* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download RBF
Survey
Document related concepts
Transcript
RBF ch1,2,3 Overview of RBF Networks • RBF networks have three layers: input layer , output layer, and hidden layer. • Output is a real value. • One neuron in the input layer corresponds to each predictor variable. • Each neuron in the hidden layer consists of a RBF function(Gaussian,etc) • Each neuron centered on a point with the same dimensions as the predictor variables • The output layer has a weighted sum of outputs from the hidden layers. Overview of RBF Networks Supervised Learning &Unsupervised Learning Supervised Learning: categorized into "regression" and "classification"problems. Unsupervised Learning: We can derive structure from data where we don't necessarily know the effect of the variables. Nonparametric Regression& Parametric Regression Parametric Regression: parameters have meaningful interpretations, such as initial water level or rate of flow Y depends on X Nonparametric Regression: parameters have no particular meaning in relation to the problems to which they are applied. Nonparametric Regression Nonparametric Regression : the primary goal is to estimate the underlying function Y depends on weight and basis function • RBF神經網路在架構上是一種3層前饋網路。 輸入層到輸出層的是非線性的(即:隱藏層 的函數),但是隱藏層到輸出層的映射卻是 線性的(即:輸出層的函數),因此可以加快 網路的學習速度。 • 高維度空間的資料分類問題,比低維度空 間更符合線性分離趨勢。 The idea y Training Data x The idea y Training Data x Basis Functions (Kernels) The idea y Function Learned x Basis Functions (Kernels) The idea y Nontraining Sample Function Learned x Basis Functions (Kernels) Linear model m Formula: f (x) wii (x) i 1 Example Linear Models • Polynomial f ( x) wi x i i ( x) x , i 0,1, 2, i i • Fourier Series f ( x) wk exp j 2k0 x k k ( x) exp j 2k0 x , k 0,1, 2, Single-Layer Perceptrons as Universal Aproximators y w2 w1 Hidden Units 1 x= wm 2 x1 x2 m xn m f (x) aswii (x) Radial Basis Function Networks i 1 Universal Aproximators y w2 w1 Hidden Units 1 x= wm 2 x1 x2 m xn With sufficient number of radial-basis-function units, it can also be a universal approximator. Linear model