Deep Learning with PyTorch: Neural Networks in Practice
- Hands-on AI and data science techniques from working practitioners
- Structured program with clear milestones and real-world datasets
- Accessible remotely — join from anywhere in Ukraine
About this program
Neural networks without the mysticism
Deep learning has a reputation for being hard to debug and harder to understand. Much of that difficulty comes from skipping the fundamentals. This program builds understanding layer by layer — literally.
You will implement networks from scratch before using higher-level APIs, so you understand what backpropagation is actually doing when a training run goes wrong.
Project areas
The program covers two application domains: computer vision (CNNs for image classification, transfer learning with ResNet) and natural language processing (tokenization, embeddings, sequence models).
Hardware requirements
All training runs are executed on cloud GPU instances provided through the course platform. A standard laptop is sufficient for writing and testing code locally.
Realistic expectations
Deep learning requires significant time investment. Participants who complete all assignments typically spend nine to eleven hours per week. The program is not suitable for those without prior Python and basic ML experience.
- Prerequisite: Python proficiency
- Prerequisite: familiarity with gradient descent and loss functions
Program structure
-
Part 1 — PyTorch Core
2 weeksTensors, autograd, building a neural network from scratch, training loops.
-
Part 2 — Convolutional Neural Networks
3 weeksCNN architecture, image augmentation, transfer learning, fine-tuning pretrained models.
-
Part 3 — Sequence Models and NLP
3 weeksWord embeddings, RNNs, LSTMs, and an introduction to the Transformer architecture.
-
Part 4 — Training at Scale
1 weekMixed precision training, learning rate schedulers, experiment tracking with Weights and Biases.
-
Part 5 — Final Projects
2 weeksTwo projects: one vision task and one NLP task, both evaluated by instructors.