Ph.D. in Computer Science, Electrical & Computer Engineering, Physics, or a related technical field with strong publication record.
Deep, hands-on expertise in high-performance computing (HPC) and parallel programming models (e.g. CUDA, MPI, OpenMP).
Expert-level proficiency in C++ for performance-critical development.
Experience in low-level systems optimization and performance profiling.
Background in optimizing numerical algorithms, computational mechanics, or compiler technologies is a plus.
Responsibilities
Re-engineer fundamental building blocks of complex ML models for substantial performance gains.
Design and implement novel numerical algorithms in C++ and CUDA to enhance model training and inference.
Profile and analyze deep learning workloads to identify and resolve performance bottlenecks across the system.
Collaborate with quantitative researchers and engineers to develop next generation ML systems.
Stay updated with cutting edge research in HPC, computer architecture, and ML systems.
Benefits
Hybrid work model requiring 3 days in the office per week.
Access to an elite technology and research group.
Autonomy to dive deep into the tech stack and innovate.
Opportunities to collaborate with world-class experts in the field.
Resources provided to push the boundaries of ML systems and hardware implementation.
Full Job Description
Research Scientist, ML Systems (HPC & Systems Optimization)
New York, NY (Hybrid, 3 days in office)
Highly competitive compensation package
The Role
We are seeking a specialist with a Ph.D. for a unique role that sits at the deep intersection of ML algorithms and high-performance hardware, much in the vein of researchers like Tri Dao. This is not a typical ML position. It's a role for a true systems builder who can optimize the core computational mechanics of complex models through low-level, hardware-aware development. You will have the autonomy and resources to dive deep into the stack, profile performance bottlenecks, and write highly optimized code to push the boundaries of what's possible on the latest hardware.
Responsibilities
Re-engineer the fundamental building blocks of complex machine learning models to achieve massive performance gains.
Design and implement novel numerical algorithms in C++ and CUDA to accelerate model training and inference.
Profile and analyze deep learning workloads to identify and solve non-obvious performance bottlenecks across the entire system, from the CPU to the GPU and interconnect.
Collaborate with world-class quantitative researchers and engineers to co-design and implement the next generation of ML systems and infrastructure.
Stay at the cutting edge of academic and industry research in HPC, computer architecture, and ML systems.
Who we're looking for
A Ph.D. in Computer Science, Electrical & Computer Engineering, Physics, or a related technical field with a strong publication record.
Deep, hands-on expertise in high-performance computing (HPC) and parallel programming models (e.g. CUDA, MPI, OpenMP).
Expert-level proficiency in C++ for performance-critical development.
Demonstrable experience in low-level systems optimization, performance profiling, and identifying hardware bottlenecks (e.g. memory bandwidth, latency).
A background in optimizing numerical algorithms, computational mechanics, or compiler technologies is a significant plus.
