TLDR: Researchers at the University of Surrey have developed a groundbreaking AI approach, Topographical Sparse Mapping (TSM), that mimics the brain’s neural wiring to create more energy-efficient and faster artificial intelligence systems. This method significantly reduces computational demands and energy consumption while maintaining or improving accuracy, addressing a critical sustainability challenge in the rapidly expanding field of AI.
Guildford, UK – In a significant leap forward for artificial intelligence, researchers from the University of Surrey’s Nature-Inspired Computation and Engineering (NICE) group have unveiled a novel approach that draws direct inspiration from the human brain’s intricate neural networks. Published in the journal Neurocomputing on October 30, 2025, their work introduces a method called Topographical Sparse Mapping (TSM), which promises to make AI systems more energy-efficient and faster without compromising accuracy.
Conventional deep-learning models, widely used in applications like image recognition and language processing, typically connect every neuron in one layer to all neurons in the next. This ‘dense’ wiring, while powerful, leads to substantial energy consumption and computational overhead. The Surrey team’s TSM method, however, rethinks this fundamental architecture by mimicking the brain’s sparse and structured neural wiring. Instead of exhaustive connections, TSM connects each neuron only to nearby or related ones, mirroring the efficient organization seen in the brain’s visual system.
This natural design eliminates a vast number of unnecessary connections and computations. An enhanced version, known as Enhanced Topographical Sparse Mapping (ETSM), further refines this process by incorporating a biologically inspired ‘pruning’ mechanism during training. This pruning is akin to how the brain refines its neural connections as it learns, allowing AI systems to achieve equal or even greater accuracy with a mere fraction of the parameters and energy typically required by conventional models.
The implications for sustainability are profound. The training of many of today’s large AI models can consume over a million kilowatt-hours of electricity, equivalent to the annual usage of more than a hundred US homes, and incur costs in the tens of millions of dollars. This unsustainable trajectory is a growing concern as AI continues its rapid expansion.
Also Read:
- Meta Leverages Advanced AI to Streamline Operations and Drastically Reduce Carbon Emissions
- New AI Personhood Framework Proposed for Governing Agentic Artificial Intelligence
According to the research, Surrey’s enhanced model achieved an impressive sparsity of up to 99%, meaning it could remove almost all of the usual neural connections. Crucially, it still matched or exceeded the accuracy of standard networks on benchmark datasets. Furthermore, by avoiding the constant fine-tuning and rewiring characteristic of other approaches, the TSM and ETSM methods train faster, use less memory, and consume less than one percent of the energy of a conventional AI system. While the current framework applies this brain-inspired mapping to an AI model’s input layer, extending it to deeper layers holds the potential for even leaner and more efficient networks in the future.
