Generative AI Basics: Fundamentals to Real-World Impacts

Discover how generative adversarial networks pair a generator and a discriminator in adversarial training to produce realistic images, text, and music, with diverse outputs and real-world applications.

Explore the architecture of generative adversarial networks, with a generator and discriminator trained adversarially on a latent vector, and study losses, binary cross-entropy, mode collapse, and image or text generation.

Explore generative adversarial networks, where a generator and a discriminator engage in adversarial training to learn the data distribution and produce realistic images, text, or music.

Explore the fundamentals of generative adversarial networks and how the generator and discriminator train adversarially to produce diverse, photorealistic images from latent vectors.

Join a hands-on session to implement a simple gan that generates images, including handwritten digits from the mnist dataset, and save the final generated images as png files.

Understand variational autoencoders that encode data into a latent space and decode reconstructions. Learn how they optimize reconstruction and KL divergence losses to model a probabilistic distribution for image generation.

Explore how encoders map inputs to latent representations and decoders reconstruct data in variational autoencoders, enabling image generation and data augmentation through Gaussian priors and joint training.

Explore the objective function of variational autoencoders, balancing reconstruction loss and KL regularization to learn a smooth latent space for data generation and manipulation.

Discover variational autoencoders as versatile generative models, enabling image generation, inpainting, style transfer, text generation and summarization, anomaly detection, reinforcement learning applications, data augmentation, and drug discovery.

Set up the environment with Python and essential libraries, build a simple VAE model using the Keras API, train it, and generate or reconstruct MNIST images.

Explore the basics of recurrent neural networks, their memory of past inputs, and their use in NLP, speech recognition, and time series.

Explore sequence generation with recurrent neural networks, showing how hidden states enable memory and how LSTM and GRU address vanishing gradients for language modeling and text generation.

Explore long short-term memory (LSTM) networks, a recurrent neural network with a memory cell and gates that capture long-term dependencies for language modeling, speech recognition, and time series forecasting.

Learn how recurrent neural networks train to generate sequences such as text, music, or code. Explore LSTM architectures, teacher forcing, and how seeds drive coherent sequence generation.

Set up a python environment with tensor flow and keras, create an rnn_text_generation.py script, prepare sample training data, build an rnn model, and add a text generation function.

leverage pretrained models to improve related tasks with limited data using feature extraction or fine tuning, enabling more accurate transfer learning across domains such as image recognition and speech processing.

Fine-tune pre-trained models to tackle new generative tasks, boosting performance with less data and resources. Learn steps, benefits, and creative improvements across text, image, and audio.

Leverage transfer learning to adapt pre-trained image generation models for new data, enabling high-quality, varied images with limited data; explore GANs and VAEs, fine-tuning, and diverse applications.

Discover how transfer learning leverages pre-trained language models to boost text generation, enabling fine-tuning on task-specific data for creative writing, dialogue, and summaries.

Engage in a hands-on session on transfer learning for generative AI, linking fundamental concepts to real-world impacts.