TensorFlow-Slim (TF-Slim) is a TensorFlow wrapper library that allows you to build and train complex TensorFlow models in an easy, intuitive way by eliminating the boilerplate code that plagues many deep learning algorithms. This course teaches you how to use TF-Slim and is intended for learners with some previous experience working with TensorFlow.
To get the most out of this training, learners should be familiar with the core concepts of data science theory (train/test splits, overfitting and underfitting, bias-variance tradeoffs, etc.), and deep learning theory (backpropogation, weight parameter tensors, neural network layers, objective and loss functions, and optimization via stochastic descent).
- Learn to build readable and maintainable deep learning models using the TF-Slim API
- Master TF-Slim's wrapper functions for variable creation and manipulation
- Be able to rapidly experiment with loss functions, optimizers, and regularizers
- Learn to implement routings for model training, evaluation, and hyper-parameter tuning
- Understand how to fine-tune a pre-trained model
- Learn how to take a model trained on a specific task and use it for another task
- Discover how to build and train a feedforward neural network
- Gain experience building and training image classification and text classification models
Marvin Bertin is a data scientist with Driver, a San Francisco based treatment access platform for cancer patients. Before that, he worked as a deep learning researcher for the AI company Skymind. Marvin holds degrees in Data Science and Mechanical Engineering, has authored a number of courses on deep learning, and is a speaker at machine learning and deep learning conferences.