The first automobile was invented in 1887, and shortly after, the first auto race took place in Paris. AI applications, like cars, are still in their early stages but are expected to have a significant impact. Operator fusion is an optimization technique in AI programs that reduces the costs of successive operators by considering them as a single operator, similar to combining two tasks into one. This technique improves the efficiency and performance of AI programs. NVIDIA’s TensorRT Optimizer is an example of a tool that performs operator fusion. Fusing convolutional layers, bias adds, and activation function layers is a common choice for operator fusion. Different hardware platforms have varying capabilities when it comes to operator fusion.

source update: Unlocking the Power of Operator Fusion to Accelerate AI – Towards AI