Negative to Positive Co-learning with Aggressive Modality Dropout

We find that by using variant. We show that in situations where there is NCL, by aggressive modality dropout we are able to applying aggressive modality dropout we are able to reverse reverse negative co-learning (NCL) to positive NCL to PCL. While there is prior work documenting the effectiveness co-learning (PCL). Aggressive modality dropout of modality modality dropout during co-learning can be used to'prep' a multimodal model for and multimodal machine learning, we are the first to show unimodal deployment, and dramatically increases that modality dropout can reverse NCL to PCL. model performance during negative co-learning, where during some experiments we saw a 20% gain in accuracy.