AdaCN: An Adaptive Cubic Newton Method for Nonconvex Stochastic Optimization

<table class="table-group" id="tab2"><tr><td><table class="table"><tr><td class="thead-hr" colspan="6"><hr/></td></tr><tr class="thead"><td class="align_left">Model</td><td class="align_center">Adam</td><td class="align_center">SGD</td><td class="align_center">AdaBound</td><td class="align_center">Apollo</td><td class="align_center">AdaCN</td></tr><tr><td class="thead-hr" colspan="6"><hr/></td></tr><tr><td class="align_left">LeNet</td><td class="align_center">0.984</td><td class="align_center">0.981</td><td class="align_center">0.987</td><td class="align_center">0.988</td><td class="align_center"><b>0.989</b></td></tr><tr class="table-tr"><td colspan="6"><hr class="tbody-hr"/></td></tr></table></td></tr><tr class="table-fn"><td><div>The best result is shown in bold.<br/></div></td></tr></table>

<div>Test accuracy of LeNet on Mnist.</div>

Computational Intelligence and Neuroscience

tab2

Table 2

Table 2: AdaCN: An Adaptive Cubic Newton Method for Nonconvex Stochastic Optimization 