Comprehend Deep Learning
Deep Learning
Deep Learning is a subset of machine learning. Technically, machine learning looks through input data for valuable representations, making use of feedback signal as guidance. The “deep” in deep learning isn’t a reference to any kind of deeper understanding attained by the approach; rather, it stands for the idea of consecutive layers of representations. The “depth” of the model is the indication of the number of layers subscribed to the model. Concurrently, shallow learning’s (another approach of machine leaning) focal point is to learn representations from merely one or two layers.
Digit classification example
The above figure shows a network with multiple layers predicting the handwritten input digit. Each layer in the network learns different features about the data and passes it as an input to the next layer. While specifying the n-dimensional intermediate layers (hidden units) we need to be mindful of not establishing any “information bottleneck”. Within the pile of dense layers, if one layer drops any relevant information; that information cannot be recuperated by the subsequent layer. For example, while predicting the output from the dataset of 10 different classes; (in case of this example, numbers from 1 to 10) if we introduce only 6-dimensional intermediate layers information bottleneck will arise because of such compact layers, which might lead to irreversible loss of pertinent information.
Overview of how deep learning functions
The transformation carried out by a layer is parameterized by its weights. To map the inputs to the associated labels, we need to find a suitable set of values for the weights. But how do we know what are the best appropriate values to be considered as weights? Well, this can seem intimidating at first however it is pretty easy to understand.
Initially, weights are assigned some random values, so the network implements random transformations on the input data and produces an output. Needless to say, the output that will be obtained will be far from the truth label (associated label to the input data). Now, this is where the loss function and optimizer play a vital role. The loss function (aka objective function) computes the distance between the predicted label and the truth label and yields a loss score, which helps to access how well the model is functioning. The values of weights are altered to lower the loss score based on the feedback signal (previous loss score). Optimizers are the ones who handle this task. This process is repeated until we’ve reached a minimal loss score. A network with minimal loss score has the predicted values close to truth values.
I hope this blog post helped you grasp the basic concept of deep learning. Thank you for reading. Have a productive day :)
