Achieving Learning Objectives Online: Not All Platforms are Equal! Nick Feamster Georgia Tech A Tale of Two MOOCs • A free Coursera MOOC on Software Defined Networking to over 50,000 enrolled students – Scalable delivery, assignments, and forums • A for-credit Udacity “MOOC” on Computer Networking for Georgia Tech’s Online MS program, to 200+ enrolled students – Not so scalable… Subtle differences in platform design have significant effects on a course’s ability to scale. MOOC #1: Software Defined Networking • Extremely hot, new topic – No existing “real world” courses on the topic (yet) – A chance to develop an archetype course (and material) that others might use in their own courses – I wanted to be copied. This seemed like a good way. • Boutique topic means that there were no “set in stone” ways of teaching the course – Easier to think outside the box, since there was no existing material anywhere http://blog.sflow.com/2012/05/software-defined-networking.html Two Offerings So Far: • Summer 2013 (6 weeks) • Summer 2014 (8 weeks) Who Takes The Course? • 50,000+ people registered each year – 10,000 people active each week – 1,000-3,000 watching lectures (3,500+ in 2014) • 500+ turn in programming assignments each year Other Demographics (Summer 2014) • 55,000+ Enrolled • 197 Countries – 31% from the United States – 15% from India • Sex and Age – 86% Male – 74% 20-39 years old • Education and Employment – 58% Employed full-time – 26% Full-time students – 38% have Undergrad degree, 31% have Masters What the Instructor Sees How Was the Course Developed? • Course lesson plan, with learning objectives • Top down design: – Figure out modules, continue subdividing until you get 10-minute “lectures” • Production – Lecture Filming and Production – Assignments – Syllabus • Operation – Forums Filming Take 1: Studio • Started in a studio at Georgia Tech • This quickly proved intractable – Studio is distracting. Lots of people watching makes recording nerve-wracking. – Studio time is limited, contentious, and expensive. – Post-production quality was not to my liking. – Faculty members need scheduling autonomy. • A lot different from a TV interview! Filming Take 2: Camtasia • Studio people will claim that quality suffers. • I believe this is bogus, for several reasons. – People can take their time recording, take breaks, record when they are “in the zone”. – Screen captures from a laptop are easy. – Recording quality from a good laptop is quite fine. – Nobody has ever complained about the quality of the recordings. Content is what matters most. • Takes a little getting used to at first (looking at the camera, etc.) • You have to do your own editing. Producing Lectures in Camtasia • General rule: 10-15 minutes of lecture == 2 hours of recording and editing – This does not include time to prep slides, etc. • This is way different than a usual lecture! – Biggest lesson: Silence is quite easy to edit! – Stumbling is OK: Just say the same thing again – You can also “practice” while the video is rolling • Need to find a quiet, well-lit location • Takes patience, but it starts to be quite fun General Lecture Production Process • Determine learning objectives • Research the topic • Write slides – General format: Summary/outline, 7-10 content slides, wrap-up • Plan demonstrations – Plan on-screen landscape • • • • Record (one sitting for consistency…big challenge!) Edit and export Upload Note: You can prepare lectures out-of-order! New Modes of Lecture Production • On-screen demonstrations – Videos lend themselves very nicely to onscreen demonstrations – User can see things in “life size” – User can pause, rewind, etc. Very helpful tutorials for working through assignments • Interviews with luminaries in the field – Google “On Air” Hangouts – Can be done anywhere, and people are willing! – Process: Invite, script questions, send, revise. Stay roughly on script – Many people have since copied this idea (Nick McKeown, Alex Orso, Mike Hicks, etc.) Lessons from Lecture Production • Lots of money is not needed. Students want content, clearly delivered. Camtasia works. • A 45–60-minute lecture is about 15 minutes of well-polished video. Content distills well. • Lecturer is always “ahead” of the students. This creates interesting time-shifting effects, with advantages and disadvantages. • MOOCs are asynchronous and autonomous for the student. This is a big reason for their success. Production should be the same. How Was the Course Developed? • Course lesson plan, with learning objectives • Top down design: – Figure out modules, continue subdividing until you get 10-minute “lectures” • Production – Lecture Filming and Production – Assignments – Syllabus • Operation – Forums Developing Scalable Quizzes: Stick to the Basics • Problem: The platform is not very good at handling anything that’s not multiple choice or simple numerical answer. – Short answers become massive regular expressions – …will likely improve as the software gets better • Stick to multiple choice if possible • Main differences from a normal MC quiz – Helps to write up explanations for each option – Need to figure out if students can re-take the quiz, and if so, how many times Developing Assignments is Hard • Wanted to teach concepts and programming – Not a “heavy-duty” programming course, but some concepts are best explained and learned through short code exercises. • Code submission possible, but grading must be done on regular expression evaluation of output • Problem: How to guarantee uniform programming environments? Uniform Environment: Virtual Machine • Asked students to: – Install VirtualBox (or equivalent) virtual machine emulator – Download 64-bit VM image that had everything prepackaged (no asking students to install software themselves). Kind of like a lab at home. • This has generally worked. A few problems – VM is huge (~1 GB), so have to get it right the first time – Not everyone has a 64-bit machine – Certain performance, even when run in an emulator, varies depending on underlying hardware How Was the Course Developed? • Course lesson plan, with learning objectives • Top-down design: – Figure out modules, continue subdividing until you get 10-minute “lectures” • Production – Lecture Filming and Production – Assignments – Syllabus • Operation – Forums Communicating with 10,000 Students Can Scale! • I feared absolute disaster – Small glitches in assignments in a classroom of 50 students create mayhem. – I feared that I would be spending my life answering minutia in online forums. • Thankfully, this did not happen! • Lesson: Giving an assignment to 10,000 students in a MOOC is easier than giving it to 50 in a classroom! Why Assignments Can Scale • • • First of all, you do your best to work out the bugs, mostly out of fear – We attempted everything that we assigned, sometimes on multiple platforms More importantly: Self-selection! – Everyone doing the assignments wants to be there – This makes a huge difference – People not only fix issues themselves, they rewrite documentation Caveat: Course difficulties vs. platform difficulties are currently hard to tease apart. Some of this is “user error”. A student’s blog post describing part of the assignment setup in detail… Some Feedback… A Tale of Two MOOCs • A free Coursera MOOC on Software Defined Networking to over 50,000 enrolled students • A for-credit Udacity “MOOC” on Computer Networking for Georgia Tech’s Online MS program, to 200+ enrolled students “MOOC” #2: Online MS at GT • Online version of CS 6250: Computer Networking • Two offerings thus far (Spring and Summer 2014) – About 120+ students in each offering – Paying tuition for a “degree” – Not specialists in networking • Technical difference: Platform How Was the Course Developed? • Top-down planning, with “scripts” • Several full-time staff members – Course developer from Udacity – Video editor from Udacity • Part-time from others – Three paid TAs/Graders – Many other people handling logistics • Course recorded in studio, ahead of time, with custom equipment – Recording done in “marathon stints” – Not possible to integrate interviews • No auto-grading, little visibility of content, etc. • Initially, all material released at once (huge mistake) Comparing the Approaches Coursera (SDN MOOC) Udacity (OMS “MOOC”) Video Delivery Camtasia-recorded videos, talking head and slides + onscreen demos Tablet writing + moving hand Grading Simple MC quizzes, autograded programming assignments. No auto-grading! TAs graded assignments “manually”. Forums Initially “home brew”. Moved to Piazza. Piazza, but the students expect more direct instructor engagement. Pacing One week at a time, synchronous (though people fall behind) All-at-once release. Demographics Professionals and students taking the course for free, for enrichment. Paying students who are not necessarily interested in the subject. Entitled. Cost 300+ hours of my time, 40+ hours of TA time (all “Free”) Estimated: $150k (staff salaries, studio time, etc.) Leave the Pedagogy to the Instructors • Substance trumps style. • A MOOC platform should allow instructors to deliver the content in whatever format they see fit. • Mandating a particular “style” emphasizes style over substance. GT OMS Degree “MOOCs” Are Neither Massive nor Open • One of the important features of MOOCs that allows scaling is community support. • Things that dilute this support: – Students who pay and expect staff contact – Students who are not interested in the content, but rather trying to get a degree – “Go at your own pace” style (students are on their own, course staff can’t keep up) • Other platform-specific factors prevent scaling – Instructor has less autonomy in creation and management of content Lesson #1: MOOCs Change the Game for Instruction • Everyone is watching you – In seven years, never got feedback (or acknowledgment) about on-campus course – Now, everyone’s eyes are on you – Huge pressure to do a good job • MOOCs are a “forcing function” for introducing new media into the classroom – I wouldn’t have thought to do “on air” interviews for a classroom; people would likely decline them anyhow – I will likely re-use some MOOC content in the “real” classroom Lesson #2: Time Investment Pays Off • Typical Course Creation (45-minute lecture) – 1-2 hours to create slides – 1 hour to deliver the lecture – Number of students: 50-100 – Total: 3 hours per lecture – … but then you have to do it again! (can’t save an awesome lecture, students can’t rewatch, etc.) • MOOC Course Creation (10-minute video) – 1-2 hours to create slides – 2-3 hours to produce the lecture – Total: 3-5 hours per lecture – Number of students: 500-100,000 – … but, you have the archive. Can keep improving it, use it to enhance in-classroom discussion, etc. Lesson #3: Certain Aspects of MOOCs Scale Surprisingly Well • Lectures are watchable by thousands – The scale can be a motivator for guest lectures! • Programming assignments can be graded automatically, if designed well • Self-motivated, self-selected students fix glitches, help each other out Conclusions • Scalable MOOCs require – Community engagement – Low barriers to creating content – Support for a diversity of delivery modes to match the material • Not all platforms are equal – A platform should give the instructor autonomy and visibility. • For-credit “MOOCs” face far different scaling challenges than true MOOCs. – At the moment: neither massive, nor open