The Effect of Chemical Representation on Active Machine Learning Towards Closed-Loop Optimization

Multivariate chemical reaction optimization involving catalytic systems is a non-trivial task due to the high number of tuneable parameters and discrete choices. Closed-loop optimization featuring active Machine Learning (ML) represents a powerful strategy for automating reaction optimization. However, the translation of chemical reaction conditions into a machine-readable format comes with the challenge of finding highly informative features which accurately capture the factors for reaction success and allow the model to learn efficiently. Herein, we compare the efficacy of different calculated chemical descriptors for a high throughput generated dataset to determine the impact on a supervised ML model when predicting reaction yield. Then, the effect of featurization and size of the initial dataset within a closed-loop reaction optimization was examined. Finally, the balance between descriptor complexity and dataset size was considered. Ultimately, tailored descriptors did not outperform simple generic representations, however, a larger initial dataset accelerated reaction optimization.

Improving Walking Economy With an Ankle Exoskeleton Prior to Human-in-the-Loop Optimization

Lower limb robotic exoskeletons have shown the capability to enhance human locomotion for healthy individuals or to assist motion rehabilitation and daily activities for patients. Recent advances in human-in-the-loop optimization that allowed for assistance customization have demonstrated great potential for performance improvement of exoskeletons. In the optimization process, subjects need to experience multiple types of assistance patterns, thus, leading to a long evaluation time. Besides, some patterns may be uncomfortable for the wearers, thereby resulting in unpleasant optimization experiences and inaccurate outcomes. In this study, we investigated the effectiveness of a series of ankle exoskeleton assistance patterns on improving walking economy prior to optimization. We conducted experiments to systematically evaluate the wearers' biomechanical and physiological responses to different assistance patterns on a lightweight cable-driven ankle exoskeleton during walking. We designed nine patterns in the optimization parameters range which varied peak torque magnitude and peak torque timing independently. Results showed that metabolic cost of walking was reduced by 17.1 ± 7.6% under one assistance pattern. Meanwhile, soleus (SOL) muscle activity was reduced by 40.9 ± 19.8% with that pattern. Exoskeleton assistance changed maximum ankle dorsiflexion and plantarflexion angle and reduced biological ankle moment. Assistance pattern with 48% peak torque timing and 0.75 N·m·kg−1 peak torque magnitude was effective in improving walking economy and can be selected as an initial pattern in the optimization procedure. Our results provided a preliminary understanding of how humans respond to different assistances and can be used to guide the initial assistance pattern selection in the optimization.

Reliability-based topology optimization of piezoelectric smart structures with voltage uncertainty

Currently, most of the piezoelectric structures are designed under deterministic conditions, where the influence of uncertain factors on the output motion accuracy is ignored. In this work, a probabilistic reliability-based topology optimization method for piezoelectric structure is proposed to deal with the working voltage uncertainty. A nested double-loop optimization algorithm of minimizing the total volume while satisfying the reliability requirement of the displacement performance is established, where the PEMAP-P (piezoelectric material with penalization and polarization) model is used for parameterization of stiffness matrix, piezoelectric coupling matrix, and polarization direction. This strategy consists of an inner loop for reliability analysis and an outer loop for topology optimization. The reliability index approach based on most probable point (MPP) is used for realizing the evaluation of reliability constraint in reliability analysis. The sensitivities of reliability constraint with respect to the random variables and design variables are detailed using the adjoint variable method. Typical examples are performed to illustrate the effectiveness of the proposed RBTO method. A comparison of the optimization results for different reliability indexes, standard deviations of the voltage, spring stiffnesses, and displacement limits are conducted, as well as the deterministic topology optimization results.

Optimization of the Convolution Operation to Accelerate Deep Neural Networks in FPGA

In recent years, machine learning algorithms related to images have been widely utilized by Convolution Neural Networks (CNN), and it has a high accuracy for recognition of an image. As CNN contains large number of computations, hardware accelerator like Field Programmable Gate Array is employed. Quite 90 % of operations during a CNN involves convolution. The objective of this work is to scale back the computation time to increase the peak, width and the pixel intensity levels in the input image. The execution time of a image processing program is mostly spent on loops. Loop optimization is a process of accelerating speed and reducing the overheads related to loops. It plays a crucial role in improving performance and making effective use of multiprocessing capabilities. Loop unrolling is one of the loop optimization techniques. In our work CNN with four levels of loop unrolling is used. Due to this delay is reduced compared with conventional Xilinix. With the assistance of strides and padding the 40 % of computation time has been reduced and is verified in MATLAB.

Joint 3-D Positioning and Power Allocation for UAV Relay Aided by Geographic Information

<div>In this paper, we study to employ geographic information to address the blockage problem of air-to-ground links between UAV and terrestrial nodes. In particular, a UAV relay is deployed to establish communication links from a ground base station to multiple ground users. To improve communication capacity, we fifirst model the blockage effect caused by buildings according to the three-dimensional (3-D) geographic information. Then, an optimization problem is formulated to maximize the minimum capacity among users by jointly optimizing the 3-D position and power allocation of the UAV relay, under the constraints of link capacity, maximum transmit power, and blockage. To solve this complex non-convex problem, a two-loop optimization framework is developed based on Lagrangian relaxation. The outer-loop aims to obtain proper Lagrangian multipliers to ensure the solution of the Lagrangian problem converge to the tightest upper bound on the original problem. The inner-loop solves the Lagrangian problem by applying the block coordinate descent (BCD) and successive convex approximation (SCA) techniques, where UAV 3-D positioning and power allocation are alternately optimized in each iteration. Simulation results confifirm that the proposed solution signifificantly outperforms two benchmark schemes and achieves a performance close to the upper bound on the UAV relay system.</div>

Application of a Combination of Machine Learning and Reservoir Physics to Increase Production in a Mature Waterflood Field in the San Jorge Basin in Argentina

This paper presents the process and results of the application of Data Physics to optimize production of a mature field in the Gulf of San Jorge Basin in Argentina. Data Physics is a novel technology that blends the reservoir physics (black oil) used in traditional numerical simulation with machine learning and advanced optimization techniques. Data Physics was described in detail in a prior paper (Sarma, et al SPE-185507-MS) as a physics-based modeling approach augmented by machine learning. In essence, historical production and injection data are assimilated using an Ensemble Kalman Filter (EnKF) to infer the petrophysical parameters and create a predictive model of the field. This model is then used with Evolutionary Algorithms (EA) to find the pareto front for multiple optimization objectives like production, injection and NPV. Ultimately, the main objective of Data Physics is to enable Closed Loop Optimization. The technology was applied on a small section of a very large field in the Gulf of San Jorge comprised of 134 wells including 83 active producers and 27 active water injectors; up to 12 mandrels per well are used to provide with selective injection, while production is carried out in a comingled manner. Production zonal allocation is calculated using an in-house process based on swabbing tests and recovery factors and is used as input to the Data Physics application, while injection allocation is based on tracer logs performed in each injection well twice a year. This paper describes the modeling and optimization phases as well as the implementation in the field and the results obtained after performing two close loop optimization cycles. The initial model was developed between October and December 2018 and initial field implementation took place between January to March 2019. A second optimization cycle was then executed in January 2020 and results observed for several months.

Human-in-the-loop optimization of visual prosthetic stimulation

Retinal prostheses are a promising strategy to restore sight to patients with retinal degenerative diseases. These devices compensate for the loss of photoreceptors by electrically stimulating neurons in the retina. Currently, the visual function that can be recovered with such devices is very limited. This is due, in part, to current spread, unintended axonal activation, and the limited resolution of existing devices. Here we show, using a recent model of prosthetic vision, that optimizing how visual stimuli are encoded by the device can help overcome some of these limitations, leading to dramatic improvements in visual perception. We propose a strategy to do this in practice, using patients' feedback in a visual task. The main challenge of our approach comes from the fact that, typically, one only has access to a limited number of noisy responses from patients. We propose two ways to deal with this: first, we use a model of prosthetic vision to constrain and simplify the optimisation; second, we use preferential Bayesian optimisation to efficiently learn the encoder using minimal trials. As a proof-of concept, we presented healthy subjects with visual stimuli generated by a recent model of prosthetic vision, to replicate the perceptual experience of patients fitted with an implant. Our optimisation procedure led to significant and robust improvements in perceived image quality, that transferred to increased performance in other tasks. Importantly, our strategy is agnostic to the type of prosthesis and thus could readily be implemented in existing implants.