Browsing by Author "Zhang, Jun"

Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemOpen Access
    Machine learning-enabled globally guaranteed evolutionary computation
    (Nature Publishing Group, 2023-04-10) Li, Bin; Wei, Ziping; Wu, Jingjing; Yu, Shuai; Zhang, Tian; Zhu, Chunli; Zheng, Dezhi; Guo, Weisi; Zhao, Chenglin; Zhang, Jun
    Evolutionary computation, for example, particle swarm optimization, has impressive achievements in solving complex problems in science and industry; however, an important open problem in evolutionary computation is that there is no theoretical guarantee of reaching the global optimum and general reliability; this is due to the lack of a unified representation of diverse problem structures and a generic mechanism by which to avoid local optima. This unresolved challenge impairs trust in the applicability of evolutionary computation to a variety of problems. Here we report an evolutionary computation framework aided by machine learning, named EVOLER, which enables the theoretically guaranteed global optimization of a range of complex non-convex problems. This is achieved by: (1) learning a low-rank representation of a problem with limited samples, which helps to identify an attention subspace; and (2) exploring this small attention subspace via the evolutionary computation method, which helps to reliably avoid local optima. As validated on 20 challenging benchmarks, this method finds the global optimum with a probability approaching 1. We use EVOLER to tackle two important problems: power grid dispatch and the inverse design of nanophotonics devices. The method consistently reached optimal results that were challenging to achieve with previous state-of-the-art methods. EVOLER takes a leap forwards in globally guaranteed evolutionary computation, overcoming the uncertainty of data-driven black-box methods, and offering broad prospects for tackling complex real-world problems.
  • Thumbnail Image
    ItemOpen Access
    Random sketch learning for deep neural networks in edge computing
    (Springer Nature, 2021-03-25) Li, Bin; Chen, Peijun; Liu, Hongfu; Guo, Weisi; Cao, Xianbin; Du, Junzhao; Zhao, Chenglin; Zhang, Jun
    Despite the great potential of deep neural networks (DNNs), they require massive weights and huge computational resources, creating a vast gap when deploying artificial intelligence at low-cost edge devices. Current lightweight DNNs, achieved by high-dimensional space pre-training and post-compression, present challenges when covering the resources deficit, making tiny artificial intelligence hard to be implemented. Here we report an architecture named random sketch learning, or Rosler, for computationally efficient tiny artificial intelligence. We build a universal compressing-while-training framework that directly learns a compact model and, most importantly, enables computationally efficient on-device learning. As validated on different models and datasets, it attains substantial memory reduction of ~50–90× (16-bits quantization), compared with fully connected DNNs. We demonstrate it on low-cost hardware, whereby the computation is accelerated by >180× and the energy consumption is reduced by ~10×. Our method paves the way for deploying tiny artificial intelligence in many scientific and industrial applications.