A biocomputing paradigm for zero-shot adaptation and learning in neural networks

The living neural networks of brains can rapidly adapt to new contexts and learn from limited data, a coveted performance characteristic that the AI community has struggled to mimic. We propose that one shortcoming of state-of-the-art AI algorithms is the underlying assumption that learning is associated with the strengthening of links. We developed a distinct learning paradigm that yields rapid adaptation and learning in neural networks while maintaining robustness and validated it on several chaotic dynamical systems. In our algorithm, the network generates the missing contextual tokens required to perform as a generalist AI architecture capable of predicting dynamics in multiple contexts. Furthermore, our algorithm allows the network to extrapolate dynamics to unseen contexts.