Responsibilities and Opportunities

• Designing a compute library(such as blas, dnn, etc.) composed of various neural network operations, which are being accelerated on the rebellions' proprietary instruction set architecture(ISA)
• From a functionality perspective, enhancing the functional coverage of each operation by considering operation-specific constraints(e.g., tensor shape variation, precision loss handling, etc.)
• From a performance perspective, enhancing the utilization of the computational units in heterogeneous compute resources by considering operation-specific characteristics

Key Qualifications

• Master's or higher degree in Electrical Engineering, Computer Science, or a related field
• Thorough knowledge of neural network operations, not only for the high-level concepts but also for the low-level computation flow
• Excellent troubleshooting, problem-solving, in-depth optimization skills
• Proficiency in programming languages: C++, Python

Ideal Qualifications

• Thorough knowledge of deep learning models for various applications, including vision, language, speech, etc.
• Experience in model/layer-level customization in terms of computation efficiency(e.g., sparsity, reduced precision, layer decomposition, etc.)
• Experience in architecture-specific parallel programming to accelerate target operations(e.g., SSE/AVX in x86, NEON in AArch, CUDA/OpenCL in GPU, etc.)
• A major in computer architecture field is preferred