve modeling of electroencephalography data. This analysis yielded robust orientation tuning to both the presented gratings and observers' decisions, and suggested that observers were storing both grating stimuli for subsequent averaging rather than computing a running average. The neural representation of the second grating was not reliably stronger than that of the first, suggesting that the recency bias is not due to a difference in the strength of encoding of the second stimulus, and instead may arise at a later decision stage where information is retrieved or integrated. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Normalized passivity control for robust tuning in real‐time hybrid tests

AbstractReal‐time hybrid testing involves the separation of a system into an experimental component and a numerically simulated substructure which are coupled and run together. The coupling between substructures is achieved using actuators and force sensors which comprise the transfer system. Close synchronization is required between substructures for reliable hybrid testing. However, actuator lag may cause tracking errors and instability in hybrid tests. Existing lag compensation schemes require identification of the coupled dynamics of the transfer system and experimental component and can be sensitive to changes in these components. Passivity control is a technique intended to maintain stability without the need for system identification or assumptions about the actuators or test specimens. Yet, the tuning of existing passivity controllers is sensitive to both the system being tested and the amplitude and frequency range excited. This paper presents a new, normalized passivity controller which behaves well across a much broader range of operating conditions once tuned for a single‐test scenario. The proposed approach uses a virtual damping element on the numerical substructure to dissipate spurious power injected by the actuator into the system, based on the ratio of net power output to mean power throughput. The scheme has been shown to result in identical performance for a linear hybrid test with a range of step excitations from 0.5 mm up to 500 mm. The proposed method can be used to improve test stability and fidelity in isolation or alongside other compensation schemes to further improve performance.

Inerter-Connected Double Tuned Mass Damper for Passive Control of Buildings under Seismic Excitation

In current work the Inerter-Connected Double Tuned Mass Damper (ICDTMD) is employed for structural control of a well-recognized benchmark 10-story linear shear building. The ICDTMD is introduced to overcome the practical limitations of the roof-top tuned mass damper inerter (TMDI), in which the second terminal of the inerter is connected to the lower floors of the building. To this end, a modification of the double tuned mass damper (DTMD) with a linking inerter is proposed to not only exploit the promising features of inerter but also surmount the architectural interference of the effective TMDI configurations. The TMDs free parameters are optimized using particle swarm optimization algorithm (PSO) for two different single objective functions, i.e., the H∞ norms of roof displacement and story drifts were minimized for robust tuning. To evaluate the robustness of the optimal damper, its performance was compared to a traditional roof-top single tuned mass damper (STMD) and DTMD in both frequency and time domains (time history analysis for four far- and near-field records) for different preselected mass and inertance ratios. The performance indices in the time domain were selected as the maximum story drifts, story acceleration and shear. Results show that the rooftop ICDTMD, unlike the rooftop TMDI, provides a similar level of response reduction as STMD, while being more reliable due to redundancy. In addition, the ICDTMD exhibits a similar level of response reduction as the DTMD with significantly smaller optimized spring stiffness.

Robust Tuning of Multiresonant Current Controllers for Grid-Tied Converters and Erroneous Use of the Naslin Polynomial Method

The majority of the paper is devoted to debunking flawed methods proposed in the literature within the context of multiresonant control systems for grid-connected converters and then a novel approach is proposed. Several flawed approaches to grid tied power electronic converters (rectifiers and shunt active power filters) are discussed to initiate a critical discussion regarding some tuning methods reported in the topical literature. In this paper we analytically show that some of them are even erroneous from the control theory point of view or are based on mathematically erroneous derivations. That is especially prevalent for papers in which the Naslin polynomial method is employed to tune proportional-multiresonant controllers. Some authors recognize that the resulting control systems are not robust, i.e. not practical, and use obtained gains only as a starting point for further tuning using the trial and error method or evolutionary global optimization. Robustness of such systems is often not guaranteed and if it occurs at all, it is by chance or thanks to a combination of luck and expert knowledge of the engineer (not by deliberate design). Therefore, the absence of such a guarantee motivates the search for better ways to tune such controllers. As a result, a practical robust controller tuning method for multiresonant grid current controllers is proposed and verified experimentally. We are convinced that one of the solutions can be based on the disk margin stability analysis and a global search algorithm. The required robustness is expressed as stability margins. The design procedure is very user-friendly and the disk size is the key design parameter to be selected by an engineer. The practicality of the tuning method is demonstrated empirically in a 10 kVA physical grid-tied converter.

Sparsest Univariate Learning Models Under Lipschitz Constraint

Beside the minimizationof the prediction error, two of the most desirable properties of a regression scheme are <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">stability</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">interpretability</i> . Driven by these principles, we propose continuous-domain formulations for one-dimensional regression problems. In our first approach, we use the Lipschitz constant as a regularizer, which results in an implicit tuning of the overall robustness of the learned mapping. In our second approach, we control the Lipschitz constant explicitly using a user-defined upper-bound and make use of a sparsity-promoting regularizer to favor simpler (and, hence, more interpretable) solutions. The theoretical study of the latter formulation is motivated in part by its equivalence, which we prove, with the training of a Lipschitz-constrained two-layer univariate neural network with rectified linear unit (ReLU) activations and weight decay. By proving representer theorems, we show that both problems admit global minimizers that are continuous and piecewise-linear (CPWL) functions. Moreover, we propose efficient algorithms that find the sparsest solution of each problem: the CPWL mapping with the least number of linear regions. Finally, we illustrate numerically the outcome of our formulations.

Adaptive FIT-SMC Approach for an Anthropomorphic Manipulator With Robust Exact Differentiator and Neural Network-Based Friction Compensation

In robotic manipulators, feedback control of nonlinear systems with fast finite-time convergence is desirable. However, because of the parametric and model uncertainties, the robust control and tuning of the robotic manipulators pose many challenges related to the trajectory tracking of the robotic system. This research proposes a state-of-the-art control algorithm, which is the combination of fast integral terminal sliding mode control (FIT-SMC), robust exact differentiator (RED) observer, and feedforward neural network (FFNN) based estimator. Firstly, the dynamic model of the robotic manipulator is established for the n-degrees of freedom (DoFs) system by taking into account the dynamic LuGre friction model. Then, a FIT-SMC with friction compensation-based nonlinear control has been proposed for the robotic manipulator. In addition, a RED observer is developed to get the estimates of robotic manipulator joints’ velocities. Since the dynamic friction state of the LuGre friction model is unmeasurable, FFNN is established for training and estimating the friction torque. The Lyapunov method is presented to demonstrate the finite-time sliding mode enforcement and state convergence for a robotic manipulator. The proposed control approach has been simulated in the MATLAB/Simulink environment and compared with the system with no observer to characterize the control performance. Simulation results obtained with the proposed control strategy affirm its effectiveness for a multi-DoF robotic system with model-based friction compensation having an overshoot and a settling time less than 1.5% and 0.2950 seconds, respectively, for all the joints of the robotic manipulator.

Robust Parameter Tuning of Antenna Structures by Means of Design Specification Adaptation

Parameter tuning through numerical optimization has become instrumental in the design of high-performance antenna systems. Yet, practical optimization faces several major challenges, including high cost of massive evaluations of antenna characteristics, normally involving full-wave electromagnetic (EM) analysis, large numbers of adjustable variables, the shortage of reasonable initial solutions in the case of topologically complex structures, and multimodality of the objective function landscapes. The last two reasons foster the use of global search routines, which might be prohibitively expensive. This article proposes a novel procedure for efficient and reliable antenna optimization by means of design specification adaptation. Using appropriate selection criteria and predictions from a linear expansion model, the ability of improving the current design is verified, and—in the case of a negative outcome—performance requirements are temporarily relaxed. The adjustment is carried out prior to each iteration of the optimization process with the specifications eventually converging to their original levels. The algorithm is intended to improve the efficacy of local optimizers if the target operating frequencies of the structure at hand are severely misaligned with respect to those at the available initial design. This eliminates the need for global routines and greatly enhances the robustness of the search process, as demonstrated using several antenna structures optimized under different scenarios.

A comprehensive review of standards and best practices for utility grid integration with electric vehicle charging stations

AbstractThe present work presents a comprehensive state‐of‐the‐art bibliographical review of standards related to utility grid integration and best practices of the electric vehicle (EV) charging stations. The presence of a robust tuning method is essential for successful utility grid integration with the charging stations. The lack of system standardization may hamper the EV uptake as well as successful utility grid integration with the charging stations. The distributed energy resources (DER) and vehicle to grid (V2G) are going to play a vital role in the power system operation and control. The applicability of criterions within the utility grid integration with charging station area is important to the grid operators, charging service providers, manufacturers, fleet operators, and so forth to ensure safety, dependability, and interoperability. Hence, this work tries to deliver a comprehensive and systematic review of standards and best practices for utility grid interaction with charging stations. It will help the specialists of power as well as transport sectors to track down every one of the norms and best practices, which are accessible at one stage to compare different guidelines.This article is categorized under: Energy Infrastructure &gt; Systems and Infrastructure Wind Power &gt; Systems and Infrastructure Fossil Fuels &gt; Economics and Policy Fossil Fuels &gt; Systems and Infrastructure

Tunable magnetic skyrmions in ferrimagnetic Mn4N

Thin films of ferrimagnetic Mn4N are candidate materials to host magnetic skyrmions that have demonstrated thermal stability up to 450 °C. However, there are no experimental reports observing skyrmions in this system. Here, we discuss the results of sputter grown 15–17 nm Mn4N thin films on the MgO substrate capped with Pt1−xCux layers. Vibrating sample magnetometry measurement of out-of-plane hysteresis loops confirmed that magnetic properties are insensitive to the cap layer composition. Imaging based on magnetic force microscopy measurements observed 300–50 nm sized skyrmions, as the Cu concentration was increased from x = 0–0.9. We performed density functional theory calculations and found that the interfacial Dzyaloshinskii–Moriya interactions (iDMI) follow a trend: Mn4N/MgO(001) &amp;lt; Cu/Mn4N(001) &amp;lt; Pt/Mn4N(001). We infer from these calculations that x in the Pt1−xCux capping layer can serve as a robust tuning knob to tailor the iDMI and control the skyrmion size. This work provides guidance to achieve smaller Néel-type skyrmions in Mn4N thin films, which is an important step forward for building thermally stable skyrmionic devices.

A PID tuning approach to find the optimal compromise among robustness, performance and control effort: implementation in a free software tool

A PID tuning approach is presented to achieve an optimal compromise among robustness, performance and control effort in terms of measurement noise amplification. The tuning strategy is based on the concept of fixed robustness tuning line, in which the derivative filter parameter is a key point in finding the compromise between control effort due to noise and performance. The approach allows to find the controller that optimises the performance while fulfils a prescribed robustness constraint and a permitted control effort due to noise. The tuning procedure has been implemented in a software tool that can be freely downloaded from https://sites.google.com/a/uji.es/freepidtools/. The robustness can be defined in three different ways, involving the phase margin, the gain margin and the sensitivity peak. The performance can be defined in several ways, including the IAE, ITAE, ISE and other. Some examples show the validity of the approach and the use of the tool.

Robustness and sensitivity metrics-based tuning of the augmented Kalman filter for single-channel speech enhancement

The inaccurate estimates of the speech and noise linear prediction coefficients (LPCs) introduce bias in augmented Kalman filter (AKF) gain, which impacts the quality and intelligibility of enhanced speech. Although current tuning methods offset the bias in AKF gain, particularly in colored noise conditions, they do not adequately address nonstationary noise conditions. This paper introduces a new tuning algorithm of the AKF gain for speech enhancement in real-life noise conditions. Due to this purpose, a speech presence probability (SPP) method first estimates the noise power spectral density (PSD) from each noisy speech frame to compute the noise LPC parameters. A whitening filter is constructed with the noise LPCs to pre-whiten each noisy speech frame prior to computing the speech LPC parameters. The AKF is then constructed with the estimated speech and noise LPC parameters. To achieve better noise reduction, the robustness metric is employed to dynamically offset the bias in AKF gain during speech absence of the noisy speech to that of the sensitivity metric during speech presence. The speech activity is obtained through adopting the speech and noise production model parameters. It is shown that the reduced-biased AKF gain achieved by the proposed tuning algorithm addresses speech enhancement in real-life noise conditions. Objective and subjective scores on the NOIZEUS corpus demonstrate that the proposed method produces enhanced speech with higher quality and intelligibility than the competing methods in real-life noise conditions for a wide range of signal-to-noise ratio (SNR) levels.

Control of high-order processes: near-optimal robust tuning of PID regulators for repeated-pole plus dead-time models

This paper aims to present an analysis of the performance and robustness characteristics of repeated-pole plus dead-time (RPPDT) model PID (Proportional Integral Derivative) regulatory control system. Their performances are evaluated with the integrated absolute error index (IAE) and their robustnesses with the maximum sensitivity ( M S ). Normalised controlled process models and controllers are used for control system design. Optimal – robustness free – and optimal robust PID regulators parameters are found. A Near-Optimal Robust tuning rule (NORpid) for PID controllers based on high-order RPPDT models is obtained and applied to the control system of an eighth-order process. Examples include the control of the superheated steam temperature system (SHSTS) of a 500-MW boiler.

Robust Iterative Feedback Tuning Control of a Permanent Magnet Synchronous Motor with Repetitive Constraints: A Udwadia–Kalaba Approach

Robust Iterative Feedback Tuning Control of a Permanent Magnet Synchronous Motor with Repetitive Constraints: A Udwadia–Kalaba Approach

Robustness and Sensitivity Tuning of the Kalman Filter for Speech Enhancement

Inaccurate estimates of the linear prediction coefficient (LPC) and noise variance introduce bias in Kalman filter (KF) gain and degrade speech enhancement performance. The existing methods propose a tuning of the biased Kalman gain, particularly in stationary noise conditions. This paper introduces a tuning of the KF gain for speech enhancement in real-life noise conditions. First, we estimate noise from each noisy speech frame using a speech presence probability (SPP) method to compute the noise variance. Then, we construct a whitening filter (with its coefficients computed from the estimated noise) to pre-whiten each noisy speech frame prior to computing the speech LPC parameters. We then construct the KF with the estimated parameters, where the robustness metric offsets the bias in KF gain during speech absence of noisy speech to that of the sensitivity metric during speech presence to achieve better noise reduction. The noise variance and the speech model parameters are adopted as a speech activity detector. The reduced-biased Kalman gain enables the KF to minimize the noise effect significantly, yielding the enhanced speech. Objective and subjective scores on the NOIZEUS corpus demonstrate that the enhanced speech produced by the proposed method exhibits higher quality and intelligibility than some benchmark methods.

Offline robust tuning of the motion control for omnidirectional mobile robots

In recent years, mobile robots have been helpful systems to perform a wide variety of complex tasks in daily life applications from industry, academy, and home. These robots carry out mobility on flat terrains, mainly in narrow spaces that are difficult to access or dangerous for humans. Therefore, increasing the efficiency of their movements through control technologies has become a topic of great interest for researchers. Among controllers, the linear ones are widely used to improve the efficiency of mobile robots because of their simplicity, reliability, and practicality, notwithstanding advanced control strategies. A well-tuned linear controller can show outstanding performances in controlled environments where the modeled and simulated conditions used for its adjustment are not too far from reality. However, actual operating environments are subject to uncertainties and disturbances that can hardly be accounted for during the controller tuning process. The above compromises the performance of the mobile robot in practice, and finding the appropriate controller parameters that enhance robustness becomes a crucial task. Therefore, this work presents a robust tuning approach for the controller of an omnidirectional mobile robot based on the solution of a nonlinear dynamic optimization problem through meta-heuristics. Robustness is incorporated in the optimization problem by minimizing the sensitivity to the control performance indexes. Simultaneously, this is included through dynamic and stochastic variations in the meta-heuristic optimizer hyperparameters. A comparative statistical analysis is performed using robust and non-robust tuning approaches. Based on simulated and experimental tests, the proposed robust approach shows notable performance improvements regarding the non-robust one while minimizing operation errors in the presence of different uncertainty magnitudes.

A probabilistic framework to achieve robust non‐fragile tuning methods: PD control of IPD‐modeled processes

AbstractWe introduce a novel probabilistic framework to achieve robust non‐fragile tuning methods in control of processes with parametric uncertainties. We consider probability distributions to model the process parameters' uncertainties. First, we propose the tuning framework in a general setting. Then, as an illustration, we apply it to PD control of IPD‐modeled processes. It is noteworthy that the proposed tuning method is robust against the considered parametric uncertainties. Also, to empower the proposed robust tuning method in the viewpoint of non‐fragility, we utilize a centroid approach. Selecting the form of the probabilistic framework, we empirically observe some of the popular tuning methods are special cases of the proposed novel framework. Moreover, we theoretically/empirically make a comparison among the tuning methods in the literature based on non‐fragility and robustness via such a probabilistic framework.

An Improved Approach for Robust MPC Tuning Based on Machine Learning

A robust tuning method based on an artificial neural network for model predictive control (MPC) of industrial systems with parametric uncertainties is put forward in this work. Firstly, an efficient approach to characterize the mapping relationship between the controller parameters and the robust performance indices is established. As there are normally multiple conflicted robust performance indices to be considered in MPC tuning, the neural network is further used to fuse the indices to produce a simple label representing the acceptable level of the robust performance. Finally, an automated algorithm is proposed to tune the MPC parameters for the considered uncertain system to achieve the desired robust performance. In addition, the regulation of the pH value of the sewage treatment system is used to verify the effectiveness of the robust tuning algorithm which is described in this paper.

A multi-objective optimization model for robust tuning of wide-area PSSs for enhancement and control of power system angular stability

A multi-objective optimization model for robust tuning of wide-area power system stabilizers (WAPSSs) aiming the improvement and control of angular stability of electric power system (EPS) is presented in this paper. In this work the WAPSSs are referred to power system stabilizers (PSSs) of fixed-structure with insertion of remote signals, from phasor measurement units (PMUs), in the control loops, that are used in a wide-area control system (WACS). The proposed model is composed of objective functions to maximize both the small-disturbance angular stability (SDAS) through the damping ratio and the large-disturbance angular stability (LDAS) using an index derived from transient energy function (TEF). Moreover, three objectives functions that contribute to the previous ones are taken into account: maximization of power system robustness for several operating conditions, minimization of eigenvalues sensitivity to small disturbances, and minimization of eigenvalues sensitivity with respect to time delay variations of remote signals. Swarm intelligence-based techniques are used to solve the multi-objective optimization model. Simulation results, with the benchmark interconnected system New England test system – New York power system (NETS-NYPS) of 16-generator and 68-bus, corroborate the applicability of the proposal.