Quantum error mitigation techniques are at the heart of quantum computation. Conventional quantum error correction codes are promising solutions, while they become infeasible in the noisy intermediate scale quantum (NISQ) era, hurdled by the required expensive resources. The variational quantum learning scheme (VQLS), which is composed of trainable quantum circuits and a gradient-based classical optimizer, could partially adapt the noise affect by tuning the trainable parameters. However, both empirical and theoretical results have shown that for most variational quantum algorithms, noise can deteriorate their performances evidently when the problem size scales. Furthermore, VQLS suffers from the barren plateau phenomenon. Here we devise a resource and runtime efficient scheme, i.e., quantum architecture search scheme (QAS), to better improve the robustness and trainability of VQLS. Particularly, given a learning task, QAS actively seeks an optimal architecture among all possible circuit architectures to balance benefits and side-effects brought by adding quantum gates, where more quantum operations enable a stronger expressive power of the quantum model but introduce a larger amount of noise and more serious barren plateau scenario. To this end, QAS implicitly learns a rule that can well suppress the influence of quantum noise and the barren plateau. We implement QAS on both the numerical simulator and real quantum hardware via the IBM cloud to accomplish the data classification and the quantum ground state approximation tasks. Numerical and experimental results exhibit that QAS outperforms conventional variational quantum algorithms with heuristic circuit architectures. Our work provides guidance for developing advanced learning based quantum error mitigation techniques on near-term quantum devices.