Greece

Real-world time series often present missing values due to sensor malfunctions or human errors. Traditionally, missing values are simply omitted or reconstructed through imputation or interpolation methods. Omitting missing values may cause temporal discontinuity. Reconstruction methods, on the other hand, alter in some way the original time series. In this paper, we consider an application in the field of meteorological variables that exploits end-to-end machine learning. The idea is to entrust the task of dealing with missing values to a suitably trained recurrent neural network that completely by-passes the phase of reconstruction of missing values. A difficult case of reproduction of a rainfall field from five rain gauges in Northern Italy is used as an example, and the results are compared to those computed by more traditional methods. The proposed methodology is general-purpose and can be easily applied to every kind of spatial time series prediction problem, quite common in many environmental studies.

EANN 2020

Reconstructing Environmental Variables with Missing Field Data via End-to-End Machine Learning

rainfall field

lstm network

time series

The purpose of the AAAI conference is to promote research in artificial intelligence (AI) and scientific exchange among AI researchers, practitioners, scientists, and engineers in affiliated disciplines. 

AAAI 2021

technical paper

21st edition of EANN (International Conference on Engineering Applications of Neural Networks) should be held in Halkidiki, Greece. Due to current Covid-19 situation the conference will be held virtually and Underline will host a repository of video talks and preserve them for the future. Videos will be released on Underline one day after the conference. 

Transition towards Industry 4.0 relies heavily on manufacturing digitalisation. Digital twin plays a significant role among the pool of relevant technologies as a powerful tool that is expected provide digital access to detailed real-time monitoring of the physical processes and enable significant optimisation due to utilisation of big data acquired from them. Over the past years a significant number of works produced conceptual frameworks of digital twins and discussed their requirements and benefits. The research literature demonstrates application examples and proofs of concepts, although the content is less rich.
This paper presents a generative model based on generative adversarial networks (GAN) for machining vibration data, discusses its performance and analyses the drawbacks. The proposed model includes process parameter inputs used to condition the features of generated signals. The control over the generator and a neural network architecture utilising techniques from style-transfer research provide the means to analyse the signal building blocks learned by the model and explore their relationship. The quality of the learned process representation is demonstrated using a dataset obtained from a machining time-domain simulation. The novel results constitute a critical component of a machining digital twin and open new research directions towards development of comprehensive manufacturing digital twins.

Towards a digital twin with generative adversarial network modelling of machining vibration

This paper introduces a semantic-segmentation guided image recoloring approach of digitized art paintings to enhance the color perception of color- blind people that suffer from protanopia and deuteranopia. Semantic segmentation using trans-fer learning between natural images and art paintings is applied to extract annotat-ed color information. By using a standard technique, the annotated colors are transformed to simulate the effects of protanopia and deuteranopia. Then, a spe-cialized objective function is minimized to recolor only the colors that are signifi-cantly different from the respective simulated ones, because these colors are per-ceived as confusing by the color blind. The effectiveness of the proposed method is demonstrated through its comparison with other algorithms in several experi-mental cases.

Image Recoloring of Art Paintings for the Color Blind Guided by Semantic Segmentation

We propose a novel monetary policy strategy in an attempt to provide an auxiliary tool to central banks, whose main predictive models are still from the Dynamic Stochastic General Equilibrium (DSGE) family, which has some flaws. We derive an objective function from three empirical relationships that have long been established in the economic literature and we seek to minimise the value of this function by choosing the interest rate via a genetic algorithm. Since the function is forward looking, we use a neural network to predict values of unemployment and inflation. Using data from Brazil, simulation results suggest that had the Brazilian central bank applied our strategy, and all other economic conditions remained equal, inflation could have been lower for 62.48% of the time. Predicted unemployment, however, was lower only for 39.69% of covered periods, as it faces a trade-off with inflation.

A Monetary Policy Strategy Based on Genetic Algorithms and Neural Networks

Texture offers an effective characterization of image shape and orientation. Thus, a predominant task is to detect and extract texture features that discriminate accurately images within different semantic classes. The challenge resides in making these features invariant to several changes, such as affine transformation and viewpoint change, in order to ensure their robustness. Besides, the training phase requires a large number of images. To deal with these issues, Genetic Programming (GP) is adopted in this work with the intention of classifying precisely texture images using some training images per class. In fact, in order to automatically generate a descriptor that is invariant to illumination, rotation and scale; the proposed method combines GP with the scale extraction technique involved by SIFT. The performance of the proposed method is validated on five challenging datasets of non-scaled as well as scaled texture images. Results show that the method is robust not only to scale but also to rotation, while achieving significant performance compared to the state of the art methods.

A Genetic Programming Method for Scale-Invariant Texture Classification

In this paper, a method to access the location of a steel strip in the rolling process was developed. The method consists of a hybrid system composed of a CNN-based semantic segmentation followed by morphological operation and outlier removal. The proposed method was capable of estimating the position of the strip with high precision and low computational burden, making it suitable for the application. The implementation of automatic estimation for the steel strip positioning, replacing the current human operation, can yield substantial costs saving. Future work will be carried out for the and integration of automatic control in the process.

Semantic Segmentation Based on Convolution Neural Network for Steel Strip Position Estimation

When convolutional neural networks are applied to imagesegmentation results depend greatly on the data sets used to train thenetworks. Cloud providers support multi GPU and TPU virtual machinesmaking the idea of cloud-based segmentation as service attractive. In thispaper we study the problem of building a segmentation service, whereimages would come from different acquisition instruments, by training ageneralized U-Net with images from a single or several datasets. We alsostudy the possibility of training with a single instrument and performquick retrains when more data is available. As our example we performsegmentation of Optic Disc in fundus images which is useful for glau-coma diagnosis. We use two publicly available data sets (RIM-One V3,DRISHTI) for individual, mixed or incremental training. We show thatmultidataset or incremental training can produce results that are simi-lar to those published by researchers who use the same dataset for bothtraining and validation

Multidataset incremental Training for OpticDisc Segmentation

Performance evaluation an image data set on a CNN from scratch and on a pre-trained network.

Factors affecting accuracy of Convolutional Neural Network using VGG-16

Hyperspectral satellite imagery is now widely being used for accurate disaster prediction and terrain feature classification. However, in such classification tasks, most of the present approaches use only the spectral information contained in the images. Therefore, in this paper, we present a novel framework that takes into account both the spectral and spatial information contained in the data for land cover classification. For this purpose, we use the Gaussian Maximum Likelihood (GML) and Convolutional Neural Network (CNN) methods for the pixel-wise spectral classification and then, using segmentation maps generated by the Watershed algorithm, we incorporate the spatial contextual information into our model with a modified majority vote technique. The experimental analyses on two benchmark datasets demonstrate that our proposed methodology performs better than the earlier approaches by achieving an accuracy of 99.52% and 98.31% on the Pavia University and the Indian Pines datasets respectively. Additionally, our GML based approach, a non-deep learning algorithm, shows comparable performance to the state-of-the-art deep learning techniques, which indicates the importance of the proposed approach for performing a computationally efficient classification of hyperspectral imagery.

A Novel Spatial-Spectral Framework for the Classification of Hyperspectral Satellite Imagery

Exercise Recognition is relevant in many high impact domains, from healthcare to recreational activities to sports sciences, which faces many challenges when deployed in the real world. 
For instance, typical lab performances of Machine Learning~(ML) models, are hard to replicate, due to differences in personal nuances, traits and ambulatory rhythms.
Thus effective transferability of a trained model depends on its ability to adapt and personalise to a new user or a user group.
We look at person-agnostic and person-aware methods of evaluation to identify transferability of generic deep learning models and introduce a personalised algorithm to improve performance in different user groups. 
Our findings show that Exercise Recognition, when compared to results with other HAR tasks, to be a far more challenging personalisation problem.

Transferability of Personalised Exercise Recognition Models

In this presentation I introduce my paper Instance Weighted Clustering: Local Outlier Factor and K-Means. I briefly explain the area, inspirations, methodology and findings. Instance Weighted clustering is easy to understand and produces robust clustering results, the technique can be applied to variety of k-means like algorithms.

Downloads

Next from EANN 2020

Towards a digital twin with generative adversarial network modelling of machining vibration

Similar lecture

Generative Semi-Supervised Learning for Multivariate Time Series Imputation

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from EANN 2020

Towards a digital twin with generative adversarial network modelling of machining vibration

Similar lecture

Generative Semi-Supervised Learning for Multivariate Time Series Imputation

Downloads