Biomimetic Visual Information Spatiotemporal Encoding Method Yields Promising Results

Strategic Insights -

Investigators at the Chinese Academy of Sciences have developed a novel biomimetic visual information spatiotemporal encoding method based on improved delayed phase encoding. This method has shown significant potential in integrating in vitro biological neural networks (BNNs) with robotic systems for information processing and adaptive learning in practical tasks. The researchers propose a biomimetic visual information spatiotemporal encoding method that transforms high-dimensional images into pulse sequences through convolution, temporal delay, alignment, and compression for BNN stimuli. The method was tested through three stages of unsupervised training on in vitro BNNs using high-density microelectrode arrays (HD-MEAs) and achieved an image recognition accuracy of 80.33% ± 7.94%, a 13.64% improvement over the first training stage.

Key Takeaways:

  • The researchers propose a new biomimetic visual information spatiotemporal encoding method based on improved delayed phase encoding, which transforms high-dimensional images into pulse sequences for BNN stimuli.
  • The method was tested through three stages of unsupervised training on in vitro BNNs using HD-MEAs, achieving an image recognition accuracy of 80.33% ± 7.94%.
  • The method demonstrated a 13.64% improvement in image recognition accuracy over the first training stage.
  • The BNNs exhibited significant increases in the connection number, connection strength, and inter-module participation coefficient after unsupervised training.
  • The proposed method significantly enhances the functional connectivity and cross-module information exchange in BNNs.
  • The research was supported by the Open Project of The National Key Laboratory of Human-computer Hybrid Augmented Intelligence, Cas Project For Young Scientists in Basic Research, Shenyang Science And Technology Innovation Talent Program For Middle-aged And Young Scholars, and other funding sources.
  • The research team consisted of Xingchen Wang, Bo Lv, Fengzhen Tang, Yukai Wang, Bin Liu, and Lianqing Liu from the State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences.

Statistics:

  • 80.33% ± 7.94%: Image recognition accuracy achieved by the proposed method after three stages of unsupervised training.
  • 13.64%: Improvement in image recognition accuracy over the first training stage.
  • 3: Number of stages of unsupervised training conducted on in vitro BNNs using HD-MEAs.

Sources:

  • Biomimetic Visual Information Spatiotemporal Encoding Method for In Vitro Biological Neural Networks. Biomimetics, 2025,10(6):359. (Biomimetics - http://www.mdpi.com/journal/biomimetics)
  • NewsRx. Chinese Academy of Sciences Researchers Yield New Study Findings on Biomimetics (Biomimetic Visual Information Spatiotemporal Encoding Method for In Vitro Biological Neural Networks). Biotech Week. July 9, 2025; p 43.