Master Thesis - Approximative Computing for CNNs
Core
Investigate how approximate arithmetic circuits affect the accuracy and hardware efficiency of CNNs for object detection and semantic segmentation.
Role type
Master's thesis researcher (hardware-aware ML)
Builds
Simulation results and analysis of trade-offs between network accuracy and power/area savings
Domain
Edge AI, approximate computing, digital design
Deliverable
research
Required skills
Python, deep learning frameworks (PyTorch/Tensorflow), CNNs, digital design, computer architecture, computer arithmetic
Preferred skills
Object detection/semantic segmentation models, neural network quantization, hardware description languages (Verilog/VHDL), C/C++, CUDA
Technologies
PyTorch, Tensorflow, Verilog, VHDL, CUDA
Responsibilities
Evaluate approximate multipliers on detection/segmentation performance, quantify accuracy vs. power/area trade-offs, explore accuracy recovery methods (layer-wise selection, calibration, fine-tuning)
Seniority
Master's student (Thesis Worker)
