Participated in a research project “Automated Design of Hardware Accelerators for Resource-Aware Machine Learning”. The goal of work is to improve an existing methodology for designing machine-learning models (e.g., neural networks). In the first part of the internship, studied the given topic and conducted experiments with neural architecture search (NAS) using genetic programming, automated methods (e.g., AutoML, AdaNet, etc.), and differentiable NAS (e.g., DARTS). The goal of these… Show more Participated in a research project “Automated Design of Hardware Accelerators for Resource-Aware Machine Learning”. The goal of work is to improve an existing methodology for designing machine-learning models (e.g., neural networks). In the first part of the internship, studied the given topic and conducted experiments with neural architecture search (NAS) using genetic programming, automated methods (e.g., AutoML, AdaNet, etc.), and differentiable NAS (e.g., DARTS). The goal of these preliminary experiments is to identify the possible improvements or modifications to the chosen NAS algorithm in order to obtain a resource-aware variant. It is expected to applying some of the modern concepts (e.g., approximate computing, Capsule Neural Networks) in the resource-aware NAS. Then implement the proposed modifications to existing NAS frameworks and investigate the impact on the accuracy-energy tradeoffs. Show less