Extended Data Fig. 3: Hebbian learning between sensory layer and scaffold also produces memory continuum.
From: Episodic and associative memory from spatial scaffolds in the hippocampus
A memory continuum is obtained in Vector-HaSH even if the weights between the sensory and hippocampal layers are bi-directionally trained using Hebbian learning (instead of pseudoinverse learning, as in Fig. 3. This continuum is also asymptotically proportional to the theoretical bound on memory capacity (forest green dashed line indicative of slope of theoretical upper bound, vertical and horizontal position of dashed line is arbitrary). However, the proportionality constant is lower, with the gradual degradation of information recall occurring well before Nh. Vector-HaSH parameters identical to Fig. 3c with λ = {3, 4, 5}.
