Resource constraints have restricted several EdgeAI applications to machine
learning inference approaches, where models are trained on the cloud and
deployed to the edge device. This poses challenges such as bandwidth, latency,
and privacy associated with storing data off-site for model building. Training
on the edge device can overcome these challenges by eliminating the need to
transfer data to another device for storage and model development. On-device
training also provides robustness to data variations as models can be retrained
on newly acquired data to improve performance. We, therefore, propose a
lightweight EdgeAI architecture modified from Xception, for on-device training
in a resource-constraint edge environment. We evaluate our model on a PCB
defect detection task and compare its performance against existing lightweight
models – MobileNetV2, EfficientNetV2B0, and MobileViT-XXS. The results of our
experiment show that our model has a remarkable performance with a test
accuracy of 73.45% without pre-training. This is comparable to the test
accuracy of non-pre-trained MobileViT-XXS (75.40%) and much better than other
non-pre-trained models (MobileNetV2 – 50.05%, EfficientNetV2B0 – 54.30%). The
test accuracy of our model without pre-training is comparable to pre-trained
MobileNetV2 model – 75.45% and better than pre-trained EfficientNetV2B0 model –
58.10%. In terms of memory efficiency, our model performs better than
EfficientNetV2B0 and MobileViT-XXS. We find that the resource efficiency of
machine learning models does not solely depend on the number of parameters but
also depends on architectural considerations. Our method can be applied to
other resource-constraint applications while maintaining significant
performance.
EdgeAI, the integration of artificial intelligence (AI) capabilities on edge devices, holds immense potential for various applications. However, due to resource constraints, many EdgeAI applications have been limited to using machine learning inference approaches. These approaches involve training models in the cloud and deploying them to edge devices, which poses challenges such as bandwidth, latency, and privacy concerns.
To overcome these challenges, on-device training has emerged as a promising solution. By enabling training directly on the edge device, the need to transfer data to another device for storage and model development is eliminated. This not only addresses bandwidth and latency issues but also enhances privacy by keeping the data on the edge device.
The proposed lightweight EdgeAI architecture, based on modifications to the Xception model, leverages on-device training in resource-constrained edge environments. This architecture enables robustness to data variations by allowing models to be retrained on newly acquired data, thereby improving performance. The multi-disciplinary nature of this concept becomes evident as it combines expertise from AI, edge computing, and resource optimization.
In order to evaluate the effectiveness of this approach, the model is tested on a PCB defect detection task. The performance of the proposed model is compared against existing lightweight models such as MobileNetV2, EfficientNetV2B0, and MobileViT-XXS.
The results of the experiment demonstrate that the proposed model achieves remarkable performance with a test accuracy of 73.45% even without pre-training. This performance is comparable to the test accuracy of non-pre-trained MobileViT-XXS (75.40%) and significantly better than other non-pre-trained models (MobileNetV2 – 50.05%, EfficientNetV2B0 – 54.30%). The test accuracy of the proposed model without pre-training is also similar to that of a pre-trained MobileNetV2 model (75.45%) and superior to a pre-trained EfficientNetV2B0 model (58.10%).
Additionally, the proposed model showcases better memory efficiency compared to EfficientNetV2B0 and MobileViT-XXS. It highlights that resource efficiency in machine learning models is not solely determined by the number of parameters, but also depends on architectural considerations.
This research has far-reaching implications for various resource-constrained applications. By demonstrating significant performance while operating in limited resource environments, the proposed method opens up avenues for wider application of EdgeAI. It also emphasizes the importance of considering architectural design choices in resource optimization efforts, showcasing the multi-disciplinary aspects of such endeavors.