On Thursday, we have another student speaker, Clare Heinbaugh who is currently a sophomore at the college. The topic of this talk is Transfer learning, stacked generalization, and Keras (focus on Python packages). Clare started off introducing the neural networks, which is consists of the input layer to hidden layer and to output layer. She then introduce a project she did with Spotify data. She focused on the Top 4 genres which consist of 80% of data from Spotify. The model is designed to identify the genre with the sonic frequencies. She then elaborate on the train-test split methodology and Keras. The next concept she talks about is convolutional neural network. We learned the concept back in the first module of this class. Hence, it was easy to understand how to convert the color band into matrices. We also use filters to convolve over the image. The new concept I learned is transfer learning. The advantage of transfer learning is that there are pre-trained models that have fixed weights to apply it on our own models. The example that Clare shared with us is using CNNs to predict road quality. She uses three transfer learning models, VGG16, InceptionV3, and ResNet50. At last, she talks about the stacked generalization, where three models for different output are stack for the prediction.