PyTorch for Deep Learning

What you'll learn

• Build deep learning systems step-by-step: Code neural networks from the ground up in PyTorch. Work directly with tensors, neural networks, and training loops to create and evaluate your first image classifier.

• Apply your skills to vision and natural language tasks: Use TorchVision and Hugging Face models, fine-tune pretrained models, compare architectures, and boost performance through hyperparameter tuning.

• Advance to modern architectures and deployment: Explore structures such as Siamese networks, ResNets, and Transformers. Interpret model behavior, and prepare your models for deployment using tools and techniques such as ONNX, MLflow, pruning, and quantization.

Why Learn PyTorch

About this Professional Certificate

Hands-on projects for a better learning experience

• Begin by building your first neural networks in PyTorch, then take on image classification as you teach a model to tell apart categories like flowers, insects, and animals while learning how neural networks learn from data.

• Improve model performance and efficiency on real datasets. Experiment with optimizers and learning-rate schedulers, automate hyperparameter tuning with Optuna, and use profiling tools to identify bottlenecks in your training pipeline.

• Apply PyTorch to computer vision projects using TorchVision to load, transform, and augment image data; fine-tune pretrained networks like ResNet and MobileNet; and visualize predictions with saliency maps and class activation maps.

• Build and fine-tune language models using Hugging Face and PyTorch. Implement text preprocessing, compare embeddings (GloVe, FastText, DistilBERT), and train classifiers that can analyze real text datasets.

• Explore advanced architectures in PyTorch by building transformer models from core components like multi-head attention, and get an introduction to modern generative approaches such as diffusion models.

• Prepare models for real-world deployment by exporting them with ONNX, tracking experiments in MLflow, and applying pruning, quantization, and quantization-aware training to reduce model size and latency.

Recommended Background