Ben Lew

Artificial neural networks (ANNs) are a machine learning (ML) method that tries to replicate how the brain can learn from new information. Neural networks are made up of neurons that each hold a number and are linked together with parameters take in data this can be a value or a image and either label the image or use the value to predict a problem. These neurons are divided into three layers: the input layer, hidden layer, and output layer. The input layer takes in the unlabeled data, the hidden layer alters it and the output layer outputs it as labeled data. ANNs are able to do this by undergoing a training phase where they are shown labeled data that is then used to adjust their parameters. The parameters are adjusted by comparing the results from the network when they are shown unlabeled data and putting it into a cost function with the correct label. With these two results in a function, we can minimize the function to find the closest possible value to the correct label. After this training phase the network is now able to label data. 

Recurrent neural networks (RNNs) are a type of ANN that can process sequences of data to give context to individual data values in a large sequence. An example is translation between languages, where each word is based on the previous words in a sentence even if the words are an exact translation. They can do this by utilizing recurrent edges which  are loops added to the network that take the value from the output and gives each part of the dataset a secondary value for context. Most research in the field is trying to optimize the process of finding the minimal value of the cost function in training efficiently.