scholarly journals On the utility of dreaming: A general model for how learning in artificial agents can benefit from data hallucination

2020 ◽  
pp. 105971231989648 ◽  
Author(s):  
David Windridge ◽  
Henrik Svensson ◽  
Serge Thill
Keyword(s):  
Machine Learning ◽  
Simulated Data ◽  
Training Data ◽  
Successful Implementation ◽  
Artificial Agents ◽  
Learning Context ◽  
Training Set ◽  
Convergence Point ◽  
And Training ◽  
General Method

We consider the benefits of dream mechanisms – that is, the ability to simulate new experiences based on past ones – in a machine learning context. Specifically, we are interested in learning for artificial agents that act in the world, and operationalize “dreaming” as a mechanism by which such an agent can use its own model of the learning environment to generate new hypotheses and training data. We first show that it is not necessarily a given that such a data-hallucination process is useful, since it can easily lead to a training set dominated by spurious imagined data until an ill-defined convergence point is reached. We then analyse a notably successful implementation of a machine learning-based dreaming mechanism by Ha and Schmidhuber (Ha, D., & Schmidhuber, J. (2018). World models. arXiv e-prints, arXiv:1803.10122). On that basis, we then develop a general framework by which an agent can generate simulated data to learn from in a manner that is beneficial to the agent. This, we argue, then forms a general method for an operationalized dream-like mechanism. We finish by demonstrating the general conditions under which such mechanisms can be useful in machine learning, wherein the implicit simulator inference and extrapolation involved in dreaming act without reinforcing inference error even when inference is incomplete.

scholarly journals Effects of Training Set Size on Supervised Machine-Learning Land-Cover Classification of Large-Area High-Resolution Remotely Sensed Data

2021 ◽  
Vol 13 (3) ◽  
pp. 368
Author(s):  
Christopher A. Ramezan ◽  
Timothy A. Warner ◽  
Aaron E. Maxwell ◽  
Bradley S. Price
Keyword(s):  
Machine Learning ◽  
Sample Size ◽  
Remotely Sensed ◽  
Training Data ◽  
Supervised Machine Learning ◽  
Sample Sizes ◽  
Remotely Sensed Data ◽  
Large Area ◽  
Training Set ◽  
Set Size

The size of the training data set is a major determinant of classification accuracy. Nevertheless, the collection of a large training data set for supervised classifiers can be a challenge, especially for studies covering a large area, which may be typical of many real-world applied projects. This work investigates how variations in training set size, ranging from a large sample size (n = 10,000) to a very small sample size (n = 40), affect the performance of six supervised machine-learning algorithms applied to classify large-area high-spatial-resolution (HR) (1–5 m) remotely sensed data within the context of a geographic object-based image analysis (GEOBIA) approach. GEOBIA, in which adjacent similar pixels are grouped into image-objects that form the unit of the classification, offers the potential benefit of allowing multiple additional variables, such as measures of object geometry and texture, thus increasing the dimensionality of the classification input data. The six supervised machine-learning algorithms are support vector machines (SVM), random forests (RF), k-nearest neighbors (k-NN), single-layer perceptron neural networks (NEU), learning vector quantization (LVQ), and gradient-boosted trees (GBM). RF, the algorithm with the highest overall accuracy, was notable for its negligible decrease in overall accuracy, 1.0%, when training sample size decreased from 10,000 to 315 samples. GBM provided similar overall accuracy to RF; however, the algorithm was very expensive in terms of training time and computational resources, especially with large training sets. In contrast to RF and GBM, NEU, and SVM were particularly sensitive to decreasing sample size, with NEU classifications generally producing overall accuracies that were on average slightly higher than SVM classifications for larger sample sizes, but lower than SVM for the smallest sample sizes. NEU however required a longer processing time. The k-NN classifier saw less of a drop in overall accuracy than NEU and SVM as training set size decreased; however, the overall accuracies of k-NN were typically less than RF, NEU, and SVM classifiers. LVQ generally had the lowest overall accuracy of all six methods, but was relatively insensitive to sample size, down to the smallest sample sizes. Overall, due to its relatively high accuracy with small training sample sets, and minimal variations in overall accuracy between very large and small sample sets, as well as relatively short processing time, RF was a good classifier for large-area land-cover classifications of HR remotely sensed data, especially when training data are scarce. However, as performance of different supervised classifiers varies in response to training set size, investigating multiple classification algorithms is recommended to achieve optimal accuracy for a project.

scholarly journals Understanding the Relationship between Interactions and Outcomes in Human-in-the-Loop Machine Learning

2021 ◽  
Author(s):  
Yuchen Cui ◽  
Pallavi Koppol ◽  
Henny Admoni ◽  
Scott Niekum ◽  
Reid Simmons ◽  
...  
Keyword(s):  
Machine Learning ◽  
Human Performance ◽  
Autonomous Agents ◽  
Training Data ◽  
Open Problems ◽  
Human In The Loop ◽  
Efficiency And Effectiveness ◽  
Interaction Types ◽  
The Relationship ◽  
And Training

Human-in-the-loop Machine Learning (HIL-ML) is a widely adopted paradigm for instilling human knowledge in autonomous agents. Many design choices influence the efficiency and effectiveness of such interactive learning processes, particularly the interaction type through which the human teacher may provide feedback. While different interaction types (demonstrations, preferences, etc.) have been proposed and evaluated in the HIL-ML literature, there has been little discussion of how these compare or how they should be selected to best address a particular learning problem. In this survey, we propose an organizing principle for HIL-ML that provides a way to analyze the effects of interaction types on human performance and training data. We also identify open problems in understanding the effects of interaction types.

Facial Attractiveness: Beauty and the Machine

2006 ◽  
Vol 18 (1) ◽  
pp. 119-142 ◽  
Author(s):  
Yael Eisenthal ◽  
Gideon Dror ◽  
Eytan Ruppin
Keyword(s):  
Machine Learning ◽  
Facial Attractiveness ◽  
Training Data ◽  
Data Sets ◽  
Learning Context ◽  
Data Set ◽  
Attractiveness Rating ◽  
Facial Beauty ◽  
Facial Images

This work presents a novel study of the notion of facial attractiveness in a machine learning context. To this end, we collected human beauty ratings for data sets of facial images and used various techniques for learning the attractiveness of a face. The trained predictor achieves a significant correlation of 0.65 with the average human ratings. The results clearly show that facial beauty is a universal concept that a machine can learn. Analysis of the accuracy of the beauty prediction machine as a function of the size of the training data indicates that a machine producing human-like attractiveness rating could be obtained given a moderately larger data set.

scholarly journals Sample Bias Effect on Meta-Learning

2020 ◽  
Author(s):  
Mariane Reis ◽  
Ana Lorena
Keyword(s):  
Machine Learning ◽  
Random Sample ◽  
Side Effect ◽  
Random Sampling ◽  
Training Data ◽  
Learning Context ◽  
Bias Effect ◽  
Sample Bias ◽  
Meta Learning ◽  
Learning Research

Sample bias is a common issue on traditional machine learning studies but rarely considered when discussing meta-learning. It happens when the training data sample lacks or overemphasizes one or more characteristics, compared to others. Herewith, models trained on such data may become inaccurate for some instances. This work aims to analyze this issue in the meta-learning context. Indeed, in most of the meta-learning literature, a random sample of datasets is taken for building meta-models. Nonetheless, there is no discussion over a possible side-effect bias not controlled in such random sampling. This work aims to analyze these effects, in order not only to discuss their consequences, but also to start a debate over the need of their consideration in meta-learning research.

Machine Learning-Driven Real-Time Topology Optimization Under Moving Morphable Component-Based Framework

2018 ◽  
Vol 86 (1) ◽  
Author(s):  
Xin Lei ◽  
Chang Liu ◽  
Zongliang Du ◽  
Weisheng Zhang ◽  
Xu Guo
Keyword(s):  
Machine Learning ◽  
Topology Optimization ◽  
Real Time ◽  
Training Data ◽  
Design Parameters ◽  
Structural Topology ◽  
Training Set ◽  
K Nearest Neighbors ◽  
External Stimuli ◽  
External Loads

In the present work, it is intended to discuss how to achieve real-time structural topology optimization (i.e., obtaining the optimized distribution of a certain amount of material in a prescribed design domain almost instantaneously once the objective/constraint functions and external stimuli/boundary conditions are specified), an ultimate dream pursued by engineers in various disciplines, using machine learning (ML) techniques. To this end, the so-called moving morphable component (MMC)-based explicit framework for topology optimization is adopted for generating training set and supported vector regression (SVR) as well as K-nearest-neighbors (KNN) ML models are employed to establish the mapping between the design parameters characterizing the layout/topology of an optimized structure and the external load. Compared with existing approaches, the proposed approach can not only reduce the training data and the dimension of parameter space substantially, but also has the potential of establishing engineering intuitions on optimized structures corresponding to various external loads through the learning process. Numerical examples provided demonstrate the effectiveness and advantages of the proposed approach.

Beautiful Fractals as a Crystal Ball for Financial Markets? - Investment Decision Support System Based on Image Recognition Using Artificial Intelligence

2020 ◽  
Vol 14 (2) ◽  
pp. 27-44
Author(s):  
Benjamin M. Abdel-Karim
Keyword(s):  
Machine Learning ◽  
Financial Markets ◽  
Image Recognition ◽  
Fractal Geometry ◽  
Stock Price ◽  
Simulated Data ◽  
Real Data ◽  
Point Of View ◽  
Training Data ◽  
Theoretical Point

The work by Mandelbrot develops a basic understanding of fractals and the artwork of Jackson Pollok to reveal the beauty fractal geometry. The pattern of recurring structures is also reflected in share prices. Mandelbrot himself speaks of the fractal heart of the financial markets. Previous research has shown the potential of image recognition. This paper presents the possibility of using the structure recognition capability of modern machine learning methods to make forecasts based on fractal course information. We generate training data from real and simulated data. These data are represented in images to train a special artificial neural network. Subsequently, real data are presented to the network for use in predicting. The results show that the forecast of time series based on stock price illustration, compared to a benchmark, delivers promising results. This paper makes two essential contributions to research. From a theoretical point of view, fractal geometry shows that it can serve as a means of legitimation for technical analysis. From a practical point of view, highly developed methods from the field of machine learning are able to recognize patterns in data through appropriate data transformation, and that models such as random walk have an informational content that can be used to train machine learning models.

scholarly journals Metadynamics sampling in atomic environment space for collecting training data for machine learning potentials

2021 ◽  
Vol 7 (1) ◽  
Author(s):  
Dongsun Yoo ◽  
Jisu Jung ◽  
Wonseok Jeong ◽  
Seungwu Han
Keyword(s):  
Machine Learning ◽  
Sampling Method ◽  
Local Environment ◽  
Training Data ◽  
Interatomic Potentials ◽  
Mathematical Form ◽  
Training Set ◽  
Large Pool ◽  
Automatic Sampling ◽  
Proper Training

AbstractThe universal mathematical form of machine-learning potentials (MLPs) shifts the core of development of interatomic potentials to collecting proper training data. Ideally, the training set should encompass diverse local atomic environments but conventional approaches are prone to sampling similar configurations repeatedly, mainly due to the Boltzmann statistics. As such, practitioners handpick a large pool of distinct configurations manually, stretching the development period significantly. To overcome this hurdle, methods are being proposed that automatically generate training data. Herein, we suggest a sampling method optimized for gathering diverse yet relevant configurations semi-automatically. This is achieved by applying the metadynamics with the descriptor for the local atomic environment as a collective variable. As a result, the simulation is automatically steered toward unvisited local environment space such that each atom experiences diverse chemical environments without redundancy. We apply the proposed metadynamics sampling to H:Pt(111), GeTe, and Si systems. Throughout these examples, a small number of metadynamics trajectories can provide reference structures necessary for training high-fidelity MLPs. By proposing a semi-automatic sampling method tuned for MLPs, the present work paves the way to wider applications of MLPs to many challenging applications.

scholarly journals Deriving Field Scale Soil Moisture from Satellite Observations and Ground Measurements in a Hilly Agricultural Region

2019 ◽  
Vol 11 (22) ◽  
pp. 2596 ◽  
Author(s):  
Luca Zappa ◽  
Matthias Forkel ◽  
Angelika Xaver ◽  
Wouter Dorigo
Keyword(s):  
Machine Learning ◽  
Soil Moisture ◽  
Soil Texture ◽  
Vegetation Cover ◽  
Learning Algorithm ◽  
Low Cost ◽  
Cost Effective ◽  
Training Data ◽  
Field Scale ◽  
Training Set

Agricultural and hydrological applications could greatly benefit from soil moisture (SM) information at sub-field resolution and (sub-) daily revisit time. However, current operational satellite missions provide soil moisture information at either lower spatial or temporal resolution. Here, we downscale coarse resolution (25–36 km) satellite SM products with quasi-daily resolution to the field scale (30 m) using the random forest (RF) machine learning algorithm. RF models are trained with remotely sensed SM and ancillary variables on soil texture, topography, and vegetation cover against SM measured in the field. The approach is developed and tested in an agricultural catchment equipped with a high-density network of low-cost SM sensors. Our results show a strong consistency between the downscaled and observed SM spatio-temporal patterns. We found that topography has higher predictive power for downscaling than soil texture, due to the hilly landscape of the study area. Furthermore, including a proxy of vegetation cover results in considerable improvements of the performance. Increasing the training set size leads to significant gain in the model skill and expanding the training set is likely to further enhance the accuracy. When only limited in-situ measurements are available as training data, increasing the number of sensor locations should be favored over expanding the duration of the measurements for improved downscaling performance. In this regard, we show the potential of low-cost sensors as a practical and cost-effective solution for gathering the necessary observations. Overall, our findings highlight the suitability of using ground measurements in conjunction with machine learning to derive high spatially resolved SM maps from coarse-scale satellite products.

Sign in / Sign up

Export Citation Format

Share Document