Papers by Xiping Xiping Hu
Layer-wise Regularized Dropout for Neural Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods to regularize dropout are consistency training and dropout is a problem in many pre-trained neural language models. |
| Approach: | They propose a layer-wise regularized dropout technique which regularizes dropout at the output layer using consistency training. |
| Outcome: | The proposed model can be regarded as a "self-distillation" framework, in which each sub-model generated by dropout is the other's "teacher" model and "student" model. |