A research work is highly significant if it lays the foundation for future work.
This may violates some newbie in research area. From industry perspectic, people tend to ask: what is the usage of this method? what is the performance, including accuracy and efficiency? If we apply the significant definition by the authors, we will realize the above question is not important indices for evaluating an academic papers. A significant paper may have no usage in industry currently, but it possesses the potential to change the world. Because of this reason, not everyone enjoy the academic research despite the difficulty. Some people is urgent to check the outcome and adjust the path to success, others enjoy the "Eureka!" process, no matter the idea is useful or not.
This idea remind me of paper RECOS and Saak transform. It mimic CNN structure, but every layer has been changed to a determined operation. It can solve two critical problems in deep learning: the theorectical explaination and the learning inefficiency. In the experiment section, the author compares the result with traditional classifier e.g. SVM, KNN on MNIST. So far it cannot handle large dataset like ImageNet. After reading the paper, my reflection is: an interesting paper, yet no good experimental results, seems not so useful. But if I recall the definition of significant, I should understand usage is never "usage" when judging a research paper, the novelty and significant matter much more.