generalization bound estimation preliminary result

[aseok]

Hi. here are published ARC-2 weight estimated generalization bounds using paper data: [hoeffding] n=1000, correct=403, delta=0.05 | emp_acc=40.30% -> acc_lower=36.01% (err_upper=63.99%) [empirical_bernstein] n=1000, correct=403, delta=0.05 | emp_acc=40.30% -> acc_lower=34.63% (err_upper=65.37%) [wilson_one_sided] n=1000, correct=403, delta=0.05 | emp_acc=40.30% -> acc_lower=37.78% (err_upper=62.22%) [pac_bayes_seeger (KL≈2.85e+06, k=562741762, prior_var=1.0)] n=1000, correct=403, delta=0.05 | emp_acc=40.30% -> acc_lower=40.30% (err_upper=59.70%)