In the first module, we'll get familiar with the Multi-Task Learning algorithms implemented inside companies like Tesla, and we'll study the main rules and barriers in Multi-Task Learning.

You'll learn:

• How to Build Advanced Multi-Task Models with PyTorch (even if you're a complete beginner) 
  
  
• 2 Beheading Techniques to add heads to any pretrained Neural Network
  
  
• How to build a Loss Function that can be optimized for 2 completely different problems, such as classification and regression.
  
  
• How to tune your hyperparameters when heads are learning at different rates
  
  
• PyTorch 101 — My cook-book to build a DataLoder, Create Models, and Train Parameters from scratch with PyTorch.
  
  
• Inside Tesla's HydraNets — A look at Tesla's Multi-Task Algorithms and how they work to make it scalable at several hundreds of thousands of cars
  
  
• 10 Advanced Deep Multi-Task Architectures to force communication inside a Neural Network— We'll go over Cross-Stitch Units, Bayesian Neural Networks, and many other ideas...
  
  
• Multi-Task Learning in Computer Vision — Deep Dive inside an untold Experimentation that reveals which task you should train together... and avoid mixing!
  
  
• The Intermediate Python Concepts you should know to make your code look more professional
  
  
• ✅ PROJECT: Build your first multi-task learning algorithm from scratch with PyTorch to process images and do binary classification, regression, and multi-class classification.

In the second module, we'll consolidate our PyTorch skills and implement an advanced research paper capable of doing Semantic Segmentation and Monocular Depth Estimation in Real-Time.

You'll learn:

• The Encoder-Decoder Architectures used to build multi-heads decoders that work better and faster than single task networks
  
  
• MobileNet v2 from scratch — Build an mobile optimized encoder using Inverted Residual Blocks, Linear Bottlenecks, and ReLU6 Operations 
  
  
• Separable Convolutions and the future of Convolutional Neural Networks
  
  
• Ultra Fast Multi-Head Decoders for Monocular Depth Estimation and Semantic Segmentation (we'll go over CRP Blocks, CRU Blocks, and many other advanced Deep Learning techniques used in the field)
  
  
• ✅HYDRANET PROJECT — Build a HydraNet trained with 2 to 3 heads that does real-time semantic segmentation and depth estimation in self-driving cars.
  
  
• 3D Segmentation — My technique to fuse a Depth Map with a Segmentation Map and reconstruct a 3D Point Cloud already segmented. (Hint: You can also add other "heads" outputs to this point cloud, such as object detection boxes)

That last point in particular is very interesting: 3D SEGMENTATION.
If we have a depth map and a segmentation map, and assuming we know our camera parameters, we can build a 3D Point Cloud that has already been segmented.
Here's what it looks like:

In module 2, we'll use a pretrained model and focus on building a model, then running it and analyzing the output. In Module 3, we'll learn how to train it.

In the last module, we'll implement the training techniques used to come up with a state-of-the-art multi-task encoder/decoder network such as the one we used in Module 2.

You'll learn:

• The Loss Function of a model doing Monocular Depth Estimation Neural Networks and Semantic Segmentation Neural Networks (plus what are the independent losses for single-task models)
• How to train a HydraNet with or without Transfer Learning
• A simple trick to to influence training to learn one task more than the other and finetune a model.
• How to use multiple optimizers with PyTorch.
• ✅ Home Robot Project — Train a HydraNet for Home-Robot Segmentation and Depth Estimation.

Before joining, my biggest obstacle is the time required to complete this course. But I enrolled, and I have discovered a network architecture for multitasking that I did not know how to implement. I loved the quality of the material and the practical content, and I liked to learn about possible practical application fields of multitasking.

This is a magnificent practical course for discovering real areas for the application of multitasking architectures. The didactic quality of the course and its material were great.