Backend Engineer, Deep Render

• At a minimum, a Bachelor's degree in computer science or related field (Mathematics, Physics, Engineering)
• At a minimum, 3-5 years of experience in performance optimisation
• Formal training could come through education, work experience and/or extensive private projects
• Expertise in C++
• Some experience with optimisation techniques. Examples include SIMD (SSE, AVX), vectorisation, loop dependencies, multithreading, multi-processor usage, and tensor cores
• The ideal candidate will have a deep understanding of optimisation methodologies to reduce runtime and memory footprint, preferably for neural networks; and/or experience implementing high-performance entropy coding algorithms such as Huffman Coding, Arithmetic Coding, Range Coding, or Asymmetric numeral systems
• The ideal candidate will have some experience with taking algorithms from research to deployment

Desirable

• We’re looking for Performance Engineers to join us to be part of a highly talented team bringing the next step-change in compression technology to billions of users, creating enormous global impact and value
• We're looking for engineers who will help us deliver AI-based compression to end-users by porting our codec from GPU-based systems to (primarily) mobile platforms with NPUs and (secondarily) mid-range GPU/CPU systems, thus going from research to production
• You’ll enjoy working with low-level code and are comfortable with programming across multiple platforms
• Work in a team to port ML research algorithms to edge devices with an initial focus on smartphones (Android, iOS)
• Profile various algorithms to analyse performance and identify any bottlenecks. Profiling includes data loading, data movement, data caching, operation count, execution chipset, warm-up latency and others
• Implement solutions to the identified bottlenecks
• Implement a high-performance entropy coding algorithm, e.g. Range Coding or Asymmetric Numeral Systems, across different hardware architectures
• Optional: Write custom operations using the low-level API for Android (OpenGL ES) and iOS (Metal) systems
• Optional: Apply standard neural network runtime optimisation methods such as pruning, low-bit quantisation, architecture tuning, batching and others