A Practical Introduction to Convnets for Image Classification
|Big Data & Machine Learning|
An introduction to developing a complete pipeline based on convolutional neural networks (CNNs) to solve supervised image classification tasks. We consider the task of classifying hand gestures from images as rock, paper or scissors, and present the whole pipeline discussing pratical issues such as dataset acquisition, splitting training/testing data, CNN architectures, training and evaluating models. The journey will encounter a few setbacks, which introduces important machine learning concepts such as overfitting, generalization ability, dataset augmentation strategies. The talk is concluded by discussing a few spectacular applications of similar techniques in various fields (robotics, industry, biomedicine).
The talk is delivered through slides and a Python Jupyter notebook, using keras and tensorflow as a backend. The main messages, however, are conceptual and language-independent. The talk has no prerequisites.
PhD in Computer Vision at Politecnico di Milano, now senior researcher at the Dalle Molle Institute for Artificial Intelligence. Develops cutting edge applications of AI and Deep Learning to industry, robotics, biomedicine. Data science and dataviz nerd, Pythonista, lecturer at Politecnico di Milano, Zurich University of Applied Sciences, USI and SUPSI.