The feedforward control scheme has proven to be effective in boosting the performance and robustness of control systems. In this paper, we propose a new neural network design concept that leverages feedforward control to improve pattern recognition performance and robustness. Specifically, the feedforward loop is added in parallel with existing blocks and aids the entire network in converging faster, achieving greater accuracy, and being more robust against data distribution shifts without increasing the number of parameters to be trained. We provide a conceptual equivalence of the proposed feedforward network to feedforward control and evaluate its performance on the MNIST handwritten digit image dataset as a feasibility study. Our results demonstrate that the proposed design concept outperforms existing methods in terms of accuracy, convergence speed, and robustness.