Assess graduate-level ability to apply model compression and quantization techniques to optimize neural network inference on edge devices. Scope includes pruning, weight and activation quantization (PTQ, QAT, int8, mixed-precision), knowledge distillation, architecture and operator-level optimizations (batch‑norm folding, per-tensor vs per-channel scaling), calibration, hardware-aware trade-offs (latency, memory, accuracy), and practical tooling/profiling (TensorFlow Lite, ONNX, PyTorch Mobile); tasks require selecting methods, designing workflows, and analyzing evaluation metrics and common pitfalls.