Lin Hsin Hsin Artificial Intelligence Center: Supervised Training from the founder of LIN HSIN HSIN ART MUSEUM--Digital Art Museum, First Virtual Museum in the World, 1994.Wikipedia, Digital Media Center: Technology, Digital Art, Digital Paintings, Digital Sculptures, Digital Music, Digital Musical Instruments, Sound, Poetry, Animated Music, Web-enabled, Interactive, Digital Media Poineer
  Lin Hsin Hsin
   Artificial Intelligence Center





Supervised Learning






INPUT

        {(x1, y1), (x2, y2),......,(xN, yN)}
            s. t. xi = feature vector
                    yi = label

            where each data point contains
            features (vectors, covariates)
            denoted by:

            📍 discrete
            📍 discrete ordered
            📍 counts
            📍 continuous values

            & an associated label (class), which
            determines training data to be used
            determines the input features




        ACCURACY         

        is determined by the heterogeneity of the data
        & hence it
        determines the structure of the learned function



        ALGORITHM        

              maps
              vectors (input)
              to
              labels (output)

              based on the given
              input-output pairs



        TYPES of ALGORITHM                
              naive Bayes

              support-vector machines
              support-vector machines
              with Gaussian kernels

              logistic regression
              linear regression
              linear discriminant analysis

              neural networks
              decision trees
              K-nearest neighbor
              similarity learning


        ACCURACY of ALGORITHM

            bias-variance tradeoff
            function complexity
            amount of training data
            dimensionality of the input space
            generates noise in the output values





        STATISTICAL QUALITY  

            of an algorithm is measured
            through the generalization error



        OUTPUT

          The Inference

          infers a function from labeled training data
          consisting of a set of training examples

          a pair consisting of an input object (a vector)
          &
          a desired output value (supervisory signal)

          analyzes the training data
          produces an inferred function

          which can be used for mapping new examples

          correctly determine the class labels
          for new instances


          👟    
          GENERATIVE TRAINING   

            Discriminative Model


            discriminative training method
            that seek to find a function
            that discriminates well between

            the different output values
             f (x, y) = P (x, y)



          ⚡    
          RISKS           

                🛎 Empirical risk
                🛎 Structural risk



          GENERAL ISSUES     

                inductive bias
                overfitting
                uncalibrated class
                membership probabilities