Responsibilities:
· Lead research into state-of-the-art optimization techniques, including Quantization-Aware Training (QAT), Pruning, Knowledge Distillation, and Neural Architecture Search (NAS) to minimize latency.
· Design and implement scalable AI deployment architectures that can handle high-throughput data streams from multiple high-resolution cameras and process sensors simultaneously.
· Conduct hardware-software co-design to optimize models for specific deployment targets (e.g., NVIDIA Jetson, TensorRT, FPGAs, or specialized AI accelerators).
· Develop and manage asynchronous data pipelines that ensure zero-bottleneck performance from image acquisition to "final sentencing" decisions.
· Establish rigorous performance profiling benchmarks to track model latency and memory footprint across various manufacturing environments.
· Work with the System Integrator (SI) to ensure that optimized models are seamlessly integrated into the factory-level software stack.
JOB REQUIREMENTS
· Ph.D. in Computer Engineering, Computer Science, Electrical Engineering, or a related field with a focus on High-Performance AI.
· At least 3 years of experience
· Deep understanding of AI Inference Engines (e.g., TensorRT, ONNX Runtime, OpenVINO).
· Mastery of Model Compression techniques (Pruning, Quantization, Distillation).
· Expertise in C++ and Python for high-performance implementation.
· Hands-on experience with Parallel Computing (CUDA, OpenCL).
· Familiarity with Mixed-Precision Training and FP16/INT8 deployment.
· Proven ability to architect end-to-end AI systems that balance the trade-off between throughput, latency, and model precision.
The above eligibility criteria are not exhaustive. A*STAR may include additional selection criteria based on its prevailing recruitment policies. These policies may be amended from time to time without notice. We regret that only shortlisted candidates will be notified.
