1. The document proposes using Number Theoretic Transforms (NTTs) to accelerate quantized convolutional neural networks. NTTs allow fast convolution algorithms by exploiting the finite field of quantized values. 2. The approach applies NTTs with Fermat number transforms (FNTs) to quantized convolution operations. Benchmarks on a Raspberry Pi Zero show speedups compared to a naive convolution implementation. 3. Future work includes implementing SIMD optimizations, mapping operations to finite field matrix multiplication, an FPGA implementation, and techniques like binary segmentation to further accelerate quantized convolutional neural networks.