Monadic Pavlovian associative learning in a backpropagation-free photonic network

Over a century ago, Ivan P. Pavlov, in classic experiment, demonstrated how dogs can learn to associate ringing bell with food, thereby causing ring result salivation. Today, it is rare find the use of Pavlovian type associative learning for artificial intelligence (AI) applications even though other concepts, particular backpropagation on neural networks (ANNs) have flourished. However, training using method 'conventional' ANNs, especially form modern deep (DNNs), computationally and energy intensive. Here we experimentally demonstrate backpropagation-free single (or monadic) hardware element. We realize this an integrated photonic platform phase-change materials combined on-chip cascaded directional couplers. then develop scaled-up circuit network our monadic that delivers distinct machine-learning framework based single-element associations and, importantly, architectures address general tasks. Our approach reduces computational burden imposed by conventional approaches, increasing speed, whilst also offering higher bandwidth inherent implementation.