+ All Categories
Home > Documents > Closing Remarks - ETH ZSummary Top-5 accuracy is improved 1st place 82.54% (79.25% in 2018)...

Closing Remarks - ETH ZSummary Top-5 accuracy is improved 1st place 82.54% (79.25% in 2018)...

Date post: 21-Feb-2021
Category:
Upload: others
View: 1 times
Download: 0 times
Share this document with a friend
4
Closing Remarks
Transcript
Page 1: Closing Remarks - ETH ZSummary Top-5 accuracy is improved 1st place 82.54% (79.25% in 2018) Text/meta-data is the new focus Bert model Semantic relation based on WordNet Other issues

Closing Remarks

Page 2: Closing Remarks - ETH ZSummary Top-5 accuracy is improved 1st place 82.54% (79.25% in 2018) Text/meta-data is the new focus Bert model Semantic relation based on WordNet Other issues

Summary● Top-5 accuracy is improved

○ 1st place 82.54% (79.25% in 2018)

● Text/meta-data is the new focus○ Bert model○ Semantic relation based on WordNet

● Other issues○ Label noise○ Class imbalance○ Model ensemble○ Long-tail distribution

Page 3: Closing Remarks - ETH ZSummary Top-5 accuracy is improved 1st place 82.54% (79.25% in 2018) Text/meta-data is the new focus Bert model Semantic relation based on WordNet Other issues

Open Questions● Definition of classes

○ What classes are representative?

● Number of classes○ Are 5,000 classes enough or too many?

● Larger scale○ Shall we increase the dataset size?

● More tasks○ Webly supervised semantic segmentation / object detection○ Domain adaptation / transfer learning○ Video recognition

Page 4: Closing Remarks - ETH ZSummary Top-5 accuracy is improved 1st place 82.54% (79.25% in 2018) Text/meta-data is the new focus Bert model Semantic relation based on WordNet Other issues

Thank you!


Recommended