Replication Study: Neural Network Quantization for Edge Deployment — Field Validation in Resource-Constrained Environments

This study replicates MIT CSAIL's neural network quantization research (Bengio et al., 2018) under real-world field conditions in rural India. While laboratory studies demonstrate the viability of INT8 quantization for edge deployment, few validate these techniques under actual resource-constrained scenarios with intermittent power and variable network conditions. This replication study implements MobileNetV2 quantization following MIT's methodology, achieving 4.0x model compression and 2.4x inference speedup. Testing under field conditions reveals 2-57x higher network latency variance compared to laboratory simulations, while maintaining comparable accuracy (