eduzhai > Applied Sciences > Engineering >

Fine-Tuning DARTS for Image Classification

  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: Neural Architecture Search (NAS) has gained attraction due to superiorclassification performance. Differential Architecture Search (DARTS) is acomputationally light method. To limit computational resources DARTS makesnumerous approximations. These approximations result in inferior performance.We propose to fine-tune DARTS using fixed operations as they are independent ofthese approximations. Our method offers a good trade-off between the number ofparameters and classification accuracy. Our approach improves the top-1accuracy on Fashion-MNIST, CompCars, and MIO-TCD datasets by 0.56 , 0.50 , and0.39 , respectively compared to the state-of-the-art approaches. Our approachperforms better than DARTS, improving the accuracy by 0.28 , 1.64 , 0.34 ,4.5 , and 3.27 compared to DARTS, on CIFAR-10, CIFAR-100, Fashion-MNIST,CompCars, and MIO-TCD datasets, respectively.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×