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 |
- 연결 리스트
- 알고리즘
- 배열
- 코딩테스트
- RxJS
- 큐
- 타입스크립트
- cors
- RT scheduling
- APOLLO
- 프로그래머스
- C
- 브라우저
- GraphQL
- 자바스크립트
- 컨테이너
- 연결리스트
- alexnet
- 이진탐색
- 해시테이블
- vue3
- pytorch
- 포인터
- Machine Learning
- 자료구조
- 릿코드
- 프론트엔드
- 스택
- 프로세스
- 웹팩
- Today
- Total
목록Machine Learning (5)
프린세스 다이어리

Melnik, Mikhail, and Denis Nasonov. "Workflow scheduling using neural networks and reinforcement learning." Procedia computer science 156 (2019): 29-36. 1. 연구의 필요성 컴퓨팅 시스템 최적화를 위해 scheduling이 필요하다. 효과적으로 scheduling 문제를 해결하기 위해선 계산 모델과 resource의 높은 heterogeneity 등 다양한 컴퓨팅 환경에 적응하고 학습할 수 있어야 한다. 아직 ML은 scheduling 문제를 해결하기엔 충분치 않다. 기존의 RL을 활용한 scheduling 연구는 (1) 외부 Grid 자원에 따라 task distribution하는 것..

T. Huybrechts, S. Mercelis, and P. Hellinckx, “A new hybrid approach on WCET analysis for real-time systems using machine learning,” in Proc. 18th Int. Workshop Worst-Case Execution Time Anal. (WCET), 2018, pp. 1–12. (1) 연구의 필요성 Worst-Case Execution Time (WCET) of code는 Real-Time 시스템 개발 시 반드시 필요한 정보다. 개발 초기부터 WCET를 추정함으로써 pessimistic한 WCET 추정을 막고 과한 hardware 스펙을 방지해야 한다. WCET 분석을 위한 방법 중 hybrid ..

T. Huybrechts, T. Cassimon, S. Mercelis, and P. Hellinckx, “Introduction of deep neural network in hybrid WCET analysis,” in Proc. Int. Conf. P2P Parallel Grid Cloud Internet Comput., 2018 1. 연구의 필요성 하드 리얼타임 사이버-물리 시스템(CPS)에서는 각 태스크의 WCET(최악 시나리오 실행 시간)을 파악해야 해당 태스크가 예측 가능한지를 판단할 수 있다. WCET 분석을 위한 방법에는 static, measurement-based, hybrid approach가 있다. 이 중 hybrid 방법론은 계산 complexity(static의 장점)과 acc..

PyTorch DataParallel 모듈을 사용하여 병렬컴퓨팅을 구현할 수 있다. 참고로 구현에 사용한 모델은 규모가 작은 AlexNet이다. 1. DataParallel 적용 전 def main(): device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 데이터셋 불러오기 transform = transforms.Compose([ transforms.Resize(size=(227, 227)), transforms.ToTensor(), #이미지를 pytorch tensors 타입으로 변형, 0.0~1.0 으로 변환 transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) # rgb, -1~..

PyTorch Distributed: Experiences on Accelerating Data Parallel Training (Shen Li et al.) PyTorch의 DistributedDataParallel 모듈을 소개한 2020년 논문입니다. (1) 왜 이런 논문을 쓰게 되었는지, (2) 해결하고자 한 문제점은 무엇인지, (3) 어떻게 해결했는지, (4) 평가는 어떻게 이루어졌는지를 중점적으로 정리해보았습니다. 1. Background In the field of Deep Neural Networks (DNN) training, there are three primary steps: the forward pass to compute loss, the backward pass to calcul..