A new algorithm will allow AI to build routes more efficiently
Researchers at HSE University in Russia have enhanced the ability of neural networks to orient themselves in space by introducing an attention mechanism. Mathematicians found that adding an algorithm functioning similarly to human attention improves the capability of neural networks by 6-15%, making them more efficient in finding optimal trajectories for robots.
The developed training method using an attention mechanism is aimed at increasing the efficiency of graph-based supergraph neural networks. These algorithms are well-suited for creating artificial intelligence capable of optimally navigating complex environments. They accelerate the learning process and make neural networks more adaptable to changing environmental conditions.
Experiments conducted on navigation tasks confirmed the effectiveness of the proposed method. Scientists highlight the potential of this approach for delivery systems and robots capable of spatial orientation.
Photo: Freepik