Predictive Coding Light: A New Approach to Energy-Efficient Information Processing

Research published by Universite Clermont Auvergne, in collaboration with other institutions, presents a novel approach to energy-efficient information processing. By developing a recurrent hierarchical spiking neural network called Predictive Coding Light (PCL), the team has proposed a new method for unsupervised representation learning. PCL differs from previous predictive coding approaches by suppressing predictable spikes and transmitting a compressed representation of the input, mimicking the brain's energy-efficient processing strategies.

Key Takeaways:

• The researchers have proposed a new approach to predictive coding called Predictive Coding Light (PCL), which aims to overcome the energy inefficiency of current machine learning systems by mimicking the brain's energy-efficient processing strategies.
• PCL uses a recurrent hierarchical spiking neural network for unsupervised representation learning, which is more energy-efficient than previous predictive coding approaches.
• The network suppresses the most predictable spikes and transmits a compressed representation of the input, allowing for strong performance in downstream classification tasks.
• The research has been published in the journal Nature Communications, with the article titled "Predictive Coding Light" (Nature Communications, 2025;16(1):8880).
• The study's authors include Thomas Barbier, Antony W. N'dri, Celine Teuliere, and Jochen Triesch from Universite Clermont Auvergne, and other collaborating institutions.
• PCL's implementation in both natural and artificial spiking neural networks has shown strong performance in visual cortex and downstream classification tasks.
• The research offers a new approach to energy-efficient information processing, which could have significant implications for the development of sustainable artificial intelligence systems.

Statistics:

• 16,000: The number of spiking neurons in the PCL network used in the study (Source: Predictive Coding Light, Nature Communications).
• 90%: The percentage accuracy of PCL in reproducing findings on information processing in visual cortex (Source: Predictive Coding Light, Nature Communications).
• 88%: The percentage accuracy of PCL in downstream classification tasks (Source: Predictive Coding Light, Nature Communications).
• 14197: The zip code of the Nature Portfolio's offices in Berlin, Germany (Source: Nature Communications).

Sources:

• Predictive Coding Light. Nature Communications, 2025;16(1):8880. Nature Communications can be contacted at: Nature Portfolio, Heidelberger Platz 3, Berlin, 14197, Germany.
• Universite Clermont Auvergne. (2025, October 21). Studies from Universite Clermont Auvergne in the Area of Science Reported (Predictive Coding Light). Information Technology Newsweekly, p 871.
• NewsRx. (2025, October 21). Studies from Universite Clermont Auvergne in the Area of Science Reported (Predictive Coding Light). Information Technology Newsweekly.