This document summarizes two papers presented at NIPS 2018 on anomaly detection and out-of-distribution detection. The first paper proposes a simple unified framework using geometric transformations and Dirichlet density estimation to detect anomalies and adversarial examples. The second paper introduces a method that uses an ensemble of neural networks to detect out-of-distribution samples and adversarial attacks with state-of-the-art performance on CIFAR-10, SVHN and FGSM attacks. It also explores applications to class-incremental learning.
This document summarizes two papers presented at NIPS 2018 on anomaly detection and out-of-distribution detection. The first paper proposes a simple unified framework using geometric transformations and Dirichlet density estimation to detect anomalies and adversarial examples. The second paper introduces a method that uses an ensemble of neural networks to detect out-of-distribution samples and adversarial attacks with state-of-the-art performance on CIFAR-10, SVHN and FGSM attacks. It also explores applications to class-incremental learning.
[DL輪読会]Domain Adaptive Faster R-CNN for Object Detection in the WildDeep Learning JP
The document discusses domain adaptive faster R-CNN for object detection. It proposes a method to adapt a model trained on labeled data from a source domain to detect objects in an unlabeled target domain. The method uses an end-to-end deep learning model with two stages. First, it reduces differences in image distributions between the source and target domains. Then it performs object detection on the target domain images using the adapted model.
論文紹介:Omnivore: A Single Model for Many Visual ModalitiesToru Tamaki
Rohit Girdhar, Mannat Singh, Nikhila Ravi, Laurens van der Maaten, Armand Joulin, Ishan Misra, "Omnivore: A Single Model for Many Visual Modalities" CVPR2022
https://openaccess.thecvf.com/content/CVPR2022/html/Girdhar_Omnivore_A_Single_Model_for_Many_Visual_Modalities_CVPR_2022_paper.html
[DL輪読会]Domain Adaptive Faster R-CNN for Object Detection in the WildDeep Learning JP
The document discusses domain adaptive faster R-CNN for object detection. It proposes a method to adapt a model trained on labeled data from a source domain to detect objects in an unlabeled target domain. The method uses an end-to-end deep learning model with two stages. First, it reduces differences in image distributions between the source and target domains. Then it performs object detection on the target domain images using the adapted model.
論文紹介:Omnivore: A Single Model for Many Visual ModalitiesToru Tamaki
Rohit Girdhar, Mannat Singh, Nikhila Ravi, Laurens van der Maaten, Armand Joulin, Ishan Misra, "Omnivore: A Single Model for Many Visual Modalities" CVPR2022
https://openaccess.thecvf.com/content/CVPR2022/html/Girdhar_Omnivore_A_Single_Model_for_Many_Visual_Modalities_CVPR_2022_paper.html
2. Intro EigenTrust Proposed Method Evaluations
. ........ ..... .........
論⽂紹介
The EigenTrust Algorithm for Reputation
Management in P2P Networks
Sepandar D. Kamvar, Mario T. Schlosser, and Hector
Garcia-Molina
WWW2003 (ACM 1-58113-680-3/03/0005), 2003
2/25
16. Intro EigenTrust Proposed Method Evaluations
. ........ ..... .........
Algorithm 3: Distributed EigenTrust
• Ai set of peers which have downloaded files from
peer i
• Bi set of peers from which from peer i has
downloaded files
§
1 Each peer i do
(0)
2 Query a l l peers j ∈ Ai f o r tj = pj ;
3 repeat
(k+1) (k) (k)
4 Compute ti = (1 − α)(c1i ti + · · · + cni tn ) + αpi ;
(k+1)
5 Send cij ti to a l l peers j ∈ Bi ;
6 Compute δ = ||t(t+1) − t(t) || ;
(k+1)
7 Wait f o r a l l peers j ∈ Ai to return cji tj ;
8 until δ < ϵ;
9 end
13/25
20. Intro EigenTrust Proposed Method Evaluations
. ........ ..... .........
Algorithm 4: Secure EigenTrust
§ 最終バージョン
1 Each peer i do
2 ⃗
Submit l o c a l t r u s t values ci to a l l score managers at
position hm (posi ), m = 1 · · · M − 1 ;
3 ⃗
Collect l o c a l t r u s t values cd and sets of acquaintances
Bi of daughter peers d ∈ Di ;
d
4 Submit daughter d’s l o c a l t r u s t values cdj to score
managers hm (posd ), m = 1 · · · M − 1, ∀j ∈ Bi ;
d
5 Collect acquaintances Ai of daughter peers ;
d
6 foreach daughter peer d ∈ Di do
7 Query a l l peers j ∈ Ai f o r cjd pj ;
d
8 repeat
(k+1) (k) (k) (k)
9 Compute td = (1 − α)(c1d t1 + c2d t2 · · · + cnd tn ) + αpd ;
(k+1)
10 Send cdj td to a l l peers j ∈ Bi ;
d
(k+1)
11 Wait f o r a l l peers j ∈ Ad to return cjd tj
i ;
(k+1) (k)
12 until ||td − td || < ϵ;
13 end 15/25
14 end