At Edgecortix we are a deep-tech startup revolutionizing edge computing with artificial intelligence and novel high efficiency silicon on chip design. Originating from multiple years of research, our unique AI hardware & software co-design principle and the Dynamic Neural Accelerator ® AI processor IP are geared towards positively disrupting the rapidly growing artificial intelligence edge hardware space and bring the power of AI and machine learning to all kinds of devices. Our operations are headquartered in Tokyo, Japan, with offices in Singapore, Virginia, and California in the US.
As an engineering driven company we are working to define and solve the hardest problems in AI including computer vision, speech, and natural language, geared towards real-time capabilities on small to medium form factor devices. We originated out of multiple years of research, as such at our core we value learning, intellectual curiosity, and self-starters. We have the ambitious goal of enabling cloud-level performance with significantly better energy-efficiency for AI inference at the edge.
This role is part of the EdgeCortix’s SAKURA AI Chip team
Work closely with our compiler and hardware engineers to accelerate the adoption of state-of-the-art neural networks on EdgeCortix's proprietary hardware and software stack. Engage in customer-oriented activities guiding deploying of custom and out-of-the-shelf models to meet accuracy and performance targets of customer’s AI system. Grow and maintain a collection of ready-to-use models for an ever-growing set of perception tasks. Work closely with our compiler team to continuously improve stability and user-experience of our software stack.
In this role, you are expected to be an ML expert providing valuable insights and feedback to our architecture team shaping the development of new features. Engage in studies on how various hardware features such as pruning, quantization and mixed-precision impact various aspects of neural network performance across a diverse set of tasks. Stay up to date with common operators and architectural patterns in neural networks from bleeding-edge research to industry-standard models and drive their adoption on EdgeCortix’s stack.
Your day-to-day activities will include but are not limited to dealing with existing neural network implementations, implementing models from scratch, training, quantization, pruning, deployment of models on EdgeCortix hardware, committing accuracy studies, reviewing academic literature, working with public and proprietary vision datasets, constantly communicating with hardware and compiler teams. You will use industry-standard PyTorch and TensorFlow frameworks and do a lot of Python programming.