Abstract: Supervised machine learning requires large labeled datasets - a prohibitive limitation in many real-world applications. Being able to teach machines with examples is a powerful capability, but it hinges on the availability of vast amounts of data.
First, the data not only needs to exist but has to be in a form that allows relationships between input features and output to be uncovered. Creating labels for each input feature fulfills this requirement but is an expensive undertaking. What if machines could learn with fewer labeled examples?
Second, certain real-world problems have long-tailed and imbalanced data distributions which may mean that it is simply hard to collect training examples in such instances. The talk focuses on a new paradigm - meta-learning. Meta-learning not only learns from a handful of examples but also learns to classify novel classes during model inference.
We explore algorithmic approaches that drive both capability and provide practical guidance for translating this capability into production. We provide intuition for how and why these algorithms work, share our experiments and some of the challenges associated with these techniques.
Bio: Nisha Muktewar is a Research Engineer at Cloudera Fast Forward Labs, where she spends time researching latest ideas in machine learning, builds prototypes that showcase these capabilities when applied to real-world use cases, and advises clients in this space. Prior to joining Cloudera, she worked as a Manager in Deloitte’s Actuarial & Modeling practice leading teams in designing, building, and implementing predictive modeling solutions for pricing, consumer behavior, marketing mix, and customer segmentation use cases for insurance and retail/consumer businesses.