Challenge closed-book science exam
Problem being addressed
Standardized science exam is a challenging AI task for question answering, as it requires rich scientific and commonsense knowledge.
A model-agnostic meta-learning method to train a meta-classifier which is able to capture the meta-information that contains similar features among different tasks, so that the meta-classifier can quickly adapt to new tasks with few samples. The questions are regarded with the same knowledge points as a task. The model learns to reason testing questions with the assistance of question labels and example questions (examine the same knowledge points) given by the meta-classifier. Experiments show that the proposed method achieves impressive performance on the closed-book exam with the help of the few-shot classification information.
Advantages of this solution
Experimental results show that the proposed method greatly improves the QA accuracy by +17.6%, and exceeds the performance of retrieval-based QA method.
Solution originally applied in these industries
Possible New Application of the Work
Electronics and Sensors Industry
Meta learning helps to execute training when large training datasets aren't available. For example, a mini robot completes the desired task on an uphill surface during test even through it was only trained in a flat surface environment.
Meta learning can be used to detect fraudulent transactions based on other tasks which the low-level models were trained on.
Source URL: #############