Researchers from Bar-Ilan University have published findings in Scientific Reports that suggest improving AI by making global decisions, rather than just adding more layers to deep learning (DL) architectures, can enhance performance significantly.

The study, led by Prof. Ido Kanter, shows that choosing the most influential paths to outputs in DL systems can outperform traditional methods that rely on local decisions at each layer. This approach is likened to using a map to see the entire route to a destination, rather than making decisions at each intersection, which can lead to more efficient and effective learning processes. This method could allow AI systems to perform better on classification tasks without the complexity of additional layers.

The research also ties into broader efforts to bridge machine learning with biological insights, suggesting that understanding and mimicking biological processes can further enhance AI capabilities.

Click here if you would like to read the full article.