I want to use nengo and nengo-bio to build a bionic model, in which I am concerned that PES is a very popular learning algorithm. After my detailed study, I learned that the PES learning rules are achieved by adjusting the decoder of the spiking trains to achieve the correct output. Is this bio-interpretable? We all know that the STDP/BCM learning algorithm has advantages in terms of biological interpretability. It influences the change of connection weights through the sequence of neuron activity before and after, so as to achieve the purpose of learning. So, does the PES learning rule have a biological counterpart? I want to build a bionic supervised learning algorithm. Any good ideas or examples are welcome.
I you are looking to read more about the biological plausibility of the PES learning algorithm, I’d advise you check out @tbekolay’s Master’s thesis, and in particular, Chapter 4, where he describes the PES rule. Note that it wasn’t called the PES rule in the thesis, but the formulation is there. The name “prescribed error sensitivity” (PES) was coined after his thesis was published.
I also found a discussion here that might be relevant.