4 accepted papers
In this work, we investigate how to sparsify a pre-trained dense large language model into a mixture-of-experts (MoE) architecture for faster inference. Our approach applies mask matrix to the activations for each expert, constrained by L0 regularization to minimize the number of activated parameter…
We present an efficient BERT-based multi-task (MT) framework that is particularly suitable for iterative and incremental development of the tasks. The proposed framework is based on the idea of partial fine-tuning, i.e. only fine-tune some top layers of BERT while keep the other layers frozen. For e…
Conditional Random Field (CRF) based neural models are among the most performant methods for solving sequence labeling problems. Despite its great success, CRF has the shortcoming of occasionally generating illegal sequences of tags, e.g. sequences containing an “I-” tag immediately after an “O” tag…
We propose Independent Component Autoencoder (ICAE), a deep neural network-based framework for nonlinear Independent Component Analysis (ICA). The proposed method consists of a penalized autoencoder and a training objective that is to minimize a combination of the reconstruction loss and an ICA cont…