Earlier this month, Dan Kuster gave a talk discussing why businesses should consider adopting deep learning solutions. Key takeaways include simplicity, accuracy, flexibility, and some hacks for working with the tech.
In this case study, we evaluate four different strategies for solving a problem with machine learning. In terms of both technical performance and practical factors like economics and amount of training data required, customized models built from semi-supervised "deep" features using transfer learning outperform models built from scratch, and rival state-of-the-art methods.
Last week Dan Kuster held a workshop at General Assembly Boston on how machine learning is changing -- and improving -- the way digital marketers do their jobs. Video and slides available for those of you who missed it.
At the beginning of the month, three members of our Advanced Development team attended the International Conference on Learning Representations in Puerto Rico. Luke discusses some key takeaways and his favorite papers.
Luke expands on the data input methods he discussed in Part 1 of this mini blog series, namely highlighting a hybrid approach of those methods that allows for faster training, as well as some extensions to the demo.
TensorFlow is a great new deep learning framework that supports the symbolic construction of functions (similar to Theano) to perform some computation, generally a neural network based model. Luke, one of our machine learning researchers, discusses several methods for feeding data into a machine learning model using this framework.