Presented By O'Reilly and Cloudera
Make Data Work
Dec 4–5, 2017: Training
Dec 5–7, 2017: Tutorials & Conference
Singapore

Bootstrap custom image classification using transfer learning

Danielle Dean (Microsoft), Wee Hyong Tok (Microsoft)
12:05pm12:45pm Wednesday, December 6, 2017
Average rating: ***..
(3.25, 4 ratings)

Who is this presentation for?

  • Data scientists and developers

Prerequisite knowledge

  • A basic understanding of machine learning and deep learning

What you'll learn

  • Understand the basics of convolution neural networks (CNNs) and how you can leverage transfer learning to bootstrap your use of deep learning
  • Learn how to use the pretrained models in Microsoft Cognitive Toolkit (CNTK) and TensorFlow to build a custom image classifier

Description

Convolution neural networks (CNNs) have been used in many image classification tasks and are usually trained on large image datasets, such as ImageNet and CIFAR. CNNs have been shown to be very effective in extracting the features of images from diverse domains.

In practice, most people do not train a CNN from scratch, due to time constraints. Transfer learning enables you to use pretrained deep neural networks (e.g., AlexNet, ResNet, and Inception V3) and adapt them for custom image classification tasks. Danielle Dean and Wee Hyong Tok walk you through the basics of transfer learning and demonstrate how you can use the technique to bootstrap the building of custom image classifiers using pretrained CNNs available in various deep learning toolkits (e.g., pretrained CNTK models, Caffe Model Zoo, and pretrained TensorFlow libraries).

Topics include:

  • Using pretrained CNNs and the last fully connected layer as a feature extractor: Once the features are extracted, any existing classifier (SVMs, decision trees, LightGBM, etc.) can be used for image classification, using the extracted features as inputs.
  • Using backpropagation to further tune the weights of the pretrained CNN and adapt it for the new set of images.
  • A quick introduction to deep neural network architecture
Photo of Danielle Dean

Danielle Dean

Microsoft

Danielle Dean is a principal data scientist lead in AzureCAT within the Cloud AI Platform Division at Microsoft, where she leads an international team of data scientists and engineers to build predictive analytics and machine learning solutions with external companies utilizing Microsoft’s Cloud AI Platform. Previously, she was a data scientist at Nokia, where she produced business value and insights from big data through data mining and statistical modeling on data-driven projects that impacted a range of businesses, products, and initiatives. Danielle holds a PhD in quantitative psychology from the University of North Carolina at Chapel Hill, where she studied the application of multilevel event history models to understand the timing and processes leading to events between dyads within social networks.

Photo of Wee Hyong Tok

Wee Hyong Tok

Microsoft

Wee Hyong Tok is a principal data science manager with Microsoft. Wee Hyong has worn many hats in his career, including developer, program and product manager, data scientist, researcher, and strategist, and his track record of leading successful engineering and data science teams has given him unique superpowers to be a trusted AI advisor to customers. Wee Hyong coauthored several books on artificial intelligence, including Predictive Analytics Using Azure Machine Learning and Doing Data Science with SQL Server. Wee Hyong holds a PhD in computer science from the National University of Singapore.