Introduction to Generative AI Models
Generative AI is one of the most significant advances in the artificial intelligence field due to its ability to create new things. Generative AI models are mostly unsupervised algorithms. In contract discriminative techniques that learn to classify the data, generative AI techniques are mostly involved in creating new data from the training data.
This Framework enables computers to learn the underlying pattern associated with the input and then generate similar content.
There are three popular types of generative models:
- Generative Adversarial Networks(GANs)
- Autoregressive Convolutional neural networks (AR-CNNs)
- Transformer-based Models
Generative Adversarial Networks:
GANs employ two neural networks that compete with one another, a generator and a discriminator. The generator, also known as the generative network, is a neural network that is in charge of generating new data or content that is similar to the source data. The discriminator, also known as the discriminative network, is a neural network that distinguishes between the source data and the generated data. The Competition between these two networks to improve their methods until they manage to generate Indistinguishable data from the source content.
Use cases:
- Drug Discovery
- 3D Object Creation
Models:
- DCGAN(Deep Convulutional GAN)
- ProGAN(Progressively Growing GAN)
- BigGAN
Autoregressive Convolutional neural networks:
Autoregressive means regression on self. Here Autoregression term is used to imply predicting future outcomes of a sequence from the previously observed outcomes of that sequence.AR-CNNs are used to examine systems that evolve over time and assume that the probability of certain data is only dependent on what has happened previously. They rely on previous time-series data to generate accurate new data. The main autoregressive architectures are RNNs and casual convolutional networks.
Use Cases:
- Weather Forecasting
- Stock Prediction
- Music Generation
Models:
- PixelRNN
- PixelCNN
- WaveNet
Transformer-based Models
Transformer-based models are most commonly used to analyze data with a sequential structure (such as the sequence of words in a sentence). In the modern period, transformer-based techniques have become a common tool for modeling natural language. The ability of the transformer models to attend to different positions of the input sequence in order to compute a representation of that sequence is core to their architecture.
Use Cases:
- Text Generation
- Text Summarization
- Text Translation
Models:
- BERT(Bidirectional Encoder Representation from Transformers)
- RoBERTA(Robustly Optimized BERT)
