Neural networks in neuroscience are a circuit of neurons that power the human brain. In the field of computer science, however, neural networks are defined as a tool or technique so a computer program can make decisions from observational data, similar to how humans make decisions.[5] Deep learning is the field of machine learning focused on how these programs learn based on the data provided.
There are 3 main types of learning: supervised, unsupervised, and reinforcement learning. Supervised learning means that the algorithm is trained on a set of labeled data. Supervised learning is especially useful when it comes to classification problems; however, it is not always possible to access labeled data sets. In this case, computer scientists may use unsupervised learning.
In unsupervised learning, a deep learning model explores a dataset, finds patterns, and predicts an output. Data sets on unsupervised learning models are not organized like data sets used for classification/supervised learning. Unlike supervised learning, unsupervised learning, there is no specific output or correct answer. Because of this, these two types of learning algorithms have very different uses. Often, unsupervised learning is used for the analysis of large datasets. [5]
Lastly, reinforcement learning, influenced by the behavioral psychology term, is a technique that trains models to make predictions and decisions based on an established end goal. Reinforcement learning requires the developer to reward the program for the correct decisions and penalize the algorithm for incorrect choices, similar to a game. To make decisions, the model relies on past feedback and new learnings through data. Slowly, the algorithm is trained and eventually finds an optimal way to accomplish the goal. On a large scale, the goal of reinforcement learning models is to “predict the best next step to take to earn the biggest final reward.” [5] The process to get accurate predictions depends on how much reinforcement or feedback the machine has received. Today, reinforcement learning is popular in the field of robotics. Neural networks and deep learning currently provide the best solutions to many areas in image recognition, speech recognition, and natural language processing. [7]
Log in or sign up for Devpost to join the conversation.