kenics.net

Technical notes on perl, python, php, sql, cgi, c/c++, q/kdb+, unix/shell, revision control tools, data structures & algorithms, and their applications into web services and other various forms of software engineering.

omscs story

Enrolled:  2014 Aug 
Graduated: 2018 May 
 
Specialization: Machine Learning 
 
Course list 
2014 Fall  : KBAI, CN 
2015 Spring: CCA 
2015 Fall  : IOS 
2016 Spring: CP 
2016 Fall  : ML4T 
2017 Spring: DVA 
2017 Summer: HCI 
2017 Fall  : ML 
2018 Spring: RL 
 
 
########################### 
###  (1) application   #### 
########################### 
 
## 
##  statement of purpose / background essay 
## 
- this is really the bread and butter of your application. i've seen in many online forums, such as google plus, reddit, etc where very qualified applicants (like CS undergrad degree from a top notch school, or years of industry experience as software engineer at companies like big four) got rejected because their essay didn't address the most important question: why you want this degree. 
 
- i simply wrote about my situation: at the time i worked in IT department at morgan stanley, working on market data systems. it required deep CS expertise around the implementation/maintenance of ultra low latency data distribution platform, how to make it robust, how to make it scale, how to monitor, etc etc. i briefly talked about how the trading business in financial industry has become essentially a technology race, and expressed my passion of the subject. also because the innovation is rapid, i didn't wanna leave for full time on-campus study, so omscs is a perfect opportunity to study while staying in the industry so i could apply what i learn into practice, blah blah. 
 
- PS: i wrote the above in 2014 spring. i believe that was the prevailing sentiment shared by many at the time, when the entire program had only a few hundred people and we had only 3 cources (they added 2 more courses when i enrolled). there were more than enough applicants who met the bar and the admission committee had to read the statement of purpose etc to make an ultimately subjective final pick. but today in 2018, where they got the capacity to handle several thousands of students and ~30 courses, it looks more like the admission committee really wants to mechanically evaluate the applicants' likelihood of success in the program, and accept anybody who is deemed at least minimally qualified and promising, based on tangible criteria like CS degree with a sane GPA, accreditted CS courses, relevant industry experience. obviously that's the point of omscs, accept and educate people in a larger scale than it was possible in traditional on-campus program. I am curious though, how certain metrics like retention rate, grade distribution, withdrawal rate, etc are changing as the institute began accepting in mass. but it is also important to note that these traditional metrics don't quite relevantly apply to onilne campus programs. (more on this later) 
 
 
## 
##  transcripts 
## 
- there is no specific GPA cutoff threshold as far as i hear. 
- obviously admission committee will take into considaration various factors, e.g. if your GPA is 4.0 in history in some unknown college, then it may still be regarded less competitive than 3.1 GPA in CS from Stanford. 
- also overall GPA can be deceiving so they may look at how you did in CS/maths courses in particular. 
- the only concerning case is if you have a super low GPA from a less known univ. in a less relevant major, but that can be compensated with industry experience plus evidence of you having taken accredited courses, as many people argue. 
 
## 
##  recommendation letters (LoR) 
## 
- i carefully chose my portfolio. 
(1) my thesis advisor from my undergrad days, who can speak for my demonstrated research ability. i had a couple of reviewed conference papers published thanks to his tutelage. 
(2) an assistant prof. from cmu cs dept. i spent a year as an exchange student there during my undergrad year, so i wrote to the instructor from the systems programming course. i got him to write about my academic CS aptitude. 
(3) my supervisor at morgan stanley, IT department. i got him to write about my demonstrated skill in industry, applying CS concepts, technology into practice, etc etc. 
 
- unless your recommender is super motivated and supportive, usually it helps them if you provide a nice template so they can expand on it. 
 
## 
##  resume 
## 
- one common question people ask is "should I put my MOOC enrollment in my academic history?" - it will not bear much value unless you provide some certificateof completion. as it has been discussed and answered billion times in reddit, it is far more desirable to get course credits from accredited institutions like your local community college. 
 
## 
##  toefl 
## 
- i think what's important is that the admission committee sees your english is ok. out of max 120 toefl score, hopefully you get more than 100. imho, anything above 90 should be ok. but usually they figure out how good/bad you are from your essay, so make sure you keep your grammar/style impeccable. if an american student makes a typo, it is construed as a harmless human mistake, but if an international student makes a typo, it can easily be interpreted as linguistic incompetence. 
 
## 
##  waiting 
## 
- i know it feels like an eternity, i checked my emails 1000 times every day. just stay patient. 
- also, i know many students who got in after getting rejected the first time. so keep applying while improving your application. 
 
 
############################ 
###  (2) course review   ### 
############################ 
 
for each course, i will assign the overall difficulty on a scale of 1 to 10.  (1 easiest, to 10 hardest) 
 
note: "difficulty" here is a mixture of different aspects. 
- intricacy/complexity of the subject matter 
- sheer amount of work 
- harshness of grading 
- vagueness of homework/assignments instruction 
 
 
## 
##  KBAI: 4.5      # fall 2014    (cs7637 - knowledge based artificial intelligence) 
## 
- can be time consuming but not hard. 
- this is a survey course into KBAI topics. at many points it feels like a cognitive science class, not so much about CS, which is not necessarily a bad thing as you get to study how those disciplines overlap in the context of AI. no hard maths, no hard algorithms, but just broadly covering various concepts, some of which are rather abstract. 
- weekly assignment: you can write a couple of pages about what you learnt about the topic that week, how you might go about applying it to solving KBAI problems. usually you get full marks. 
- midterm/final: very broadly defined questions where you can write about anything and you normally get 95+% easily. it's take-home format. 
- 4 projects: all about solving the RPM problem. they test your solver against in-sample and out-of-sample data to decide the score. quite a lot of programming in python|java. project can be super time consuming. it took me 40hrs on the first one, then approx. 10hr for each of the rest. in a way, you can work on it without understanding KBAI concepts. (I heard they cut down to 3 projects in later iteration of this course) 
- because of super lenient grading, it becomes an easy A. 
 
## 
##  CN: 1.0        # fall 2014    (cs6250 - computer networking) 
## 
- an easy A. 
- great lecture videos, sometimes covering advanced sophisticated stuff. to really digest the content, it requires significant time commitment. 
- 11 assignemts. each consists of mini python programming plus a few quizes, requiring a few hours at most. 
-- programming part is about doing some experiments with mininet NW simulator. the only potentially tricky part is to get the hang of somewhat obscure mininet API but TA will clarify bits and pieces in piazza. 
-- quiz part is about key concepts from the lecture/paper, sort of. 
- since the assignments were too easy (you could complete them without really watching the lecture videos, because it's all about figuring out miminet API), i hear they added exams, etc to increase the load in later iteration. 
 
## 
##  CCA: 9.5       # spring 2015    (formerly cs6505, now cs6515 - computability, complexity and algorithms) 
## 
- very hard, but not impossible. you definitely need time commitment unless you are a maths connoissuer. 
- it is really a theory course. i was expecting an advanced version of different algorithms and their computation complexity, etc but no, the course really deeply goes into the theory side of things that often have absolutely nothing to do with coding, but rather set theory, probability, formal proof by induction, etc. 
- lecture videos are of commendable quality. despite hostile reviews online, i thought they covered enough ground/details to get you started on assignments. 
- 8 assignments: 33% of the class grade. very hard, easily takes 20hrs each. some assignments are pure maths problems you write them on paper and scan to submit or other assignments contain a bit of python problems but still the gut of the problem is about theoretical aspects. 
- midterm/final: 33% of the class grade for each. again, very hard, but they apply a super lenient curve so in the end most students get a B. 
- so after so many people bleeding to death, gatech redesigned the course to be less about complexity/computability, and more about algorithms, which i think is pertinent to the goal of many students in omscs, they are more inclined to learn techniques to help with their industry career, instead of preparing for phd. so it seems. 
 
## 
##  IOS: 8.0       # fall 2015    (cs6200 - intro to OS) 
## 
- very hard. 
- good lecture videos, covering various fundamental OS concepts/principles. 
- lots of heavy C programming assignments, which i think are well designed. you must get familiar with debug tools like valgrind, and use the full debugging capability of proper IDE like Eclipse to help t-shoot. you cannot get by with simple embedded print statements. 
-- specifically 4 projects. only the second one is relatively light, the rest is all mammoth. 
- midterm/final. they are hard, you really need to remember the details from the lecture as well as the papers. closed book/notes. fortunately they apply a generous curve to adjust. 
 
## 
##  CP: 6.0        # spring 2016   (CS6475 - computational photography) 
## 
- relatively easy. can be a bit time consuming. 
- lecture videos are light but broad. they don't go deep into the theoretical aspects, which are researved for the other course "computer vision", so i hear from others. 
- 11 assignments: relatively easy. most take only a few hours. several hours at most. mostly implementing python code to do some sort of image processing you learnt in the lecture. it can be time consuming or tricky because to get good expected results, you need pertinent photos. e.g. for edge detection, you need photos that at least have good edges for your edge processing python code/library. as for the coding itself, they provide step-by-step instruction of what to do plus complete unit testing script, so technically you can get it done without understanding any concepts. no real heavy coding involved. mostly getting the right openCV functions/parameters running properly. 
- final: not trivial. actually quite hard. closed book/notes. you have to review both lectures and papers. 
- project: coming up with a good project is hard as there is a boundless list of topics and doing a substantial work can be very time consuming. but the grading is lenient enough. 
- portfolio: mostly cut and paste of all the assignment results. 
- one big aspect of this course is the 90% grading rule, where you do everything asked for, then you only get 90%. to get the remaining 10%, you must go above and beyond, but that turned out to be quite subjective depending on what TAs you get. so overall, this course is easy to get a B, but rather not so easy to get an A. 
 
## 
##  ML4T: 3.5      # fall 2016    (cs7646 - machine learning for trading) 
## 
- an easy A. not because it's rudimentary, but mostly because they are very clear on what they want you to implement, and let you try the autograder they use for grading. 
- a good mix of finance + ML. well crafted videos. 
- they spend quite a bit of time(first 2 weeks) and resource at the beginning to train you up on python pandas/numpy (plus a bit on scipy). 
- assignments are well designed. they build on prev assignments to get to a fairly interesting ML trading engine toward the end of the semester. each assignment relatively easy but still you can learn a lot about the concept, as well as exposing you to hands-on details of pandas/numpy. 
- midterm: fairly easy. watching lecture videos and studying a practice question bank are more than enough to ensure 90%+. closed book/notes. in my semester no final, but the prof sounded like he might add a final in place of the final project. 
- because the course touches on both finance & ML, you don't go deep into either subject. so in a way it's a good intro course for both subjects. 
 
## 
##  DVA: 5.5     # spring 2017    (cse6242 - data visualization and analytics) 
## 
- a lot of work. not difficult but it can be time consuming: processing/restructuring your original data in R for diff visualization and analysis, running the test/model-training with different parameters, etc, comparing the results. the whole exercise can be time consuming. 
- relatively light in the lecture video length: approx 60~90 min each for 6 topics: (1) R programming, (2) visualization, (3) data processing, (4) logistic regression, (5) linear regression, (6) regularization 
- work: 3 homework assignments (10% each), 2 projects(20% each), 1 final exam(30%), plus a few extra credit activities (1% each). 
-- i thought they were all reasonably well crafted to expose students to, and get hands-on on, the relevant concepts/techniques from the lecture. 
-- there is no real difference between a homework and a project. supposedly a project is a bigger engagement but hw3 was much much bigger than pr2, for example. 
-- the final is open-book, closed-internet, on proctortrack. it was not impossible. easy to get over 80. 
- grading is not too strict, maybe on a relatively generous side. but this kinda depends on which TA grades you. 
 
## 
##  HCI: 4.0     # summer 2017    (cs6750 - human computer interaction) 
## 
- a good summer class. 
- 8 semi-weekly reports. 5% each. you write some essay using the vocabulary/concepts from the lecture videos. here we basically go thru the entire design cycle of user needfinding, brainstorming, prototyping, evaluation. some assignments involve conducting actual surveys/interviews in the real world, so it can be time consuming. 
- participation credit: 10% of the course grade. 3 sources of points: 1. peer review of class mates weekly reports. 2. piazza contribution. 3. helping other class mates surveys/interviews. every students is assigned several tokens which he/she can give out to class mates who help with surveys/interviews. so you collect tokens from your class mates, where each token is worth 0.1% of the course grade. 
- exam: midterm and final on proctortrack. 12.5% each. relatively easy. open book, open internet. no live communication with other human being. the meadian score was like 89. 
- individual project: this is essentially nothing but another weekly report to me, but worth 12.5% of the course grade. 
- team project: 12.5%. as i described above, weekly assignment lets you go thru the design cycle, then you re-do the whole design cycle, this time, as a team. obviously, there is an element of luck in getting good team mates. I was blessed with the best. 
 
- [aside] when i write course reviews, i try not to review instructor per se. admittedly instructor influences the course a lot, in terms of the assignments, exams, grading standard, etc etc. so the same course can be a drastically different experience once the instructor changes. but for OMSCS, at least the lecture video stays the same, and that, to a certain extent, keeps the core content/learning material of the course consistent across various instructors that might come and go. and obviously i try to keep my reviews focused on that very core part, and instructor-agnostic. but for this course, i must say, it will never be the same without david joyner, original developer of the course, as our instructor. he is so well engaged, so prepared, so positively passionate about teaching, and the nicest person you will ever know. 
 
## 
##  ML: 8.0    # fall 2017    (cs7641 - machine learning) 
## 
- hard because of its extremely open ended assignments. 
- ML is definitely the fad of the decade in CS both academia and industry. 
- unique in its delivery format. two profs "chatting" through each topic, which is great because if one explains in a way that is even slightly obscure then the other prof immediately asks questions for more clarity. the downside is the lecture video length tends to be very prolonged. it feels like often you see they just joke around for 30 seconds, and come back to the lecture. 
- the sheer amount of topics covered in the lecture is of biblical proportion. the total video length is like 28 hours. as the instructor says in the lecture, the goal is to get as much breadth as possible with enough depth so students can go out there and dig in any particular field. 
- this is one of those classes where lecture videos only give you a starting point, just help you build an intuitive understanding, then you have to dive into literature to study theoretical aspects deeper and more formally. it is an intensive survey course after all. 
- it is notorious for open-ended assignments in which you are graded solely for the analysis and not the code. 
- exams are hard in that the prof doesn't care about derivation of theorems and formulas, but only cares about synthesis, i.e. how you can connects different parts of the lecture together. 
- a rather generous curve at the end. 
 
## 
##  RL: 7.0  # spring 2018    (cs7642 - reinforcement learning) 
## 
- a good course. well polished (in its 7th or so iteration by the time i took it, many seasoned/helpful TAs too). 
- 6 homeworks (5% each), 3 projects (15% each), 1 final exam (25%). every hw has autograder for students to verify their answers. it's so easy to get stuck on certain bits and pieces and waste hours/days before you come to an 'aha' moment. projects are hard, and requires you to thoroughly read papers, understand and implement algorithms, replicate experiments and analyze. 
- TA grading standard is not so lenient. but not entirely unreasonable. I guess it depends on which TA you get. 
- RL is a hard subject. but unlike the ML class, this class has a much narrower specialized scope and each assignment is very specific as to what they want you to do. 
 
 
############################# 
####  (3) misc thoughts  #### 
############################# 
 
## 
##  logistics - time allocation 
## 
- the mere act of watching and digesting the lecture videos is already an enormous commitment of time and mental energy. so never underestimate the workload when reviews say light workload and easy assignment. 
 
- as you calculate how much time you think you will be able to allocate, always think in terms of "mentally fresh and productive" hours, not any theoretically free hours. for example, if you work all day, drive home, take care of kids/family, do some other chores, then even if you have 2 free hours before going to bed, at this point your brain might be fried. you may have time to watch some TV/movie, listen to music, read books, but maybe not able to work on linear algebra in high dimensional vector space in the SVM algo. 
 
- for anybody working more than 40 hours a week, and if you care about being able to do other things in life like going out for drinks/dinner with friends/family, exercising, sleeping, reading books, watching netflix, etc, then only take one course per semester. 
 
- summer semester is tough. i believe summer = 13 weeks while regular semester = 17 weeks. so no matter what course you take, it is really a lot of work to digest the same content in a much condensed time window. 
 
- i know a letter grade C can be counted toward satisfying the requirement for graduation as long as it's not part of his/her specialization core courses. but you need GPA 3.0 in the end, so you will need an A for every C, which can complicate the situation. you don't wanna be in a situation where you must get an A in a hard course to compensate for a C in other course where getting a B in that course would've been far easier than getting an A in the current course. so i think we should always aim for at least B in every class. 
 
- the required minimum amount of time and mental energy commitment differs (obviously) for each grade. generally you have to work harder to earn an A than to earn a B. but by how much? this really depends on the course (and its instructor/TA grading scheme). and this is where people might wanna get strategic. 
 
here are a couple of intuitive simplified examples. 
 
course ABC: 
grade | the relative min amount of commitment required 
------------------------------------------------------ 
   A  |  1.3 
   B  |  1.0 
   C  |  0.7 
   D  |  0.0 
 
course XYZ: 
grade | the relative min amount of commitment required 
------------------------------------------------------ 
   A  |  2.5 
   B  |  1.0 
   C  |  0.5 
   D  |  0.0 
 
==> it's not like you can precisely aim for 1.0, so you usually commit 1.1 ~ 1.2 or so, then it makes sense to invest a little extra bit of effort to obtain an A in the course ABC, but it's not gonna be the same story in the course XYZ. 
 
 
## 
##  logistics - online delivery 
## 
- i think enough technology is there. not just video archiving/streaming, but the whole online class forum like piazza, and other administrative portals for class registration,announcement,assignments etc which on-campus students also use. even with exam proctoring, they can video monitor, so it's rigorously invigilated. 
 
- but the conundrum with online class seems to be its sheer massive size. if 500 students are enrolled in a class instead of 50, then online forums will be flooded with so many question threads especially before assignment due dates. sure they can hire an army of TAs to take care of these, but as an individual student, im not sure if it's the most effective forum structure. it feels overwhelming. some talk about splitting a class into multiple sections like it's done on-campus, but i believe it beats the purpose somewhat, also the tuition fee would have to go up. 
 
 
## 
##  sequence of courses to take 
## 
 
seems like many people ponder on this. i can only speak for the courses i took. 
 
good first class : hci, kbai 
- joyner class is so well structured. every assignment is so well defined. (almost to the level of handholding every step, which is not always the best thing) and grading is lenient. perfect as first class as you learn the whole dynamics of omscs platform. 
 
sequence for ML core classes: ml4t -> dva -> ml -> rl 
- ml4t is the easiest. it is a perfect intro to ml. hands on. practical. amazingly fun (prof tucker magic). 
- dva touches on more theory and details. a bit terse. R-based. but at least it gives you fiarly concrete instruction for the assignments/exams. a lot of work but at least you wont get the kind of frustration as you would from isbell ml class. 
- now you are ready to take on ml, which gives you super open-ended projects, which is exactly the intention of the prof. it forces you to really explore and synthesize the material. but it's not easy, it's painful. so my take is you can best prepare by getting the foundation from ml4t/dva combo. 
- rl : it just flows as a sequel to ml. 
 
as for CCA/GA, which is mandatory for most specializations, it is the hardest class. so you definitely dont wanna take it as your first class. make sure you take it in a semester when you can afford both time and mental energy. one note is apparently these days so many people try sign up for it, so it is almost impossible to take it in early semesters where your priority in the registration queue is lower. 
 
 
## 
##  how to prepare for course XYZ in particular / OMSCS courses in general 
## 
 
I see people ask this on reddit/google plus almost every week. obviously it ultimately depends on the courses and student background knowledge/skill. and people tend to go for someting veeeeery generic like "brush up my linear algebra/probability/discrete maths" and/or "study java/python/C" - sure it certainly helps, but in my opinion, the best way to get started on preparation is to simply start watching the lecture videos. just watch and digest as much as possible, taking notes. then you will know much more specifically what you need to review. once you finish this, the rest is just assignment/exam. so you not only prepare for the class but you actually finish a big portion of it. 
 
## 
##  can i get in without CS undergrad ? what are my chances ? 
## 
- just apply and find out. as long as you have relevant course work under you belt, you usually get accepted. 
- if you are unsure, just complete as many courses as possible. e.g. data structures and algorithms, OS/systems, computer networks, security, database, graphics, theory, machine learning, you name it. 
- there are various accredited institutions that offer those courses online. 
 
## 
##  on a more philosophical side of things 
## 
- i believe it was the dean Galil Zvi who said "why are we doing this? because we can, and we should." : indeed, omscs is really revolutionary in its endeavour, providing higher education online, to mass audience, in lower price. of course, there is always a commercial aspect to its administration, to be sustainable. omscs certainly turned out to be a success. it is personally fulfilling for me to be able to continue to study in school even after age 30 while maintaining my professional career and family life. omscs should inspire more positive changes in education. 
 
- classmates: school is where many people meet their life long best friends. while it is not as easy to make true friends online, i have made friends with a couple of class mates who I will keep in touch after graduation. 
 
 
## 
##  metrics to evaluate the program 
## 
- traditional metrics like retention rate, grade distribution, withdrawal rate, don't quite apply the same way to measure the online campus program. because for on campus program, students are mostly full time and making far more significant commitment than online students. on-campus students have to relocate to attend physical classroom lectures, paying far bigger tuition. whereas online students in contrast are often part-time, and usually at a much older age with full time jobs, often with family, even with kids sometimes, and smaller tuition because of online platform. For a 23yr old full time on-campus student, classes are his/her biggest commitment. he/she has to do well, hoping to get a good internship opportunity and a job upon graduation. whereas for many online students, often they have to prioritize their job, family, and may have to sacrifice their class, the cost of quitting is not nearly as great as it is for on-campus students. 
 
- assuming the degree obtained thru online campus is equally rigorous, which gatech is working hard to ensure, a good absolute metric may be simply how many students graduate from online platform. because that's really the spirit of omscs, providing higher education to mass audience, especially for certain types of students who otherwise couldnt have pursued this academic learning opportunity. 
 
 
## 
##  other frequently discussed points 
## 
 
[quality of courses/instruction] 
 
the OMSCS program began in Spring 2014, and I joined Fall 2014, so I'm one of the earliest cohorts. one orientation email very honestly addressed that this is somewhat uncharted territory for gatech and there will be a few bumps we will run into and have to fix along the way, so please be patient, and so on. i really liked its honesty and it set the right tone and expectation. i didn't really see a major issue. sure a few glitches here and there, but nothing fundamentally specific to omscs. regardless of online or on-campus, there will always be too easy/hard classes, some poorly run class by a mean instructor, or some error in the lecture, or mistakes in the homework instruction/grading, etc, but we are all human so we just try improve next time. in terms of the lecture content, i really liked the recorded lecture, because they are very well concisely video-editted, with some nice visual effects/aids, so on. you can stop and go back and re-watch a particular section, stop and take notes or check certain keywords, and so on. it's great. 
 
an occasional rant I see from certain students is along the line of "i took the class XYZ and it was so easy thus this program is not rigorous." duh. so one paradigm im beginning to notice is at masters level, there is a lot of freedom for you to put in as much as you want to get out. for example, in reinforcement learning class, the project 2 is about openai gym lunar lander. you can kind of hack away by using a standard q learning plus some simple method to discretize continuous state space. but i saw many classmates read up on deep reinforcement learning where you use neural network to approximate the q value of input state. there is this seminal paper by google deepmind team, and also the subsequent Nature paper on double deep q learning, and so on. some students did an amazing reading/learning and doing hands-on implementation of these cutting-edge newly resarched algo. sure, while some classes are still simply mammoth injection of knowledge followed by tons of homework, at masters level you really should drive your own learning. it's just my interpretation. 
 
[lack of personal interaction] 
 
yes, you really need a good lasting personal interaction with prof if you are doing research. and it's not easy to develop that kind of personal relation online. but as far as course-work based masters program goes, online format is sufficient i think. even on-campus, if you are sitting on 5th row, 6th row or 29th row, in the class room, you are as close to the prof as someone online (a prof said something to that effect). i think however if your ultimate goal is to goto phd, omscs may not be the best option. but i know some students became TA and got to know prof well and got a letter of rec to eventually get accepted into phd. so it has happened. also i know they opened an opportunity to get a course credit for doing research if prof agrees. 
 
[quality of student] 
 
from time to time, i hear of a feedback "quality of my classmates are poor, i saw a student post this pathetic question on the forum, which shows he doesn't understand the very basic of XYZ, which degrades the program."  duh. so there are two perspectives to this. first of all, omscs intends to make gatech cs education available to wider audience. in that spirit of democratizing education, one needs to retire from the mentality of elitism of exclusivity. it's ok we accept many, as long as the academic rigor and standard are preserved, then only the worthy capable students will graduate. in other words, i believe it is perfectly healthy if the program has a higher acceptance rate and a higher dropout rate than traditional on-campus program. remember the cost of signing up for and dropping out of online program is very small. secondly, from a commercial perspective, if we accept more students who get weeded out in a couple of classes, that's good money that can potentially help bring down the tuition for surviving students. 
 
[feedback mechanism] 
 
dont let a rotten apple convince you the whole barrel is corrupt. one thing i found interesting is they take end of semester survey aka cios veeeeery seriously. it affects promotion, compensation/bonus, etc for the faculty members. instructors do change the courses based on the cios feedback from students. trust the sysem. help the system improve. 
 
[how to prevent cheating] 
i think it is not unique to the online delivery format per se, but this is an age old topic. regardless of whether it is online or on physical campus, a student can get help from someone and submit a hw as his own. as for the exam, omscs employs online exam proctoring service (used to be proctorU, now proctorTrack) that verifies student ID and essentially video tapes the exam. Before the exam, they ask you to use a mirror to make sure nobody is in the room, etc. I hear they use some computer vision analysis to detect any suspicious move to be reviewed carefully. the bottom line is don't do it. 
 
 
## 
##  other interesting stuff 
## 
- here is a video by the big cheese. (as of 2015) 
https://youtu.be/3vFopBgBKtg 
- a few interesting points i note from the video. 
-- the biggest concern originally was how to ensure quality. 
-- online course makes a prof famous. like a celebrity. 
-- most courses dont need frequnt updates, the estimate is an average course needs some refresh/update of the content every 5,6 years. 
 
-- creating an online course costs 150to300k, including everything, recording, video editting, etc. just like creating a movie. this is where at&t helped by investing 2mil. 
 
-- a faculty member who creates a course gets paid 20k, and gets paid 10k every time he/she teaches a course (regardlss of who created it), and gets paid 2500 (effectively as royalty) every time somebody else teaches it. 
 
-- the diff btwn online and on-campus admission process is GRE. GRE is completely uncorrelated to student performance or success. zero predictive power. on-campus admission process requires GRE out of habit, we should really stop it. in omscs, because we dont require GRE, we made it conditional acceptance and students need to complete two foundational courses to get fully admitted status. 
 
-- in on-campus program, only 10% of the applicants get accepted because of the capacity of the class room, but we consider 60 to 70% of applicants to be qualified and empirically learnt the 10% we accept and the next 20,30% perform equally as good if accepted. pure "lottery" and omscs wants to solve this problem. accept everyone who is qualified and deemed able to succeed. 
 
-- the majority (80+% if i heard correctly) of omscs students is us citizen, average age 35, while the majority of on-campus cs masters students is international student, average age 24, especially from india/china. in omscs, course drop out rate is 20%, twice as much as the rate of on-campus. but what does that mean? to a guy whorking full time job in california with a wife and two kids only paying $700 for an online course, dropping a course is no big deal, while dropping a course for an average on-campus student could very well mean he threw away thousands of dollars and also his visa could lose full time student status, and get deported. so comparing the same metrics like drop out rate btwn online and on-campus is a tricky thing. 
 
-- does producing more cs masters hurt school reputation? no, as long as we ensure their quality, no prob. univ of michigan is twice the size of gatech, UCLA even bigger. doesnt hurt their reputation. it's not like our society/industry is saying jeez we have so many cs graduates and dont know what to do with them. 
 

  1. 2017-05-13 10:37:21 |
  2. Category : gatech
  3. Page View:

Google Ads