|
|
此文章由 Pipiv 原创或转贴，不代表本站立场和观点，版权归 oursteps.com.au 和作者 Pipiv 所有！转贴必须注明作者、出处和本声明，并保持内容完整
AI 目前火热, 主要是因为其中的一个学科"机器学习”-尤其是 是"机器学习”中的一个分支-“深度学习”发展飞快, 这几年, 深度学习在图像识别、语音识别、语言处理、医学自动诊断方面都取得了很多成果, 澳洲有不少大学有不错的 Deep Learning research groups,阿德莱德大学, 墨大, 昆大, Monash, UTS, 悉尼大学, UNSW 都有不错的"机器学习”研究小组, 另外当然还有NICTA and CSIRO 的Data-61 (跟不少澳洲大学有partner 关系 http://www.data61.csiro.au/en/Co ... artner-Universities) 但是总的来说还是缺少国际上的领军人物, Deep Learning领先还是在美国, 以及牛津剑桥, 加拿大倒是在这方面比澳洲强上很多, 主要是有大师移民了加拿大, 后来又带出弟子又成为大师, 所以Montreal and Toronto 大学在Deep Learning 领域都有世界领先的方面在做, 在Deep Learning 算一代宗师的Geoffrey Hinton是剑桥拿的实验心理学的学士学位，爱丁堡大学的人工智能phd，在卡耐基梅隆大学计算机系工作过5年, 后来移民的加拿大, 他创立的NACP项目, 参加NCAP项目研讨会的许多研究人员，比如Yann LeCun、Yoshua Bengio和Andrew Ng，后来都是大牛
楼主, 下面有个link 是Yoshua Bengio 大牛有一次参加了Reddit的机器学习板块 “Ask Me AnyThing”活动时回答Deep Learning 爱好者的问答, 有很多很不错有用的信息, 其实也回答了你的一些问题, "本科学什么"
有人问他下面三类背景学生, 他最喜欢哪一种:
Given three candidates, none of which have much experience in ML, who would you rather chose as a potential student (other dimensions being equal):
•Someone experienced in applied statistics (say, psychology research, or epidemiology), knows R
•Someone who is very good at software development and knows some numpy/scipy, Matlab
•Pure math undergrad who has little exposure to either programming or "real world" data
他的回答是:
I can afford many students. I would not evaluate based on the above features but also based on an interview, in which all aspects come together. Strength in math is an excellent predictor of success in machine learning research, and so math undergrads with good programming skills are very high on my list of preferences. Strong software development is also very important for many of the projects we have, which involve big data and big models, where computational efficiency and top-notch collective programming are really important.
另一个不错的回答关于如何入门的问题,
问:
I'm currently finishing up my undergrad in philosophy of science and logic and I am trying to make the switch to computer science for masters work with the intention of pursuing machine learning at the phd level. Besides filling in the obvious knowledge gaps in mathematics and basic programming skills, what are some of the things a person in my position could do to make themselves a more attractive candidate for your field of work? Thanks so much for visiting us
大师答:
Read deep learning papers and tutorials, starting from the introductory material and moving your way up. Take notes on your reading, trying to summarize what you learned.
Implement some of these algorithms yourself, from scratch, to make sure you understand the math for real, implementing variants of these, not just a copycat of a pseudo-code you found in a paper.
Play with these implementations on real data, maybe competing in Kaggle competitions. The point is that a lot is learned by actually putting your hands in data and playing with variants of these algorithms (this is true in general for machine learning).
Write about your experiences and results and thoughts in a blog. Initiate contact with researchers in the field and ask them if they would like to you to work remotely on some of the projects and ideas they have. Try to do an internship.
Apply to graduate school in a lab that actually does these things.
Is the roadmap clear enough?
https://www.reddit.com/r/Machine ... /ama_yoshua_bengio/
|
评分
-
查看全部评分
|
西餐大赛------ 凑个数，一如既往支持美食版活动 (2011-1-6) youpassit:) 
楼主:jinluo
发表于 2017-4-27 13:36
