How to simplify ResNet50 in a classification task?
Machine Learning / AI
What practical tricks speed up transformer inference (speculative decoding, medusa)?
How to increase the amount of training data with a small number of labeled objects?
What is attention?
How to recognize text in images using NLP?
Was dropout implemented based on loss or by randomly dropping weights?
How to implement an algorithm for counting and shifting zeros in an array in Python?
What is the difference between level-wise and leaf-wise tree training?
Is it necessary to put shuffle in DataLoader?
What types of model distillation exist?
What is a partition in ClickHouse?
When should you choose backbones, and when can you use ResNet50?
What is distillation?
What errors in the training cycle can worsen the learning and quality of the model?
What are EFB and GOSS in LightGBM?
Is YOLOv8 a one-stage or two-stage detector?