Updated time: 2020-07-08 21:09:50
05T-1081017
2019-11-04
Copy right
Continual lifelong learning is essential to many applications. Our system achieve unforgetting lifelong learning with model weight pruning, critical weights selection, and progressive networks expansion. By enforcing their integration in an iterative manner, we introduce an incremental learning method that is scalable to the number of sequential tasks in a continual learning process. Three benefits includes: first, it can avoid forgetting previous tasks; second, it allows model expansion while can maintain the model compactness when handling sequential tasks; besides, through our compaction and selection/expanding mechanism, the knowledge accumulated through learning previous tasks is helpful to adapt to a better model for the new tasks.
- Avoid forgetting previous tasks.
- Allows model expansion while can maintain the model compactness when handling sequential tasks.
- The knowledge accumulated through learning previous tasks is helpful to adapt to a better model for the new tasks through our compaction and selection/expanding mechanism.
- Smart logistics, Smart Manufacture
- Intelligent robot
- Smart city, security surveillance

Fig1. Compacting Picking and Growing (CPG) flow chart. Given well-trained model for the Task 1, we will gradually prune the model weights. When new task come, a new model is learned by simultaneous picking proper weights from old model and filling new weights in the pruned model “hole”. If the new model cannot satisfy the requirement, we can expand the model with new weights to improve the performance.
Chu-Song Chen
If you are interested in this technology, please fill in the information on the webpage under the manufacturer option (Request for Technology Licensing, Material Transfer, or Sponsored Research) on the website of this office

Compacting, Picking and Growing for Unforgetting Continual Learning