New AI Technology Uses 1% Of The Power Of Current Technology! New Paper: Increases AI Efficiency Using Human Brian-Like Technology Topographical Sparse Mapping Explained AI models are growing larger and using more energy, raising concerns about sustainability. A new paper from the University of Surrey, published in Neurocomputing, introduces a method inspired by the human brain to make AI more efficient. Titled "Topographical Sparse Mapping: A Neuro-Inspired Sparse Training Framework for Deep Learning Models," it shows how to reduce connections in neural networks while keeping or improving accuracy. Training big AI models can cost millions and use as much power as hundreds of homes. By 2030, data centers may match a country's energy use. Researchers Mohsen Kamelian Rad, Ferrante Neri, Sotiris Moschoyiannis, and Bauer drew from the brain's organized wiring in the visual system. They created Topographical Sparse Mapping (TSM), which connects input features, like image pixels, only to nearby neurons in the next layer. This cuts unnecessary links from the start, speeding up training and improving results. They also developed Enhanced TSM (ETSM), which prunes weak connections during training, like the brain refining pathways. This leads to 99 percent sparsity—removing most links—while matching or beating dense models on tests like MNIST (handwritten digits), Fashion-MNIST (clothing), CIFAR-10 (simple images), and CIFAR-100 (detailed categories). On CIFAR-100, ETSM improved accuracy by 14 percent over other sparse methods, using fewer resources. The key benefit is energy savings: These models use less than 1 percent of the power of standard ones, with faster training and lower memory needs. This could help build brain-like hardware and reduce AI's environmental impact. Key points… 1. Brain-inspired design: TSM limits connections to local neurons, reducing waste and aiding generalization without extra steps. 2. Pruning in ETSM: Removes low-value links during training, reaching 99 percent sparsity without losing accuracy. 3. Strong results: Matches or exceeds dense models on benchmarks, with big gains on complex tasks like CIFAR-100. 4. Efficiency gains: Cuts energy to under 1 percent of norms, speeds training, and lowers memory use. 5. Future potential: Could apply to other AI types, supporting sustainable growth and new hardware. This work points to a shift: Smarter structures may outperform bigger models, aiding AI's long-term development. I am beginning to build these ideas out and will test them. It is really interesting. Citations: 1. sciencedirect.com/science/ar…