Code and data for the ACL'2021 paper "Continual Learning for Task-oriented Dialogue System with Iterative Network Pruning, Expanding and Masking"

requirements

• Python >=3.7
• Pytorch 1.2.0

1.You can run TPEM with

$ bash experiment/run_TPEM.sh 

After completing the training process, you can use the following bash to obtain all middle results

$ bash experiment/eval_TPEM.sh 

2.To observe the “catastrophic forgetting” of base model, you can run

$ bash experiment/run_GLMP_continual.sh 

Obtain all middle results with

$ bash experiment/eval_GLMP_continual.sh

3.To run Re-init which need to save all 7 models:

$ bash experiment/run_GLMP_Re-init.sh 

Obtain all middle results with

$ bash experiment/eval_GLMP_Re-init.sh

4.Run TPEM with random task order

$ bash experiment/run_TPEM_with_random_task_order.sh 

To evaluate shuffle order results

$ bash experiment/eval_TPEM_with_shuffle_order.sh 

If you find our work helpful, you can also refer to

SIGIR'2021 paper "Iterative Network Pruning with Uncertainty Regularization for Lifelong Sentiment Classification"

IPRLS: https://github.com/siat-nlp/IPRLS