CNN Mini-Experiment: How Kernel Size & Filters Affect Accuracy and Training Time (PyTorch) скачать в хорошем качестве

CNN Mini-Experiment: How Kernel Size & Filters Affect Accuracy and Training Time (PyTorch)

This video walks through a hands-on CNN mini-experiment in PyTorch. We train a small convolutional neural network (CNN) and change one architectural setting at a time — kernel size or number of filters — to observe how it affects: • Model parameters • Training time per epoch • Validation accuracy The goal is not perfect modelling, but building engineering intuition. You will see how small architectural changes increase model capacity, affect compute cost, and sometimes improve (or fail to improve) performance. Key ideas covered: – CNN weight sharing and why CNNs are strong baselines for images – How kernel size and filter count impact parameter count – Trade-offs between capacity and generalization – Why accuracy gains may not always justify extra compute This video is part of GenAISA Module 2.3: Observing architectures in action. Engineering lens: After each run, ask: • What changed in parameters and runtime? • How did validation behave? • Was the extra compute worth it? GenAISA project funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Education and Culture Executive Agency (EACEA). Neither the European Union nor EACEA can be held responsible for them.