Research

We make it radically easier for engineers to adopt machine learning techniques by deeply integrating machine learning into Facebook’s programming languages, developer tooling, and infrastructure.

Team focus areas

Programming languages and modeling techniques for machine learning

• Differentiable programming: Producing a new platform for deep learning and differentiable programming by integration into a statically typed language such as Kotlin
• Probabilistic programming: Producing a universal programming language for interpretable and composable Bayesian modeling and inference
• Uncertainty: Increasing the robustness, trustworthiness, and efficiency of machine learning systems by ascribing uncertainty measurements to models

Machine learning for people

• Software artifacts insights: Automating routine and tedious work from our developers’ workflow by learning from software engineering artifacts
• Code assistance: Minimizing the latency between idea and implementation by AI-based autocomplete and code synthesis
• Automated better engineering: Aiming for a self-healing codebase that learns from improvements made by its human masters, such as machine learning-based regression and fault prevention and detection; leveraging machine learning to provide better monitoring, problem detection, and alerting for service owners

Machine learning for systems

• Smart decisions: Enabling fully customized experiences by continuously adapting and auto-tuning service and product parameters to improve the effectiveness and operational efficiency of our infrastructure and products
• Privacy, integrity, reliability: Building a virtualized version of Facebook populated with virtual users to increase reliability, integrity, and privacy of the real Facebook via simulation