These slides elaborate on Deep Learning model creation and application.
Hello and welcome to this section. In this section we will develop a high-level understanding of how a deep learning model is created.
1. We have different colors in this graph that should help the understanding. Inputs are highlighted in green. Model parameters are areas where you can take action and optimize the model by adapting the parameters. Model results are results of calculations that come from the simulation process itself. Ready? Let’s start.
2. We start with our data X. This is what is fed into the model. This data excludes our target variable. This is fed at a different point. The data usually needs to be proprocessed to be handled, but all these details are covered later.
3. Now, the data is passed to a neural network. The neural network is specified by different layers. In subsequent sections we will learn what types of layers are available, how many nodes per layer should be chosen, and much more. This is the core of the algorithm, the engine of our model. But in itself it is useless, so there are more parts necessary to build the whole model.
4. As a result of the calculation we get updated weights of all nodes.
5. The combination of data, our neural network layers and according weights results in a prediction. Do the prediction have a high or low quality? We can’t know.
6. For this we need to take our true values into account.
7. The true values and the predictions are inputs for the loss function. The objective of the loss function is to evaluate how well the model performs. This is done by minimizing the losses. There are different types of loss functions. The most appropriate loss function depends on the problem at hand.
8. After evaluating loss function we get a loss score. This loss score is used for fine-tuning the weights as we will see.
9. This adaptation is performed by an optimizer. The optimizer applies a so called Backpropagation algorithm and updates the weights.
10. Initially, weights are set randomly. The corresponding losses are high. With more and more training loops the weights are adapted in a way, that losses are decreased. When the loss function is minimized, the predictions are as close as possible to true values. The network is finally trained.
11. How can we apply our model and create predictions? We take test data X, again without our target variable. We feed this into our trained model, that consists of layers, nodes, and weights. As a result we get predictions. Important to mention is that the model should never have seen this data before. Now we can evaluate the performance of our model and possibly take it to production.
12. In subsequent sections we will dive deeper into the details of this model. We will study different neural network layer architectures, different loss functions, or different optimizers. But for a high-level understanding this is all you need to know for now.
That’s it for this lecture. Thank you and see you in the next one.
1. We have different colors in this graph that should help the understanding. Inputs are highlighted in green. Model parameters are areas where you can take action and optimize the model by adapting the parameters. Model results are results of calculations that come from the simulation process itself. Ready? Let’s start.
2. We start with our data X. This is what is fed into the model. This data excludes our target variable. This is fed at a different point. The data usually needs to be proprocessed to be handled, but all these details are covered later.
3. Now, the data is passed to a neural network. The neural network is specified by different layers. In subsequent sections we will learn what types of layers are available, how many nodes per layer should be chosen, and much more. This is the core of the algorithm, the engine of our model. But in itself it is useless, so there are more parts necessary to build the whole model.
4. As a result of the calculation we get updated weights of all nodes.
5. The combination of data, our neural network layers and according weights results in a prediction. Do the prediction have a high or low quality? We can’t know.
6. For this we need to take our true values into account.
7. The true values and the predictions are inputs for the loss function. The objective of the loss function is to evaluate how well the model performs. This is done by minimizing the losses. There are different types of loss functions. The most appropriate loss function depends on the problem at hand.
8. After evaluating loss function we get a loss score. This loss score is used for fine-tuning the weights as we will see.
9. This adaptation is performed by an optimizer. The optimizer applies a so called Backpropagation algorithm and updates the weights.
10. Initially, weights are set randomly. The corresponding losses are high. With more and more training loops the weights are adapted in a way, that losses are decreased. When the loss function is minimized, the predictions are as close as possible to true values. The network is finally trained.
11. How can we apply our model and create predictions? We take test data X, again without our target variable. We feed this into our trained model, that consists of layers, nodes, and weights. As a result we get predictions. Important to mention is that the model should never have seen this data before. Now we can evaluate the performance of our model and possibly take it to production.
12. In subsequent sections we will dive deeper into the details of this model. We will study different neural network layer architectures, different loss functions, or different optimizers. But for a high-level understanding this is all you need to know for now.
That’s it for this lecture. Thank you and see you in the next one.