http://imatge-upc.github.io/telecombcn-2016-dlcv/
Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of big annotated data and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which had been addressed until now with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks and Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles and applications of deep learning to computer vision problems, such as image classification, object detection or text captioning.
[데이터를 부탁해] 항공기상 데이터 분석으로 운항 스케줄 예측하기 by 신진환FAST CAMPUS
2015년 11월 20일, 패스트캠퍼스가 개최한 [데이터를 부탁해] 오픈 세미나의 2번째 세션에서 발표하신, [파이썬을 활용한 데이터 분석 CAMP]를 수강하셨던 신진환 님의 자료입니다.
http://www.fastcampus.co.kr/dab_openlecture_151120/
[파이썬을 활용한 데이터 분석 CAMP] 자세히 보기 ↓
http://www.fastcampus.co.kr/data_camp_pda/
http://imatge-upc.github.io/telecombcn-2016-dlcv/
Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of big annotated data and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which had been addressed until now with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks and Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles and applications of deep learning to computer vision problems, such as image classification, object detection or text captioning.
[데이터를 부탁해] 항공기상 데이터 분석으로 운항 스케줄 예측하기 by 신진환FAST CAMPUS
2015년 11월 20일, 패스트캠퍼스가 개최한 [데이터를 부탁해] 오픈 세미나의 2번째 세션에서 발표하신, [파이썬을 활용한 데이터 분석 CAMP]를 수강하셨던 신진환 님의 자료입니다.
http://www.fastcampus.co.kr/dab_openlecture_151120/
[파이썬을 활용한 데이터 분석 CAMP] 자세히 보기 ↓
http://www.fastcampus.co.kr/data_camp_pda/
http://imatge-upc.github.io/telecombcn-2016-dlcv/
Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of big annotated data and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which had been addressed until now with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks and Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles and applications of deep learning to computer vision problems, such as image classification, object detection or text captioning.
Dragonflow is an integral part of OpenStack that provides distributed SDN capabilities for Neutron including scale, performance, and latency. It uses a lightweight and easily extensible distributed control plane with pluggable database support. Current features include L2/L3 networking, tunnels, distributed DHCP, and selective database distribution. The roadmap includes adding container, SNAT/DNAT, reactive database, and service chaining support.
This document summarizes Dragonflow, an OpenStack Neutron plugin that implements a distributed SDN controller. Some key points:
- Dragonflow provides a full implementation of the Neutron API and acts as a lightweight distributed SDN controller using a pluggable database.
- It aims to provide advanced networking services like security groups, load balancing, and DHCP in an efficient, scalable way.
- As an integral part of OpenStack, it is fully open source and designed for performance, scalability, and low latency. Its distributed control plane can sync policies across compute nodes.
Neutron Done the SDN Way
Dragonflow is an open source distributed control plane implementation of Neutron which is an integral part of OpenStack. Dragonflow introduces innovative solutions and features to implement networking and distributed network services in a manner that is both lightweight and simple to extend, yet targeted towards performance-intensive and latency-sensitive applications. Dragonflow aims at solving the performance
This document summarizes lecture material on face recognition. It discusses face detection, alignment, identification, and verification. It also reviews several popular face recognition systems like DeepFace, FaceNet, and Deep ID. Experiments were conducted at UPC on various databases using deep neural networks like VGG, GoogleNet, and ResNet. The best results achieved 97% accuracy on a database of 3,500 identities and 100,000 images. Ongoing work involves verification using advanced techniques like joint Bayesian models, siamese networks, and triplets.
DragonFlow sdn based distributed virtual router for openstack neutronEran Gampel
Dragonflow is an implementation of a fully distributed virtual router for OpenStack® Neutron™ that is based on a light weight SDN controller
blog.gampel.net
OpenStack Korea 2015 상반기스터디(devops) 스크립트로 오픈스택 설치하기 20150728jieun kim
※ 본 발표자료는 DevOps팀의 codetree님이 주도적으로 작성하신 shell script를 리뷰하여 작성하였습니다.
[OpenStack Korea Community Study Group, DevOps]
2015년 상반기 두번째 스터디, DevOps Class
"쉘 스크립트를 활용한 오픈스택 Kilo 설치 - 10분만에 끝내기"
D2에서 진행한 스터디 마무리 발표, 2번째 발표에대한 자료입니다.
http://imatge-upc.github.io/telecombcn-2016-dlcv/
Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of big annotated data and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which had been addressed until now with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks and Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles and applications of deep learning to computer vision problems, such as image classification, object detection or text captioning.
Dragonflow is an integral part of OpenStack that provides distributed SDN capabilities for Neutron including scale, performance, and latency. It uses a lightweight and easily extensible distributed control plane with pluggable database support. Current features include L2/L3 networking, tunnels, distributed DHCP, and selective database distribution. The roadmap includes adding container, SNAT/DNAT, reactive database, and service chaining support.
This document summarizes Dragonflow, an OpenStack Neutron plugin that implements a distributed SDN controller. Some key points:
- Dragonflow provides a full implementation of the Neutron API and acts as a lightweight distributed SDN controller using a pluggable database.
- It aims to provide advanced networking services like security groups, load balancing, and DHCP in an efficient, scalable way.
- As an integral part of OpenStack, it is fully open source and designed for performance, scalability, and low latency. Its distributed control plane can sync policies across compute nodes.
Neutron Done the SDN Way
Dragonflow is an open source distributed control plane implementation of Neutron which is an integral part of OpenStack. Dragonflow introduces innovative solutions and features to implement networking and distributed network services in a manner that is both lightweight and simple to extend, yet targeted towards performance-intensive and latency-sensitive applications. Dragonflow aims at solving the performance
This document summarizes lecture material on face recognition. It discusses face detection, alignment, identification, and verification. It also reviews several popular face recognition systems like DeepFace, FaceNet, and Deep ID. Experiments were conducted at UPC on various databases using deep neural networks like VGG, GoogleNet, and ResNet. The best results achieved 97% accuracy on a database of 3,500 identities and 100,000 images. Ongoing work involves verification using advanced techniques like joint Bayesian models, siamese networks, and triplets.
DragonFlow sdn based distributed virtual router for openstack neutronEran Gampel
Dragonflow is an implementation of a fully distributed virtual router for OpenStack® Neutron™ that is based on a light weight SDN controller
blog.gampel.net
OpenStack Korea 2015 상반기스터디(devops) 스크립트로 오픈스택 설치하기 20150728jieun kim
※ 본 발표자료는 DevOps팀의 codetree님이 주도적으로 작성하신 shell script를 리뷰하여 작성하였습니다.
[OpenStack Korea Community Study Group, DevOps]
2015년 상반기 두번째 스터디, DevOps Class
"쉘 스크립트를 활용한 오픈스택 Kilo 설치 - 10분만에 끝내기"
D2에서 진행한 스터디 마무리 발표, 2번째 발표에대한 자료입니다.
3. 최적화
What? 문제에 대한 최적의 해답을 찾기
When? 해답의 수가 너무 많아 전체를 시도하기 어려울 때 사용
How?
여러 다른 해답을 시도 결과 품질을 판단 해답에 점수를 매김
특정 문제에 대한
해답을 도출
4. 최적화
최적화 과정의 적용
여러 다른 해답을 시도
결과 품질을 판단
해답에 점수를 매김
특정 문제에 대한
해답을 도출
최적화 과정
최적화 알고리즘 적용
해답의 표현
해답의 비용 계산
비용이 가장 작은
해답을 도출
최적화 알고리즘 적용
5. 단체여행
예시. 가족 단체 여행
시카고 오헤어 국제공항
ORD
애크런-캔턴 공항
CAK
뉴욕라과디아 공항
LGA오마하 공항
OMA
보스턴로건 국제공항
BOS
마이애미 국제공항
MIA
달라스/포르 워스 국제공항
DAL
가족 여행을 위해 가족 구성원들이 만나기로 함
각각 전국에 퍼져있고, 뉴욕에서 만나길 원함
같은 날 출발하여 같은 날 도착하고, 가능하면 공항 이용 운임을 공유하고 싶어함
6. 비용함수
변수 - 가격, 행사, 거리, 양, 질
평가단위 - 비용
“algorithms
”
시장
대형마트 편의점
동네마트
7. 무작위 검색
Random Searching Algorithm
Random Search for Hyper-Parameter Optimization
James Bergstra, Bengio
비용 구하기
무작위로 해답 생성
반복
현재 최적값과 비교
최적값 도출 - Direct Search Method
- 지속적인 것을 찾기위해 연관성을 가지지 않음
- 정확한 계산이 가능함
- 임의의 초기화에 사용하기 좋음
- http://www.sorting-algorithms.com/shell-sort
[Matlab Central]
Random Search Algorithm by Reza Ahmadzadeh
9. 시뮬레이티드 어닐링
Simulated Annealing Algorithm
# Annealing : 합금을 가열한 후 천천히 냉각하는 과정
# 시나리오
1. 울창한 숲이 우거진 가장 숲에서,
가장 높은 곳을 오르고 싶을때 (방향도 모름)
2. 무작정 느낌으로 가는데, 그 길이 맞을 확률은 매우 낮음.
3. 방향을 잡고 어느정도 오르다가 주사위를 굴림
4. 특정한 수가 나오면 특정 확률로 절벽이나 비탈아래로 미끌어짐
5. 다시 올라감
6. 반복
http://sens.tistory.com/404 https://www.youtube.com/watch?v=iaq_Fpr4KZc
Hill Climbing Method + Heuristics
- Local Minima에 빠짐
- 현재 위치보다 높은 곳으로 이동하지 않음
- 살짝 흔들어줌 à Heuristics
- Global Minima 도착
10. 유전자 알고리즘
Genetic Algorithm, GA
# 자연선택 또는 적자생존의 원칙에 입각한 알고리즘
- 해법 집합 중 비용이 낮은 것을 추림
- 추린 집합을 변이/교배하여 새로운 해법 집합 생성
- 일정 반복