Data
Science

Discover the evolution of data science workflows implemented at Expedia with a special emphasis on Learning to Rank problems. This session will explore the process of industrializing the data science workflow and best practices on how to keep your data productive, or even pull your organization out of the data swamp.

Journey from exploration and visualization to machine learning and natural language processing. Discover how Return Path built a cloud-based, production ready, enterprise scale data solution without a dedicated Dev Ops team. Leveraging modern distributed computing frameworks like Spark and managed services like EMR and Qubole were key to the process.

In the biological sciences, hypothesis-driven experiments and bottom-up design experiments rely on predicting what will happen with new cells and molecules. Machine learning excels at prediction and has become more democratized, making it an important component in the biotech toolkit. We use Merck's Kaggle competition as a representative task in this domain that involves predicting molecular activity from numeric descriptors of chemical structure. Our approach utilizes deep neural networks using the Keras library in a Qubole notebook, which is conveniently attached to an autoscaled Spark cluster. We use Spark to distribute the hyperparameter search for optimizing the neural net.

This session will cover how Azure Machine Learning and R can help data scientists overcome the following challenges: - Development Time - Dramatically reduce the time of running initial ML experiment validations. - Performance - Option for best in class performance. For deeper data science needs the session will explore how hard core data scientists can leverage R to attack the most complex scenarios. Finally we will explore Python integration on the Microsoft stack and what is new between CNTK and TensorFlow in Azure.

There is a surge in hype around Artificial Intelligence. Startups are raising hundreds of millions of dollars by bedazzling investors with Deep Learning, word embeddings, and reinforcement learning. This is a distraction from the very real problems that data and AI can solve if done right. By working across dozens of machine learning problems that are live in the real world, I’ve worked out the most common problems encountered and recurring design patterns on how to solve real-world problems using AI as a tool. This talk will arm you with a perspective on how to get pragmatic solutions with AI today.