Applied Deep Learning: Vision, Language, and Generation

Description:

This immersive three-day course provides a hands-on introduction to the foundations, applications, and frontiers of neural networks and deep learning. Participants begin by understanding how data becomes intelligence through neural networks, exploring how models learn, train, and generalize. The course then advances into applying deep learning to vision, text, and sequential data before culminating in the study of generative models that create text, images, and multimodal outputs.

By combining theory with guided coding labs using TensorFlow and open-source models, learners gain both conceptual clarity and practical skills to build, train, and experiment with deep learning systems.

Duration: 3 Days

Course Code: BDT 526

Learning Objectives:

After this course, you will be able to:

1. Explain the key principles of deep learning, including how neural networks represent and learn complex patterns.
2. Build, train, and evaluate neural networks using modern frameworks such as TensorFlow.
3. Identify and address common challenges in training deep models, such as overfitting, bias.
4. Apply convolutional networks to computer vision tasks and understand how they extract hierarchical features from images.
5. Understand sequence models, including RNNs, LSTMs, and Transformers, and apply them to language problems.
6. Use pre-trained language models (e.g., BERT, GPT) for tasks such as classification, summarization, and text generation.
7. Explain how generative AI models—including diffusion models and large language models—create new text and images.
8. Build and experiment with generative AI applications such as chatbots or text-to-image tools using open-source models.

Basic familiarity with Python programming and machine learning

Course Outline:

Day 1: Getting Grounded in Deep Learning

1. The Core Idea

• How data becomes intelligence
• What is a neural network?
• What “deep” means
• Typical applications of deep learning: vision, text, speech

2. The Building Blocks of a Neural Network

• Neural networks: layers, neurons, and activations
• How neural networks learn patterns
• Forward pass and loss
• Backpropagation
• Parameters, weights, and biases

3. Training and Evaluating Models

• The training loop explained step-by-step
• Common problems: overfitting, underfitting, bias, variance
• Practical tricks: normalization, dropout, data augmentation
• How to interpret results (loss curves, accuracy metrics)

4. Hands-On: Building a Neural Network

• Using TensorFlow to build a neural network to classify images
• Visualizing model performance and debugging learning issues

Day 2: Using Deep Learning

1. Seeing and Understanding (Computer Vision)

• Convolutional Neural Networks (CNN)
• Why convolution is powerful for images
• CNNs as feature extractors: edges to objects
• Filters, feature maps, and pooling

2. Learning from Sequences (Language and Time Series)

• Why sequences matter
• RNNs, LSTMs and remembering context
• Time series data

3. Practical NLP

• Tokenization and embeddings
• Vocabulary
• Practical application (e.g., sentiment analysis, classification)

4. Hands-On: Applied Deep Learning

• Build, train, and use an image recognition model
• Evaluate the model’s performance

Day 3: Generative AI

1. Text Generation and Large Language Models (LLMs)

• Attention and Transformers explained
• How transformer (e.g. BERT, GPT) create coherent text
• Prompting, retrieval-augmented generation, and fine-tuning
• Practical use: summarization, code generation, chat systems
• Bias, and hallucination issues

2. 2. Image Generation and Diffusion Models

• Diffusion models (e.g. Stable Diffusion, DALLE)
• Text-to-image architecture
• Hands-on: generating and editing images using open-source models

3. Beyond Text and Images: Multimodal Generation

• Combining text, vision, and sound
• The rise of video generation and voice synthesis (e.g., Sora, Veo)
• Responsible AI and ethical considerations

4. Hands-On: Build Generative AI App

• Build a text-to-image interface or a chatbot using an open-source LLM
• Discussion of results

Training material provided: Yes (Digital format)