2) DQ-HSES vs. HSES on CEC 2014 To validate the capability of DQ-HSES for generalization, more experiments on the CEC 2014 test suite are conducted. Note that no function from CEC 2014 is used for training. Table VI shows the results obtained on the benchmark in 10D, 30D, and 50D over 51 runs. It is clear that for 10D problems, DQ-HSES performs significantly better than HSES on seven functions and worse on two functions. DQ-HSES performs significantly better than HSES on 10 functions and worse on two functions for 30D problems. For 50D problems, it performs significantly better on seven functions and worse on one function. In general, DQ-HSES obtains greater values ofBM than HSES. Thus we may conclude that the learned agent can generalize well on unseen optimization problems. The curves ofoptimization of some functions from the CEC 2014 test TABLE V Results obtained by HSES and DQ-HSES on functions of CEC 2018 in 100D, averaged over 20 runs. 100D HSES MEAN F1 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 BM y= =x 0.00e+00 1.28e-08 6.15e+00 3.73e+00 8.47e-08 1.10e+02 4.32e+00x 2.78e+00 1.29e+03 2.04e+01 7.64e+02 4.39e+01 2.04e+01 1.05e+02 1.09e+03 7.11e+02 4.94e+00 2.30e+01x 5.41e+02 2.21e+02 1.00e+02y 5.45e+02 8.44e+02 7.47e+02 2.35e+03 6.38e+02 4.72e+02 1.19e+03 2.67e+03 14(12) 1/26/2 (1/21/2) STD 0.00e+00 3.58e-08 2.19e+01 1.43e+00 2.03e-08 1.59e+00 2.12e+00 5.57e+00 3.79e+02 3.21e+01 2.45e+02 5.23e+00 4.84e+00 2.30e+01 3.89e+02 2.84e+02 7.42e+00 1.59e+01 2.52e+02 4.11e+00 3.51e-07 5.92e+00 5.35e+00 4.72e+01 1.00e+02 8.01e+00 1.01e+02 2.76e+02 1.28e+02 MEAN 0.00e+00 0.00e+00 1.37e+01 4.47e+00 9.33e-08 1.09e+02 2.78e+00 5.00e-01 1.25e+03 3.82e+01 8.60e+02 4.73e+01 1.92e+01 9.85e+01 1.55e+03 5.88e+02 3.57e+00 1.42e+01 4.93e+02 2.23e+02 4.37e+02 5.43e+02 8.44e+02 7.36e+02 2.34e+03 6.39e+02 5.08e+02 1.34e+03 2.65e+03 17(13) DQ-HSES STD 0.00e+00 0.00e+00 4.15e+01 2.17e+00 2.36e-08 1.04e+00 1.63e+00 8.24e-01 3.13e+02 4.06e+01 2.75e+02 1.14e+01 6.82e+00 2.19e+01 9.10e+02 3.28e+02 5.96e+00 1.99e+00 2.08e+02 3.33e+00 8.29e+02 8.13e+00 8.21e+00 3.28e+01 1.37e+02 9.81e+00 8.39e+01 3.35e+02 1.22e+02 96 IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | FEBRUARY 2023