Why might NAS outperform human-designed architectures?
학생이 쓴 원문
I can\'t ensure that NAS\'s result always outperforms than human-designed architectures. This situation is related with sub-optimal point. If we use gradient-based search strategy, the result might fail to reach to global optimal point due to natual of gradient update method. Moreover, if we limit a search space due to computational cost, NAS\'s result might not the best architecture due to lack of candidates. Therefore, if you want to get optimal architecture, you have to set enough search space and great search algorithm.
강사 첨삭
Hi, Mary! You explained NAS’s strengths and limitations well and showed good understanding of gradient-based search and search space issues. Your reasoning is strong, but grammar, prepositions, and word choice need some polishing. With clearer phrasing and careful verb forms, your explanation will be very professional and precise. Great technical insight!
-Teacher Wela
I can\'t ensure that NAS\'s result always outperforms than human-designed architectures.
>> I cannot guarantee that the result of NAS will always outperform human-designed architectures.
This situation is related with sub-optimal point.
>> This situation is related to the possibility of reaching a sub-optimal point.
If we use gradient-based search strategy, the result might fail to reach to global optimal point due to natual of gradient update method.
>> If we use a gradient-based search strategy, the result might fail to reach the global optimum due to the nature of gradient update methods.
Moreover, if we limit a search space due to computational cost, NAS\'s result might not the best architecture due to lack of candidates.
>> Moreover, if we limit the search space because of computational costs, NAS may not find the best architecture due to a lack of candidates.
Therefore, if you want to get optimal architecture, you have to set enough search space and great search algorithm.
>> Therefore, to obtain the optimal architecture, you need to provide a sufficiently large search space and a strong search algorithm.
파워잉글리쉬 영어첨삭교정 서비스에 실제로 접수된 글입니다. 작성자와 글에 등장하는 사람·회사 이름, 연락처 등 개인을 알아볼 수 있는 정보는 모두 지웠습니다. 문법 오류는 학습 자료로서의 가치를 위해 원문 그대로 두었습니다.