Hands-on Deep Learning
From mathematical derivations of backpropagation to implementing Vision Transformers (ViT) and GANs.
The Hands-on Deep Learning course is an advanced module at Hobot Academy, focusing on opening the “black box” of neural networks. We dive deep into the internal mechanics of state-of-the-art models, ensuring students understand both the calculus behind the gradients and the engineering behind the training loops.
The complete curriculum, including handwritten mathematical derivations, NumPy-based neural networks from scratch, and PyTorch implementations of complex architectures, is available on GitHub.
🚀 VIEW PROJECTS & CODE ON GITHUB
Technical Syllabus
The course is organized into intensive modules covering the evolution of Neural Networks:
- Neural Foundations: Implementing Backpropagation from scratch and mastering Multi-Layer Perceptrons (MLP).
- Computer Vision: Advanced CNNs, Residual Connections (ResNet), and medical image segmentation using U-Net.
- Modern Architectures: Transitioning from RNNs to Vision Transformers (ViT) and Attention mechanisms.
- Generative Models: Building and training Generative Adversarial Networks (GANs) for image synthesis.
- Object Detection: Exploring region-based CNNs (R-CNN) for precise localization.
Technologies used: PyTorch, TensorFlow, NumPy, OpenCV, Jupyter.
