Using the Albumentations library for image augmentation in PyTorch, with example pipelines for training and TTA.

Last-minute competition score boosting: test-time augmentation (TTA) with soft voting and half-precision training.

Common NumPy pitfalls: dtype=object issues with mixed-type lists and axis behavior in np.mean.

Ensemble methods for AI competitions: hard voting, soft voting, and weighted voting to improve model performance.

How gradient accumulation works in PyTorch for effective large-batch training on limited GPU memory.

Practical training techniques including AMP, label smoothing, ArcFace loss, class pivot adjustment, and Wandb logging for image classification.

Clarifying common mistakes in PyTorch training loops, validation ordering, and K-fold cross validation usage.

CutMix data augmentation in PyTorch: loss computation, accuracy, and F1 score calculation.

TIL notes on wandb logging, fine-tuning pre-trained models, early stopping, and ensemble learning during an image classification competition.

Day 2 competition log covering ensemble strategies (bagging vs. boosting) and CNN model selection for image classification.

Tips on efficient data feeding, transform ordering for performance, and using Albumentations for image augmentation.

Image preprocessing and data augmentation techniques for deep learning competitions, including resizing, bounding box cropping, and torchvision transforms.

Training plan for a mask classification competition: per-feature models with ensemble learning, plus Jupyter vs. CLI workflow notes.

Common PyTorch troubleshooting tips for GPU memory issues, OOM errors, tensor accumulation, and inference best practices.

Overview of hyperparameter tuning methods including grid search, random search, and Bayesian optimization, with an introduction to Ray for parallel tuning.

Quick tip on displaying training accuracy and loss inside tqdm progress bars in PyTorch.

Guide to multi-GPU training in PyTorch covering model parallelism, DataParallel, and DistributedDataParallel with code examples.

Transfer learning in PyTorch: freezing layers, stepping frozen strategies, and model file formats.

Guide to PyTorch Dataset and DataLoader: data flow, tensor conversion, samplers, collate_fn, and transforms.

How PyTorch's nn.Module.apply() maps a function across all submodules via postorder traversal.

How to use PyTorch hooks for inspecting and modifying forward/backward passes in nn.Module.

PyTorch project tips: template structure, getattr for dynamic config, and abstract methods.

Essential PyTorch operations: view vs reshape, squeeze, matrix multiplication, index_select, gather, and more.

Hands-on PyTorch implementation of Scaled Dot-Product Attention and Multi-Head Attention for the Transformer.

Why weight initialization matters in deep learning and why zero initialization should be avoided.

Practical comparison of SGD, Momentum, and Adam optimizers on function approximation with noisy data in PyTorch.

Practical PyTorch tips covering parameter initialization, model.eval(), tensor views, and the training loop.