In this paper, we propose a Flexible Coded Distributed Convolution Computing (FCDCC) framework to enhance the tolerance against straggler nodes that cause delays and to improve the numerical stability when deploying CNNs in resource-constrained environments. We extend the existing Coded Distributed Computing (CDC) for matrix multiplication with Circulant and Rotation Matrix Embedding (CRME) and apply it to high-dimensional tensor convolutions. The proposed technique, called Numerically Stable Coded Tensor Convolution (NSCTC), presents two new coding partitioning techniques: Adaptive-Padding Coded Partitioning (APCP) for input tensors and Kernel-Channel Coded Partitioning (KCCP) for filter tensors. These strategies enable linear decomposition and encoding of tensor convolutions into CDC subtasks, and combine model parallelism and coded redundancy for robust and efficient execution. Theoretical analysis is used to find the optimal trade-off between communication and storage costs, and experimental results verify the computational efficiency, straggler resistance, and scalability of the framework on various CNN architectures.
