Notice
Recent Posts
Recent Comments
Link
| 일 | 월 | 화 | 수 | 목 | 금 | 토 |
|---|---|---|---|---|---|---|
| 1 | ||||||
| 2 | 3 | 4 | 5 | 6 | 7 | 8 |
| 9 | 10 | 11 | 12 | 13 | 14 | 15 |
| 16 | 17 | 18 | 19 | 20 | 21 | 22 |
| 23 | 24 | 25 | 26 | 27 | 28 | 29 |
| 30 |
Tags
- DualPrompt
- 베이지안 정리
- L2P
- Mask-and-replace diffusion strategy
- learning to prompt for continual learning
- Energy-based model
- Facial Landmark Localization
- learning to prompt
- img2pose: Face Alignment and Detection via 6DoF
- Markov transition matrix
- mmcv
- timm
- state_dict()
- CIL
- VQ-VAE
- Mask diffusion
- Face Pose Estimation
- Continual Learning
- requires_grad
- Vector Quantized Diffusion Model for Text-to-Image Synthesis
- VQ-diffusion
- CVPR2022
- Img2pose
- Discrete diffusion
- ENERGY-BASED MODELS FOR CONTINUAL LEARNING
- Class Incremental
- PnP algorithm
- prompt learning
- Class Incremental Learning
- Face Alignment
Archives
- Today
- Total
Computer Vision , AI
[One-page summary] RECL: Responsive ResourceEfficient Continuous Learning for Video Analytics (NSDI 2023) by Khani et al. 본문
Paper_review[short]
[One-page summary] RECL: Responsive ResourceEfficient Continuous Learning for Video Analytics (NSDI 2023) by Khani et al.
Elune001 2024. 1. 15. 20:43● Summary: Get fast and accurate expert models from edge devices with model selection and model retraining strategies
● Approach highlight
Model selection: using a gate network to measure the performance of each model zoo. based on this score, select the top-k of model zoo
Model retraining: retrain each incoming job from an edge device by 1 epoch and retrain the model with the highest improvement for the remaining time.
● Main results
● Discussion
- Is it possible to use a large-scale teacher model in a real-world online setting?
'Paper_review[short]' 카테고리의 다른 글
'Paper_review[short]' Related Articles
more
