Learnware: On the future of Machine Learning（2016）

[JChew7568]

Learnware: On the future of Machine Learning（2016） Author: Zhi-Hua Zhou

https://dl.acm.org/doi/10.1007/s11704-016-6906-3

[JChew7568]

Problems:

Machine Learning deficiencies

• Large amount of data required; expensive or difficult when collecting data in many real tasks
• Model hardly perform well / useless if environment changes
• Trained models nowadays are ususally black-boxes

Methods:

Learnware was proposed

Definition: Well-performed pre-trained machine learning model with a specification which explains the purpose and/or speciality of the model. Properties:

• Reusable: pre-trained model should be able to enhanced or adapted, by its new user through a slight modification or refinement using information
• Evolvable: learnware should be able to get accustomised to environment change.
• Comprehensible: Models need to be transparent at some extent, at least enabling eht writing of specification for the learnware.

Conclusions:

• Strong Machine Learning model can be achieved for tasks with small data
• Data Privacy will become a less serious issue
• Enable common End Users to attain tricky learnin gresults that can only be attained by machine learning experts nowadays.

Questions:

Further Reading:

Few-shot learning？