فلش‌کارت‌های آموزشی PyTorch از صفر تا صد: ساخت شبکه‌های عصبی با گرادیان‌محور

🎓 دوره آموزشی جامع

📚 اطلاعات دوره

عنوان دوره: دوره PyTorch از صفر تا صد: ساخت شبکه‌های عصبی با گرادیان‌محور

موضوع کلی: هوش مصنوعی، حاکمیت داده و بازارهای کار

موضوع میانی: فریم‌ورک‌های یادگیری ماشین

📋 سرفصل‌های دوره

• 1. PyTorch environment installation and setup
• 2. Introducing tensors: creation and basic properties
• 3. Tensor operations: arithmetic and indexing
• 4. Reshaping and transforming tensors
• 5. Autograd mechanism: tracking operations
• 6. Building your first computational graph
• 7. Gradients and backpropagation fundamentals
• 8. Linear model implementation from scratch
• 9. Loss functions: Mean Squared Error
• 10. Loss functions: Cross-Entropy for classification
• 11. Optimizer introduction: SGD basics
• 12. Optimizer introduction: Adam and adaptive learning rates
• 13. Training loop: forward, loss, backward, step
• 14. Validation loop and accuracy calculation
• 15. Model checkpointing: save and load
• 16. Regularization: Dropout layer usage
• 17. Custom datasets: inheriting from Dataset class
• 18. DataLoaders: batching and shuffling
• 19. Image preprocessing: ToTensor and Normalize
• 20. Data augmentation: RandomCrop and Flip
• 21. Convolutional layers: Conv1d, Conv2d, Conv3d basics
• 22. Padding, stride, and receptive field
• 23. Building a complete CNN architecture
• 24. Activation functions in CNNs: ReLU, LeakyReLU
• 25. Pooling layers: MaxPool and AvgPool
• 26. Transfer learning: using pretrained models
• 27. Fine-tuning strategies for transfer learning
• 28. CNN training workflow end-to-end
• 29. RNN fundamentals: SimpleRNN basics
• 30. LSTM cells: architecture and usage
• 31. GRU cells: gated recurrent unit overview
• 32. Sequence processing with PyTorch RNNs
• 33. Bidirectional RNNs and their advantages
• 34. Attention mechanism: additive attention
• 35. Scaled dot-product attention basics
• 36. Transformer encoder layer fundamentals
• 37. Building a custom transformer block
• 38. Positional encoding implementation
• 39. Encoder-only models: BERT-style overview
• 40. Decoder-only models: GPT-style overview
• 41. Sequence-to-sequence models with attention
• 42. Text tokenization and vocabulary building
• 43. Embedding layers for categorical and text data
• 44. Text classification models: CNN-RNN-Transformer comparison
• 45. Hyperparameter tuning: learning rate and epochs
• 46. Early stopping and model selection
• 47. Mixed precision training with torch.cuda.amp
• 48. Distributed data parallel (DDP) introduction
• 49. Custom loss functions for specialized tasks
• 50. Metric tracking: precision, recall, F1-score
• 51. Gradient clipping and exploding gradient prevention
• 52. Model export to TorchScript format
• 53. Exporting to ONNX for interoperability
• 54. PyTorch Lightning: organizing code with trainers
• 55. Optuna integration for hyperparameter search
• 56. Time series forecasting with LSTM
• 57. GAN fundamentals: generator and discriminator
• 58. DCGAN architecture and training loop
• 59. Conditional GAN: class-conditional generation
• 60. Autoencoder: encoder-decoder structure
• 61. Variational Autoencoder: probabilistic latent space
• 62. Research prototyping best practices in PyTorch
• 63. Integration with NumPy, Pandas, and Scikit-learn
• 64. Code modularity and reusability patterns
• 65. Final capstone project: end-to-end PyTorch workflow