Master Thesis – Designing Ternary Neural Networks for Edge Vision
Core
Design and train ternary neural networks (weights: -1, 0, +1) for edge-vision object detection to optimize resource usage on embedded hardware.
Role type
Master's thesis researcher (Computer Vision / Model Compression)
Builds
Ternary neural network models optimized for edge devices
Domain
Computer Vision, Embedded Systems, Model Compression
Deliverable
production ML models
Required skills
Deep learning, Computer vision, Model compression, Quantization-aware training, Python programming, PyTorch
Preferred skills
Embedded systems, Computer architecture, Mixed-precision design
Technologies
PyTorch, Python, Edge hardware, INT8 baselines
Responsibilities
Investigate training methods for ternary weights, adapt network architectures, benchmark models against higher-precision baselines, evaluate trade-offs in accuracy, latency, and energy use
Seniority
Master's student (Research)
