1,721,080 research outputs found
Aggiornamento sulla terapia della bronchiolite
No full textNel lavoro vengono illustrati il quadro clinico, le implicazioni diagnostiche e terapeutiche della bronchiolite epidemica nel lattante e piccolo divezzo
Basi fisiopatologiche della terapia della bronchiolite
No full textNel lavoro viene fatta una revisione della letteratura sulla clinica e la terapia della bronchiolite
Neural Predictors’ Accuracy
No full textWe examine the performance of different predictors in the deterministic environment and test their robustness against noise. In particular, we mimic the classical case of measurement noise by adding a random Gaussian signal of different intensity to the deterministic output of some archetypal chaotic systems. Then, we examine the critical case of structural noise, represented by the slow variation of the growth rate parameter of the logistic map. In both cases, the presence of noise rapidly degrades all the performance indicators, but, interestingly, the best deterministic predictor, i.e., LSTM trained without teacher forcing, remains the best also in the stochastic and non-stationary environments. Finally, we examine solar irradiance and ozone concentration time series, and again the same predictor turns out to be the best and can also be reliably applied to similar datasets in the same domain (domain adaptation)
Neural Approaches for Time Series Forecasting
No full textThe problem of forecasting a time series with a neural network is well-defined when considering a single step-ahead prediction. The situation becomes more tangled in the prediction on a multiple-step horizon and consequently the task can be framed in different ways. For example, one can develop a single-step predictor to be used recursively along the forecasting horizon (recursive approach) or develop a multi-output model that directly forecasts the entire sequence of output values (multi-output approach). Additionally, the internal structure of each predictor may be constituted by a classical feed-forward (FF) or by a recurrent architecture, such as the long short-term memory (LSTM) nets. The latter are traditionally trained with the teacher forcing algorithm (LSTM-TF) to speed up the convergence of the optimization, or without it (LSTM-no-TF), in order to avoid the issue of exposure bias. Time series forecasting requires organizing the available data into input-output sequences for parameter training, hyperparameter tuning and performance testing. An additional developers’ choice explored in the chapter is the definition of the similarity index (error metric) that the training procedure must optimize and the other performance indicators that may be used to examine how well the prediction replicates test data
Artificial and Real-World Chaotic Oscillators
No full textFour archetypal chaotic maps are used to generate the noise-free synthetic datasets for the forecasting task: the logistic and the Hénon maps, which are the prototypes of chaos in non-reversible and reversible systems, respectively, and two generalized Hénon maps, which represent cases of low- and high-dimensional hyperchaos. We also present a modified version of the traditional logistic map, introducing a slow periodic dynamic of the growth rate parameter, that includes ranges for which the map is chaotic. The resulting system exhibits concurrent slow and fast dynamics and its forecasting represents a challenging task. Lastly, we consider two real-world time series of solar irradiance and ozone concentration, measured at two stations in Northern Italy. These dynamics are shown to be chaotic movements by means of the tools of nonlinear time-series analysis
Concluding Remarks on Chaotic Dynamics’ Forecasting
No full textIn this book, we compared different neural approaches in the forecasting of chaotic dynamics, which are well-known for their complex behaviors and the difficulty of their prediction. Our analysis shows that the LSTM predictor trained without teacher forcing is the most accurate approach in the forecasting of complex oscillatory time series. This predictor always provides the best accuracy in all the considered tasks, spanning a wide range of complexity and noise sources. It also demonstrates the ability to adapt to other domains with similar features without a relevant decrease of accuracy. The comparison with the real system used as predictor in a noisy environment is particularly interesting: even the complete knowledge of the system structure does not allow perfect predictions when the initial conditions are only approximately known. This allows the border between time series forecasting and system identification problems to be clearly defined
Basic Concepts of Chaos Theory and Nonlinear Time-Series Analysis
No full textWe introduce the basic concepts and methods to formalize and analyze deterministic chaos, with links to fractal geometry. A chaotic dynamic is produced by several kinds of deterministic nonlinear systems. We introduce the class of discrete-time autonomous systems so that an output time series can directly represent data measurements in a real system. The two basic concepts defining chaos are that of attractor—a bounded subset of the state space attracting trajectories that originate in a larger region—and that of sensitivity to initial conditions—the exponential divergence of two nearby trajectories within the attractor. The latter is what makes chaotic dynamics unpredictable beyond a characteristic time scale. This is quantified by the well-known Lyapunov exponents, which measure the average exponential rates of divergence (if positive) or convergence (if negative) of a perturbation of a reference trajectory along independent directions. When a model is not available, an attractor can be estimated in the space of delayed outputs, that is, using a finite moving window on the data time series as state vector along the trajectory
Neural Predictors’ Sensitivity and Robustness
No full textThe results of the application of deep neural predictors depend on a multitude of factors which compose the experimental settings. We report all the specific information to ensure the reproducibility of a wide number of numerical experiments. A sensitivity analysis on some critical aspects is provided in order to prove the robustness of our setting. Considering the long-term behavior of the predictors, those trained for the one-step forecasting are able to reproduce the statistical properties of the attractor, i.e., the so-called attractor’s climate, whereas the multi-step ones are unsuitable for replicating these statistical properties but provide an accurate forecasting up to several Lyapunov times. Lastly, we provide some remarks on the training procedure of the different predictors and introduce some advanced neural architectures to give an overview of possible advantages/disadvantages with respect to those implemented in this study
Introduction to Chaotic Dynamics’ Forecasting
No full textChaotic dynamics are the paradigm of complex and unpredictable evolution due to their built-in feature of amplifying arbitrarily small perturbations. The forecasting of these dynamics has attracted the attention of many scientists since the discovery of chaos by Lorenz in the 1960s. In the last decades, machine learning techniques have shown a greater predictive accuracy than traditional tools from nonlinear time-series analysis. In particular, artificial neural networks have become the state of the art in chaotic time series forecasting. However, how to select their structure and the training algorithm is still an open issue in the scientific community, especially when considering a multi-step forecasting horizon. We implement feed-forward and recurrent architectures, considering different training methods and forecasting strategies. The predictors are evaluated on a wide range of problems, from low-dimensional deterministic cases to real-world time series
- …
