Perceptron Neural Nets Research Articles

Assessment of ocean color atmospheric correction methods and development of a regional ocean color operational dataset for the Baltic Sea based on Sentinel-3 OLCI

The Baltic Sea is characterized by large gradients in salinity, high concentrations of colored dissolved organic matter, and a phytoplankton phenology with two seasonal blooms. Satellite retrievals of chlorophyll-a concentration (chl-a) are hindered by the optical complexity of this basin and the reduced performance of the atmospheric correction in its highly absorbing waters. Within the development of a regional ocean color operational processing chain for the Baltic Sea based on Sentinel-3 Ocean and Land Colour Instrument (OLCI) full-resolution data, the performance of four atmospheric correction processors for the retrieval of remote-sensing reflectance (Rrs) was analyzed. Assessments based on three Aerosol Robotic Network-Ocean Color (AERONET-OC) sites and shipborne hyperspectral radiometers show that POLYMER was the best-performing processor in the visible spectral range, also providing a better spatial coverage compared with the other processors. Hence, OLCI Rrs spectra retrieved with POLYMER were chosen as input for a bio-optical ensemble scheme that computes chl-a as a weighted sum of different regional multilayer perceptron neural nets. This study also evaluated the operational Rrs and chl-a datasets for the Baltic Sea based on OC-CCI v.6. The chl-a retrievals based on OC-CCI v.6 and OLCI Rrs, assessed against in-situ chl-a measurements, yielded similar results (OC-CCI v.6: R2 = 0.11, bias = −0.22; OLCI: R2 = 0.16, bias = −0.03) using a common set of match-ups for the same period. Finally, an overall good agreement was found between chl-a retrievals from OLCI and OC-CCI v.6 although differences between Rrs were amplified in terms of chl-a estimates.

Integrating multilayer perceptron neural nets with hybrid ensemble classifiers for deforestation probability assessment in Eastern India

The rapid expansion of human settlement, agricultural land and roads because of population growth in several regions of the world has contributed to the depletion of forest land. In this study, novel ensemble intelligent approaches using bagging, dagging and rotation forest (RTF) as meta classifiers of multilayer perceptron (MLP) were used to predict spatial deforestation probability (DP) in Gumani Basin, India. The success rate and correctness of prediction of the ensemble models were compared with MLP. A total of 1000 deforested pixels and 14 deforestation determining factors (DDFs) were used. The ensemble models were trained using 70% of the deforested pixels and validated with the remaining 30%. DDFs were chosen by applying the information gain ratio and Relief-F test methods. Distance to settlement, population growth and distance to roads were the most important factors. The results of DP modelling demonstrated that nearly 16.82%–12.64% of the basin had very high DP. All four models created DP maps with reasonable prediction accuracy and goodness of fit, but the best map was produced by MLP-bagging. The accuracy of the MLP neural net model was increased 2-3% after ensemble with the hybrid meta classifiers (RTF, bagging and dagging). The proposed method could be used for deforestation prediction in other areas having similar geo-environmental conditions. Furthermore, the findings might be used as a basis for future research and could help planners in forest management.

A LEARNING PROCESS OF MULTILAYER PERCEPTRON FOR SPEECH RECOGNITION

For learning artificial systems as well as for living systems, it is generally proven that the learning performances improve with the experience. This paper seeks to analyze the learning process of an artificial system: a Multi-Layer Perceptron Neural Nets (MLP-NN) used for word recognition and dedicated for robot control. As the MLP requires references for the spoken words, we have provided these references by means of a supervised classifier based on minimizing the mean square error. We are particularly interested by estimating the minimal number of trials required to ensure the recognition of some spoken words by the MLP-NN with an acceptable predefined error. To this purpose, we have experimentally performed the learning process of the recogni- tion of some specific words. For each word, we have recorded the performance improvement with respect to the number of trials enabling to draw the learning curve. The mathematical modeling of these curves presents a bi-exponential law profile while the mathematical model- ing of human performance show generally a power law profile. The obtained results have led to a better understanding of the artificial system performance under the influence of internal

Design experiments for voice commands using neural networks

This paper presents the use of a Multi-Layer Perceptron Neural Nets (MLP-NN) for voice recognition dedicated to generating robot commands. Our main goal concerns the estimation of the minimal number of elements required for the learning process in order to ensure an acceptable success of the neural nets recognition. As the MLP requires references for the spoken words, we have provided these references by means of a supervised classifier based on the mean square error. An experimental approach has been followed for the design of experiments enabling to determine the minimal elements in the sample for each voice command. Satisfactory results have been obtained leading to a better understanding of variability of the system functioning. Finally, we have noticed that the success rate of the MLP and the minimal number of elements used for the learning process depend on the spoken word structure and of the variability of the actual work situation (word length, noise, speaker, etc).

Forecasting Business Efficiency by Using Classification Techniques: A Comparative Analysis Based on a Spanish Case

A comparative, simulation-based study of the performance of a number of classificatory devices, both parametric (LDA and Logit) and non-parametric (perceptron neural nets and fuzzy-rule-based classifiers) is conducted. The paper uses as a benchmark the problem of forecasting the level of efficiency of Spanish commercial and industrial companies upon the basis of a set of financial ratios. This case illustrates well some distinctive features/complexities of many financial prediction problems, namely that of being characterized by a high dimension feature space and low degree of separability. Results indicate a higher performance of model-free classifiers, even for moderate sample sizes. The average effects of sample size variations on the predictive of each classifier are also measured by using Monte Carlo simulations, and response surfaces are estimated in order to summarize these effects.

A Machine Learning Approach to Automatic Music Genre Classification

Abstract This paper presents a non-conventional approach for the automatic music genre classification problem. The proposed approach uses multiple feature vectors and a pattern recognition ensemble approach, according to space and time decomposition schemes. Despite being music genre classification a multi-class problem, we accomplish the task using a set of binary classifiers, whose results are merged in order to produce the final music genre label (space decomposition). Music segments are also decomposed according to time segments obtained from the beginning, middle and end parts of the original music signal (time-decomposition). The final classification is obtained from the set of individual results, according to a combination procedure. Classical machine learning algorithms such as Naïve-Bayes, Decision Trees, k Nearest-Neighbors, Support Vector Machines and Multi-Layer Perceptron Neural Nets are employed. Experiments were carried out on a novel dataset called Latin Music Database, which contains 3,160 music pieces categorized in 10 musical genres. Experimental results show that the proposed ensemble approach produces better results than the ones obtained from global and individual segment classifiers in most cases. Some experiments related to feature selection were also conducted, using the genetic algorithm paradigm. They show that the most important features for the classification task vary according to their origin in the music signal.

Forecasting business profitability by using classification techniques: A comparative analysis based on a Spanish case

A comparative study of the performance of a number of classificatory devices, both parametric (LDA and Logit) and non-parametric (perceptron neural nets and fuzzy-rule-based classifiers) is conducted, and a Monte Carlo simulation-based approach is used in order to measure the average effects of sample size variations on the predictive performance of each classifier. The paper uses as a benchmark the problem of forecasting the level of profitability of Spanish commercial and industrial companies upon the basis of a set of financial ratios. This case illustrates well a distinctive feature of many financial prediction problems, namely that of being characterized by a high dimension feature space as well as a low degree of separability. Response surfaces are estimated in order to summarize the results. A higher performance of model-free classifiers is generally observed, even for fairly moderate sample sizes.

Forecasting business profitability by using classification techniques: A comparative analysis based on a Spanish case

A comparative study of the performance of a number of classificatory devices, both parametric (LDA and Logit) and non-parametric (perceptron neural nets and fuzzy-rule-based classifiers) is conducted, and a Monte Carlo simulation-based approach is used in order to measure the average effects of sample size variations on the predictive performance of each classifier. The paper uses as a benchmark the problem of forecasting the level of profitability of Spanish commercial and industrial companies upon the basis of a set of financial ratios. This case illustrates well a distinctive feature of many financial prediction problems, namely that of being characterized by a high dimension feature space as well as a low degree of separability. Response surfaces are estimated in order to summarize the results. A higher performance of model-free classifiers is generally observed, even for fairly moderate sample sizes.