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 | 31 |
- ddpm
- pedestrian detection
- stochastic contraction theory
- ssda-yolo
- fourmer
- f2dnet
- semi-supervied learning
- unconditional ddpm
- hqs-algorithm
- ddim
- ilvr
- diffusion
- ccdf
- image restoration
- linear inverse problem
- deepfake detection
- face forgery detection
- facenet
- Ai
- conditional diffusion
- IR
- focal detection network
- ddrm
- daod
- object detection
- Triplet
- neighboring pixel relationships
- diffpir
- unconditional generative models
- ML
- Today
- Total
목록전체 글 (16)
Stand on the shoulders of giants

Main IdeaKeyword : Fourier transformation을 이용한 image restorationFourier transform을 사용하면general image restoration prior를 줄 수 있음.image degradation을 disentangle할 수 있음.global properties + intrinsic priorDFT 특성 👍Amplitude component가 결정적이기때문에, amplitude와 phase component를 swap하면 degradation이 clear version으로 transferred됨.degradation은 주로 amplitude component에서 일어나고 이때 fourier transform을 사용해 image degrada..

Main Idea기존 Conditional diffusion models(CMSs)는 주로 training-required methods라 itme-dependent classifier/score estimator이기 때문에 cost가 크다. ⇒ training-free를 제안해 시간을 줄이고자함.Energy function을 sampling process에서 guide로 사용하자.단순, 효과적, 적은 비용IntroductionCDM (Conditional Diffusion Models)다양한 guide를 주면서 generation하는 모델ex. condition = text, class labels, degraded images, landmarks, segmentation maps, …종류Trainin..

Goal WF attacks가 실제로 가능할지를 확인 가정 검증 정확도에 실질적으로 영향을 주는 변수들 찾기 FP를 어떻게 줄일 수 있을지 공격자 비용 모델링 (perfect WF system을 유지하는 것은 비용이 커서) 대표적 가정 2개 공격자와 유저 동엘 TBB(Tor Browser Bundle) 동일 localized version of a limited set of pages,sigtes 방문 Model passive, local modelling Assumption 유저의 traffic 모니터링 o, 나머지 조작x 복호화 불가능 Targeted of victim = 1 victim과 비슷한 condition에서 모델 학습 가능 근데 실질적으로 유저의 setting아는건 어려움 Non-targe..

[ updated : 2024-02-17: 요약본 하단 참조]논문 선택 이유: Augmentation Project를 위한 인사이트 얻기. WeFDE의 feature들 가져다가 활용해보기 (trace에서 정보 유출양을 feature별로 보여줌) ⇒ 수학적 서술의 Idea를 얻는게 이 논문을 읽는 것의 목표. GoalTor가 traffic overhead를 minimize하기 위해 website에 대해 불가피하게 leak하는 information의 amount를 측정함.ex. leaked information : volum, timing, direction of communicationsKeysvalidating WF defenses by accuracy alone is flawedjoint inform..

KeysNSL with triplet networks새로 수집한 데이터 바로 모델학습에 사용가능/ 적은자원만으로도(weak adversary) WF 가능 [Threat landscape]Abstract & introduction기존것의 한계DF = 큰 데이터셋을 요구받음, 꾸준히 update해야함대부분의 가정train/test data가 비슷한 distribution을 갖는다.train/test data가 “같은 시간”에 “같은 type”의 network condition에서 수집되었다.공격자는 weak attacker model (realistic condition에 맞는)BackgroundBootstrap time = 시스템 초기화 및 가동 과정이 완료되는 데 소요되는 시간(논문) ready-to-u..

CCS23-SSL-Web-FingerprintBACKGROUNDlow latency이면 대기시간이 적다는거 → 전송속도 빠름 → 일관된 통신!!High bandwidth이면 →더 많은 데이터를 동시에 처리 → 일관된 통신!!ABSTRACT일반적인 WF classifier특정 high-bandwidth network link들에서 수집된 high-bandwidth sample들로 학습되었는데얘랑 다른 condition의 network connection에 적용시INTROTor = 가장 유명한 익명 대화 시스템client의 traffic을 프록시들의 집합으로 relay함으로써 익명성을 제공함.WF(Website Fingerprinting) =피해자의 network traffic을 감독하는 passive 공격..