목표

(2023.09.15 갱신)

1. ViT-PFs, CNN-GFs
2. SimVLM

당장 목표

강의
https://cse.snu.ac.kr/en/node/68293

• 기타

1. Diffusion Models Beat GANs on Image Synthesis
   https://arxiv.org/abs/2105.05233
2. T5 (논문 기니까 리뷰라도 읽어보기)
3. DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
4. Photorealistic text-to-image diffusion models with deep language understanding
5. Networks of spiking neurons: the third generation of neural network models

• object detection

1. ResNeXt
2. DenseNet
3. SqueezeNet
4. ShuffleNet
5. R-FCN

• 기타2

1. hard negative example mining - Example-based learning
   for view-based human face detection
2. IOU loss - UnitBox: An advanced object detection network

• image caption

1. show and tell
2. show, attend and tell

• 김건희 교수 랩 논문
  http://vision.snu.ac.kr/gunhee/index.html#top

• 멀티모달 리뷰논문
  https://velog.io/@pabiya/A-survey-of-multimodal-deep-generative-models

1. Tadas Baltruˇsaitis, Chaitanya Ahuja, and Louis-Philippe Morency. Multimodal machine learning: A survey and taxonomy. IEEE transactions on pattern analysis and machine intelligence, 41(2):423–443, 2018.
2. Jing Gao, Peng Li, Zhikui Chen, and Jianing Zhang. A survey on deep learning for multimodal data fusion. Neural Computation, 32(5):829–864, 2020.
3. Chao Zhang, Zichao Yang, Xiaodong He, and Li Deng. Multimodal intelligence: Representation learning, information fusion, and applications. IEEE Journal of Selected Topics in Signal Processing, 14(3):478–493, 2020.

• 어려운데 읽어야하는 논문

1. CLIP
2. diffusion
3. ddpm
4. ddim

• 멀티모달 추가

1. ALBEF
2. SimVLM
3. BLIPv2
4. SwAV
5. Unified Vision-Language Pre-Training for Image Captioning and VQA
6. DALL·E

https://keepgoingrunner.tistory.com/category/Self%2CSemi-supervised%20learning

• VLP 분야 survey

• contrastive learning

• segmentation

여기 benchmark 논문들
https://github.com/open-mmlab/mmsegmentation

논문

일단 년도별 추천 paper 찾아보고 학회 best paper award 받은 걸로도 찾아보자
ex) CVPR, ICCV, ECCV, NeurIPS, ICML, AAAI, ICLR

1. Computer Vision: 10 Papers to Start
   

2. 20+ hottest research papers on Computer Vision, Machine Learning

3. Some interesting Computer Vision papers from ICCV 2017

4. The 10 coolest papers from CVPR 2018

• Deep Learning Papers Reading Roadmap

• https://www.notion.so/c3b3474d18ef4304b23ea360367a5137?v=5d763ad5773f44eb950f49de7d7671bd

---

실습

• https://github.com/ndb796/Deep-Learning-Paper-Review-and-Practice

• https://herbwood.tistory.com/category/Computer%20Vision/Code%20Review

• GAN
  https://github.com/PHANTOM0122/GAN_pytorch

• seq2seq/word2vec

• transformer (attention is all you need)

• VAE

• NTM, DNC

• ViT

• BERT

---

파이토치

• 파이토치(PYTORCH) 한국어 튜토리얼에 오신 것을 환영합니다!
• PyTorch로 시작하는 딥 러닝 입문
• PyTorch 한국어 튜토리얼
• https://www.boostcourse.org/ai214/joinLectures/25076

---

흥미있는 주제들

멀티모달
https://blog.naver.com/jaeyoon_95/222040999747

object detection

• MultiBox (?)

1. CenterNet
2. CornerNet
3. FCOS
4. Path Aggregation Network (PAN)
5. BiFPN
6. NAS-FPN

segmentation

• https://deep-learning-study.tistory.com/category/%EB%85%BC%EB%AC%B8%20%EC%9D%BD%EA%B8%B0/Segmentation?page=2
• https://ambitious-posong.tistory.com/101
• https://velog.io/@xpelqpdj0422/7.-%EB%AC%BC%EC%B2%B4%EB%A5%BC-%EB%B6%84%EB%A6%AC%ED%95%98%EC%9E%90-Segmentation-%EC%95%8C%EC%95%84%EB%B3%B4%EA%B8%B0

instance seg = 녹색, semantic seg = 붉은색.

Disentanglement

• https://velog.io/@tobigs-gm1/Disentanglement

knowledge distillation

• https://everyday-deeplearning.tistory.com/entry/%EC%B4%88-%EA%B0%84%EB%8B%A8-%EB%85%BC%EB%AC%B8%EB%A6%AC%EB%B7%B0Knowledge-Distillation-A-Survey

contrastive learning

강화학습

Nerf

모델 경량화
https://velog.io/@ailab/%EB%94%A5%EB%9F%AC%EB%8B%9D-%EB%AA%A8%EB%8D%B8-%EA%B2%BD%EB%9F%89%ED%99%94

• 얼굴인식

1. DeepFace
2. FaceNet

---

논문으론 읽지 않더라도 조사해 볼것들

• SVM
  https://seunghan96.github.io/ml/ppt/2.SVM/

• autoencoder

• BOW(bag of words)

• Bidirectional RNNs

• Multi-layer RNNs = stacked RNNs

• 메타러닝

• 강화학습 DQN

• transformer
  https://hoya012.github.io/blog/Vision-Transformer-1/

---

궁금하지만 우선도 낮은 것들

• NLP 삼진에바
  T5, 2020 Jul, 67페이지
  GPT-3, 2020 Jul, 75페이지
  PaLM, 2022 Oct, 87페이지
  GPT-4, 2023 Mar, 100페이지

• Decoupled neural interfaces using synthetic gradients (github)

• R. K. Srivastava, K. Greff, and J. Schmidhuber, “Training very deep networks,” in Conference on Neural Information Processing Systems, 2015

  Highway networks [15] introduced a gating mechanism to
  regulate the flow of information along shortcut connections

GAN시리즈

---

제낀 것

Hinton
1. RBM (restricted boltzmann machine)
2. DBN (deep belief networks)
3. DBM (deep boltzmann machine)

논문이 옛날 거라 이해가 힘들고 주제가 흥미도 안 생긴다. 모델 자체는 설명을 찾아봐서 이해했으니 굳이 논문까지 읽을 필요는 없을 것 같다.

---

읽어볼 리뷰 블로그

1. https://www.sshowbiz.xyz/c5acf8f5-266f-48da-ab34-cec29c2b9bb5
   또는
   https://medium.com/@covy-99479

2. https://arclab.tistory.com/category/Paper%20Review?page=9

3. 리뷰 + 실습 (GAN)
   https://velog.io/@wilko97/Paper-Summary-List

4. https://www.notion.so/c3b3474d18ef4304b23ea360367a5137?v=5d763ad5773f44eb950f49de7d7671bd

• https://herbwood.tistory.com/category/Computer%20Vision/Object%20Detection

• https://velog.io/@sjinu/series

---

형식

Paper Review examples (in openreview)

• Summary
  Describe key ideas, experiment setups and how they are related to solving the posed problem of the paper.

• Strengths
  What are the strengths of this paper? For example, there could be novel experiment settings or a posing of a new problem.

• Weaknesses
  What are the weaknesses of the paper? Please state the weaknesses with the concrete reasons. For example, the settings may not be novel since they are posed in other papers(in this case, citations must be included)

• Overall rating
  Rate this paper with scores ranging from 1 to 5

• Justification of rating
  Please explain how the strengths and weaknesses aforementioned were considered for the rating. Please also mention what you expect to see from the rebuttal that may change your rating

• How to improve