Machine Learning Engineer On-Device Control and Optimization

The Energy Tech org builds systems for managing the energy flow of Apple devices in service of a great user experience. Within this org the team develops end-to-end solutions utilizing on-device machine learning and control creating new techniques from data analysis and prototyping. Our work directly impacts the behavior of Apple devices across the product families.n

DescriptionnWe are developing on-device control systems that manage power and energy tradeoffs on Apple devices. This means building models that capture device dynamics designing cost functions that encode explicit priorities and shipping control loops that adapt to real-world looking for a Machine Learning Engineer who can work across the full stack: analyzing field data to understand device behavior prototyping control and ML algorithms and getting them running on-device. The problems are messy noisy sensors changing hardware competing objectives and the solutions need to be simple enough to ship on constrained hardware.n

Dig into raw device logs and field data to build understanding of device behavior find opportunities and validate modelsnModel device power and energy dynamics using lab and field datanDevelop and evaluate ML and control systems for on-device managementnRapidly prototype end-to-end systems from data analysis to device deployment collaborating with firmware hardware and platform teamsn

MS or PhD in controls robotics electrical engineering computer science or other quantitative field or BS with relevant experiencenExperience with model predictive control optimal control or reinforcement learning (sequential decision-making)nExperience working from raw logs or sensor data comfortable building analysis from scratchnStrong Python skills; demonstrated ability to take a project from data exploration through working prototype

Experience with thermal systems battery management or energy optimizationnFamiliarity with embedded or resource-constrained environmentsnHands-on ML experience training models evaluating tradeoffs iterating on approaches rather than applying off-the-shelf solutionsnComfort with ambiguity able to scope and drive work without detailed specificationsnTrack record of shipping models or control systems into production not just research