Page 358 - AI Ver 3.0 class 10_Flipbook
P. 358

Task                                                          21 st  Century   #Information Literacy
                                                                                              Skills

                                                                                                       CBSE Handbook
                   Convolutional Neural Network

                      Convolutional           Rectified linear             Pooling            Fully Connected
                           Layer                unit (ReLU)                 Layer                   Layer



                                                      10
                    1  1  1  0  0                      8                     Max [1,1,5,6]=6
                    x1  x0  x1
                    0  1  1  1  0  4                   6                                       Car         70 %
                    x0  x1  x0
                                                       4
                    0  0  1  1  1                                   x          Max pool with
                    x1  x0  x1                                         1  1  2  4  2×2 filters and   Truck  20 %
                                                       2
                    0  0  1  1  0                                      5  6  7  8  stride 2  6  8
                                                                       3  2  1  0      3  4
                    0  1  1  0  0            –10  –5        5   10     1  2  3  4              Bicycle     10 %
                                   Convolved   Output = Max(zero, Input)
                       Image        Feature                                  y
                                                                    Rectified Features Map
                                     Reduce size, improve feature, give probability value
                Write the whole process of how a CNN works on the basis of the above diagram.




























              Testing CNN


              To test a CNN, you typically need to perform the following steps:
              1. Data  Preparation:  Collect  a  dataset  suitable  for  your  specific  task,  such  as  image  classification  or  object
                 detection. Split the dataset into training, validation, and test sets.
              2. Model Selection: Choose a CNN architecture suitable for your task, such as VGG, ResNet, or Inception. You may
                 also consider using pre-trained models to save time and resources.
              3. Model Training: Train the CNN on the training dataset using an appropriate loss function and optimisation
                 algorithm. Monitor the model's performance on the validation set during training to avoid overfitting.
              4. Evaluation: After training, evaluate the CNN's performance on the test dataset to assess its generalisation
                 ability. The common evaluation metrics for image classification tasks include accuracy, precision, recall and F1-
                 score.
                    356     Touchpad Artificial Intelligence (Ver. 3.0)-X
   353   354   355   356   357   358   359   360   361   362   363