TF-Slim is a TensorFlow wrapper that simplifies the creation and training of complex TensorFlow models. It eliminates the need for boilerplate code commonly found in deep learning algorithms. This course is designed for individuals with prior experience in TensorFlow. To get the most out of the training, learners should be familiar with data science theory, including concepts like train/test splittings, overfitting, subfitting, and bias-variance tradingoffs. Additionally, a solid understanding of deep learning theory, specifically backpropagation and weight parameter tensors, is recommended.
Throughout the course, participants will learn how to use the TF-Slim API to create deep learning models that are both readable and manageable. They will also gain proficiency in manipulating variables using TF-Slim wrapper functions. The training will enable learners to quickly experiment with different loss functions, optimizers, and regularizers. Furthermore, participants will develop skills in implementing routings for model evaluation, training, and hyper-parameter tuning.
In addition to these fundamental concepts, participants will also learn how to fine-tune pre-trained models for specific tasks and how to transfer knowledge from one task to another. The course covers building and training feedforward neural systems, as well as creating and training image classification and text classification models.
The course is led by Marvin Bertin, a data scientist at Driver, a San Francisco-based platform that facilitates treatment access for cancer patients. Marvin previously worked as a deep-learning researcher at Skymind, an AI company. He holds degrees in Data Science and Mechanical Engineering and is a frequent speaker at deep learning and machine learning conferences.
In summary, this course provides an in-depth overview of TF-Slim and equips participants with the skills needed to create, train, and fine-tune TensorFlow models for various tasks. It combines theoretical knowledge with practical hands-on experience to ensure learners are well-prepared to tackle real-world deep learning challenges.
