Fig. 6: Neural network optimization of the resetting protocol.
From: Adaptive resetting for informed search strategies and the design of non-equilibrium steady-states

a Ball-and-stick representation of the folded (left), misfolded (center), and unfolded (right) states of chignolin in explicit water solvent (5889 atoms), including a cartoon representation of the backbone in crimson. The white, gray, blue, and red spheres represent hydrogen, carbon, nitrogen, and oxygen atoms, respectively. b The free energy along the C-alpha root-mean-square deviation (RMSD) from the folded configuration. The RMSD values at the unfolded and misfolded states are indicated with blue and yellow stars, respectively. c The predicted speedup under a neural network-based resetting protocol as a function of the number of epochs. The dotted and dashed vertical lines highlight values of 25 and 100 epochs, respectively. d Neural network-based resetting probability as a function of the RMSD. Dotted and dashed curves give the results obtained after 25 and 100 epochs, respectively. Results in (c, d) are given for simulations initiated at the unfolded state (blue and purple) or the misfolded state (yellow and orange).