Subject to: Dimitris Bertsimas
- Human connection
- Positivity: Stay from negative attitude
- Good heart: Help the others
"Subject to" offers a series of informal conversations with relevant figures in the fields of Operations Research, Combinatorial Optimization and Logistics, and they are hosted by Anand Subramanian, an Associate Professor at Universidade Federal da Paraíba, Brazil.
(*) MIT Subject levels & credit
One MIT unit is approximately equal to 14 hours of work per term. The Subject Listing displays units for each subject as a series of three numbers (e.g., 3-2-7). The numbers added together (3+2+7) equal the total credit for the subject (12). In order, the three numbers represent:
units assigned for lectures and recitations.
units assigned to laboratory, design, or field work.
units for outside preparation.
MIT subjects, Management (Course 15): The usual units for most of the courses are 6 - 8 hours per week.
上次留言是2014年在荷蘭防洪INFORMS Edelman Award那篇文章，想不到許老師已經跳槽而且還持續的在台灣吸收美國ORMS的知識，這個精神是我要學習的。數年過去我今年也唸完OR的博士然後在美國業界找到工作，3/27那篇Strong mixed-integer programming formulations for trained neural networks的作者我有幸在2019參加MIP Workshop at MIT時親自聽過Dr. Huchette的報告，我自己研究也是做 MIP method（strong valid inequalities, strong formulation, branch-and-cut），所以很感興趣看他怎麼結合這種exact method來解決neural network目前只能仰賴gradient-based heuristics的問題。讀IEOR博士班感想之一就是OR老師都不喜歡做heuristics 因為發不了好期刊，雖然講求快速的業界都用，這就是學界業界的認知差別吧，祝老師在中原不管是研究生生源還是研究都順順利利
回覆刪除Nice to hear from you and thank you.
回覆刪除I will teach OR again next year, and am thinking how to incorporate these new developments (MIP in ML) into my course.
Another example: https://www.youtube.com/playlist?list=PLdYmuqrR3BQbsvmzv3ydqhz5OkRFHcWm5
When those professors work on industrial problems, they do a lot simplifications and approximations to get workable results, e.g.,
http://chhsu135.blogspot.com/2018/10/making-better-fulfillment-decisions-on.html
http://chhsu135.blogspot.com/2012/04/zara.html
http://chhsu135.blogspot.com/2022/05/garrett-van-ryzin-talks-about.html
I am not really a fan of machine learning, I still like traditional optimization problems. But during my job search I found out that many many tech companies they combine data science, operations research, machine learning and statistics into one scientist/engineer position, which is a bit absurd as they clearly have no idea what they need and who they are hiring. I applied for data scientist - operations research position but the online assessment I got from them is all about machine learning questions. I think it is not very professional. However since this is the trend, many OR PhDs including me and my friend we all have to do self-learning (not big deal for PhD students to do that) on machine learning to handle the job interview.
刪除