University:
College
Course:
College Computer Science
Academic year:

2023
Views:

230

Pages:

1
Author:

Jocy M. J.

A branch of machine learning called deep learning is concerned with creating artificial neural networks and algorithms that are modelled after the structure and operation of the human brain. It excels at handling difficult issues involving massive volumes of data, such as audio and picture recognition, natural language processing, and gaming. Deep learning's fundamental ideas and elements include: Artificial neural networks, which are made up of linked nodes or "neurons," are frequently used to create deep learning models. An input layer, one or more hidden layers, and an output layer are only a few of the layers that make up this network of neurons. A weight is assigned to each neuronal connection, and this weight is changed throughout training to improve the model's performance. Deep learning models are distinguished by their depth, which refers to the fact that they include several hidden layers. When opposed to shallower architectures, this depth enables them to automatically learn complex characteristics and representations from raw data. Activation Functions: By introducing non-linearity to the neural network, activation functions enable the neural network to recognise intricate patterns in data. Rectified Linear Unit (ReLU), sigmoid, and tanh are frequently used activation functions. Backpropagation: Deep learning models are trained with the backpropagation optimisation technique. It entails calculating the gradients of the loss function of the model with respect to the parameters of the model and changing the parameters in the direction that lowers the loss. Convolutional Neural Networks (CNNs): CNNs are a type of neural network specifically designed for processing grid-like data, such as images and videos. They use convolutional layers to automatically learn features like edges, textures, and shapes. Recurrent Neural Networks (RNNs): RNNs are designed to work with sequences of data, such as timeseries data or natural language. They have loops that allow information to be carried across different time steps, making them suitable for tasks like language generation and speech recognition. The vanishing gradient problem is addressed by Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs), two specialised forms of RNNs that can detect longer-range relationships in sequential data. Transfer Learning: Transfer learning is the process of addressing new problems by using previously learned models as a starting point. Utilising the skills acquired from one assignment to enhance performance on another can save time and resources. GANs, or generative adversarial networks: GANs are made up of two neural networks that are trained in tandem while playing games: a generator and a discriminator. For tasks including picture synthesis, style transfer, and data augmentation, GANs are employed. Natural Language Processing (NLP): Deep learning has transformed NLP activities including text creation, sentiment analysis, machine translation, and question answering. Although deep learning has shown amazing results in a number of sectors, it also needs a lot of computer power, a lot of data, and careful tuning to prevent overfitting. In order to improve, researchers are looking at novel architectures and methods.