0:000:00
<p>Todays paper: VICReg (<a href='https://arxiv.org/abs/2105.04906'>https://arxiv.org/abs/2105.04906</a>)<br/><br/><b>Summary of the paper</b><br/>VICReg prevents representation collapse using a mixture of variance, invariance and covariance when calculating the loss. It does not require negative samples and achieves great performance on downstream tasks.<br/><br/><b>Highlights of discussion</b></p><ul><li>The VICReg architecture (Figure 1)</li><li>Sensitivity to hyperparameters (Table 7)</li><li>Top 5 metric usefulness</li></ul>