Netflix Summer 2022 Machine Learning Internships - ML Areas
Causality and incrementality lie at the heart of many product and business decisions at Netflix. Either when we only have access to observational data or when we have collected our data via adaptive learning algorithms (e.g. via contextual bandits which are widespread in recommender systems), we have to employ causal modeling in order to draw reliable inferences and make the right decisions for our members and for the product overall. This requires bringing together machine learning algorithms and causal statistical theory for principled decision-making. In this area, you will be able to work on applied and theoretical projects that advance the synergy of causal inference and machine learning at Netflix.
The amazing content library we love on Netflix comes together due to numerous creative & technical decisions made just the right way: deciding what content to commission, how to produce it, and how to promote it are all critical questions that require a deep & thorough understanding of existing and upcoming content. Join us if you are excited about using the latest Computer Vision, Audio ML, Graphics, and NLP techniques to help characterize our content, to help artists during the production process, as well as to help synthesize creative artifacts that connect our titles to our members.
ML-powered systems that interact with people (e.g. recommendation systems) must be responsible actors designed to avoid undesirable biases, promote people's best interests, be open to introspection, and generally behave in a socially responsible way. Ensuring this adequate behavior poses huge research and engineering challenges given the complexity and scale of these systems, but is key for the future of AI.
Netflix operates very large systems, to enable many teams (including ML teams) to run workloads with significant scale. Operating such workloads efficiently requires new Machine Learning research, at the intersection of many different sub-fields, such as supervised learning, reinforcement learning, systems, and optimization.
The ML Platform at Netflix is responsible for maximizing the impact of all machine learning practitioners at Netflix. The platform comprises systems, solutions, libraries, and services that touch all aspects of the full machine learning lifecycle from data access, feature engineering, model development, tuning & evaluation, and model inference & serving. Working in this area you will get an opportunity to explore key engineering problems that can advance the ML Platform in innovation areas like AutoML or Bandits & Reinforcement Learning infrastructure.
Recommendation algorithms are at the core of the Netflix product. They provide our members with personalized suggestions to make it easy for them to find something they’ll enjoy watching. We continually seek to improve recommendations by advancing the state-of-the-art in the field, which involves addressing many open challenges — e.g. dealing with complex member preferences, handling partially observed feedback, or understanding what makes a great vs a good recommendation.
Because recommender systems can only offer up a limited set of suggestions in each session and receive very sparse feedback, it is natural to look at them through the lens of bandits and reinforcement learning. Making these approaches work well within the personalization domain, however, has unique challenges due to the very large and evolving action spaces, high-dimensional state spaces, and need for off-policy training and evaluation. Thus, we are conducting research in this area to push forward what is possible and develop better algorithms for long-term optimization.
Search is a critical part of content discovery on the Netflix app. Being domain-specific, real-time, interactive, multi-lingual, global, and available across a wide variety of user interfaces, it presents some unique and challenging research problems. Conversational search/recommendations give users multiple turns to refine their entertainment needs and are increasingly desired by users. Satisfying users' entertainment needs via traditional search or via conversational agents requires creative application of the latest research from Machine Learning, Search & Recommendations, Knowledge Graphs, Natural Language Processing, and Human-Computer Interaction.