Source article originally published on
Article Summary and/or Excerpt:
Researchers from Bar-Ilan University in Israel have demonstrated that efficient learning on an artificial tree architecture can achieve better classification success rates than previous deep learning (DL) models. The tree architecture, where each weight has a single route to an output unit, is a step towards a biologically plausible AI and has reduced complexity and energy consumption compared to current DL models. Efficient dendritic tree learning is based on previous research by the Bar-Ilan team that shows evidence of sub-dendritic adaptation in neuronal cultures. A new type of hardware is required for the efficient implementation of this approach as existing GPU technology is better suited for current DL models. This development paves the way for biologically inspired AI hardware and algorithms.