Low precision training achieves win-win in efficiency and accuracy
Peer-Reviewed Publication
This month, we’re focusing on artificial intelligence (AI), a topic that continues to capture attention everywhere. Here, you’ll find the latest research news, insights, and discoveries shaping how AI is being developed and used across the world.
Updates every hour. Last Updated: 14-Dec-2025 09:11 ET (14-Dec-2025 14:11 GMT/UTC)
Decreasing Precision with layer Capacity trains deep neural networks with layer-wise shrinking precision, cutting cost by up to 44% and boosting accuracy by up to 0.68%
Researchers distill multi-teacher common knowledge + gradient alignment into one substitute model
CAD-NeRF rebuilds 3D scenes from < 10 images via ShapeNet CAD retrieval + self-supervised pose-density joint optimization
Balance training patients may soon be able to get AI feedback during home exercises, with four wearable sensors and a new machine learning model developed at the University of Michigan.
In this paper, the existing AD methods for the PMSM drive system with LC sine wave filter are reviewed, including the modified AD methods based on inherent damping, conventional AD methods based on state variable feedback, modified AD methods with LPF and HPF based on state variable feedback and AD methods based on digital filter. A new expansion of AD method based on HPF-CCF is studied to ensure the effectiveness when the resonant frequency is around sixth of the sampling frequency. The stability, dynamic performance, robustness, and algorithm complexity of the AD methods are compared in detail and the suggestion of selecting the AD method in different industrial scenarios is summed as below.
1) When evaluating the stability of control system in terms of PM and GM, CCF, LPF-CCF, and the proposed HPF-CCF are comparatively more recommended.
2) In terms of the open-loop cutoff frequency, the proposed HPF-CCF is more recommended for realizing a better dynamic performance.
3) In terms of the Bode diagrams analysis and experimental results, LPF-CCF, HPF-MCF, and the proposed HPF-CCF are more recommended for ensuring control system robustness.
4) When considering the algorithmic complexity of the AD methods, only one parameter needs to be designed for CCF and ICF-SOGI.