The University of Science has developed a binarised neural network scheme with ternary gradients, tackling the computational challenges of IoT edge devices while promising enhanced AI capabilities.
The breakthrough allows for efficient, smaller wearable devices that do not need constant cloud connectivity, improving performance in health monitoring and smart home applications.
The researchers introduced a magnetic RAM-based computing-in-memory architecture that reduces circuit size and power consumption, achieving near-identical accuracy and faster training compared to traditional BNNs.
The team highlighted that integrating AI with IoT edge devices remains a challenge due to the substantial computational resources required by artificial neural networks.
Collection
[
|
...
]