Professional Documents
Culture Documents
Nishikant (Nish) Sonwalkar have given rise to expectations that MOOCs will have a major impact
on the educational world.4,5 However, much is needed to improve the
Founder and Managing Director, Synaptic Global Learning, LLC, pedagogical effectiveness of the MOOC offerings. Adaptive learning
and Adjunct Professor of Physics, University of Massachusetts, systems that customize content to the needs of individual learners
Boston, Massachusetts. have potential to provide personalized learning experiences to the
large number of learners who intend to take the MOOC courses.
Adaptive learning systems can provide the interface necessary for
Downloaded by 106.209.186.92 from www.liebertpub.com at 03/08/19. For personal use only.
T
he MOOC movement has grown in a short duration.1,2,3 The maximize learning performance through continuous intelligent
MOOC movement has arisen to provide access to high feedback.
quality education to the remote corners of the Unites States The first task for adaptive learning systems is to allow for different
and the world. The early pronouncements regarding MOOCs ways to organize content, offering different context and perspective
for learners.12 The second task is to identify the way a learner would animations, and simulations in the five learning pathways con-
like to learn by conducting diagnostic assessments on the learning forming to the pedagogy of apprentice, incidental, inductive, de-
preference. The third task is to use assessment results to provide ductive, and discovery learning models. The third axis of the cube
continuous intelligent feedback that motivates and provides guid- indicates increased interactivity between student and teachers, and
ance to overcome concept deficiencies and maximize learning per- the fourth axis is for social learning via blogs, wiki, podcasting, and
formance. Figure 1 illustrates the four iterative processes involved in YouTube to name a few. Details of the ‘‘learning cube’’ pedagogical
an adaptive learning cycle that rapidly converge to the learning style framework are presented elsewhere.20,21,22
of an individual student. Within the AMOL system, content is presented to students based
The research conducted by cognitive psychologists13–17 indicates on the learning style of preference. For example, in the incidental
that the differences in learning styles originate from the difference in learning study, learning happens primarily within a context of case
the learning strategies each learner develops based on his/her pre- studies. Content provided by the expert is sequenced in ways that
vious learning experiences. Within the AMOL system, learning explain the events involved in the case study. For students who like to
strategies are related to the five learning pedagogies; apprentice learn this way, the content makes sense. For students who prefer to
(learning through mentor–student interaction), incidental (learning learn inductively, deductively, or through discovery, AMOL se-
through case study), inductive (learning through example), deductive quences content in ways that best facilitate the given style of
Downloaded by 106.209.186.92 from www.liebertpub.com at 03/08/19. For personal use only.
(learning through application), and discovery (learning through ex- learning. These learning strategies allow students to pursue the study
perimentation).18,19 of difficult subjects within the pedagogical environment that works
The ‘‘learning cube’’ pedagogical model suggested by the au- best for them. The five learning strategies used in the adaptive MOOC
thor18,20 in Figure 2 provides a multidimensional framework to or- pedagogical framework—apprentice, incidental, inductive, deduc-
ganize learning objects developed in text, graphics, audio, video, tive, and discovery—adhere to learning models representing
Fig. 1. The learning cycles for self-improvement in the adaptive learning system.
Fig. 2. The four-dimensional learning (4DL) framework for adaptive macroscopic scale—can now be studied from microscopic and at-
MOOCs. omistic points of view.25–27 Computational science is a critical and
highly desirable skill in growing sectors of advanced manufactur-
ing. An increasing number of engineering and science curricula are
cognitive approaches from behaviorist, constructivist, and coagni- now including modeling and simulation courses as part of under-
tivist learning theories.1–5 graduate and graduate programs. However, access to quality in-
A student using the adaptive learning system determines which struction in molecular dynamics remains hard to come by due to the
strategies work best through interaction with an inference engine shortage of well-qualified instructors and of practitioners trained in
that administers diagnostic tests for making the assessment. The the field. Dr. Sonwalkar brings to this course both his years of expe-
methodology involves using the assessment to measure performance rience teaching molecular dynamics and the benefits of his own on-
in real-time and providing continuous intelligent feedback necessary going practice in the field. As the only public university in the city of
for online learning assistance.6,7 Diagnostic quizzes are embedded Boston, The University of Massachusetts Boston extends its public
within each concept to assess the level of competency that a learner education mission by offering this free, globally available course.
has achieved. Based on the student’s results, an accelerated revision This course was designed to help learners develop a deep under-
path with intelligent suggestions may be provided to each individual standing of the fundamental concepts of computational molecular
learner. dynamics while gaining experience with the most popular methods
As illustrated in Figure 1, a learner circles through this powerful for atomistic simulations leading to discoveries in science. After
learning and revision cycle until the learner achieves the desired level completing the course successfully, learners were able to:
of competency. The adaptive learning systems:
. Understand and apply statistical mechanics and thermo-
- Provide instructional support to faculty to create multiple dynamics and concepts of thermodynamics applied to MD.
teaching pathways. . Learn computational algorithms for ordinary differential
- Deliver content with different learning pathways and reduce equations (ODE).
the dependency on ‘‘sage-on-the-stage’’ faculty lectures used in . Understand software programming and packages behind MD
most of the MOOCs today. simulations.
- Allow students to learn better by customizing their own . Apply MD simulation methods for property calculations.
learning strategies in a MOOC environment. . Select thermodynamic ensembles for complex physical problems.
- Provide increased cognitive opportunity to students to learn . Analyze MD simulation data for complex science experiments.
difficult concepts by presenting multiple perspectives on each . Prepare yourself for a career in computational science.
concept.
The course consisted of twelve modules, each made available at the
- Provide continuous feedback to students, freeing faculty to be
beginning of the week. In each week, MOOC participants explored
involved with student learners in the role of ‘‘guides on the
two main topics consisting of text content, short video lecture seg-
side’’ on social media interaction sites—blogs, chats, discus-
ments with accompanying transcripts, practice quizzes, and weekly
sions.
quizzes. In addition, students were required to practice and simulate
- Monitor learning progress of the students and provide remedial
the concepts learned using various (free) simulation software pro-
content in real time.
grams. Discussion forums were available through which MOOC
Adaptive learning systems thus support instructional learning participants were encouraged to communicate and collaborate with
strategies,20 pedagogically driven instructional design, pedagogical other learners in the course.2
24 MOOCs Forum
THE FIRST ADAPTIVE MOOC
Downloaded by 106.209.186.92 from www.liebertpub.com at 03/08/19. For personal use only.
Adaptive learning strategy. At the beginning of the course, learners had to depend on weekly autograded assignments, practice quizzes, and
were presented with a diagnostics quiz and were required to answer a few graded assessments. The continuous improvement of competency was the
questions about how they learn. This process identified each learner’s goal of the current adaptive MOOC; hence, there were two forms of as-
preferred learning strategy, based upon which each learner then was sessments that were provided to the students: (1) Nongraded practice
guided on an adapted learning path throughout the course, by which quizzes that allowed students to self-test their competency with a real-
process designers hoped to accelerate learning and improve score results.2 time intelligent feedback system that dynamically rendered revision
content in different learning strategies, and (2) graded assessments of
COURSE ASSESSMENT competency. There was no provision for a proctored exam for this course.2
The assessment rubric for MOOCs required special consideration as the
large number of participants would not allow the instructor to grade Practice quiz. The course included a graded practice quiz after
students manually. Hence the assessments both formative and summative each lecture segment. Practice quizzes were designed to assist
Table 1. Response Time and Throughput for Adaptive Mobile Learning Cloud Server for Adaptive MOOCs
RESPONSE RPS RPS THROUGHPUT THROUGHPUT TOTAL
(AVERAGE) ERRORS REQUESTS (AVERAGE) (PEAK) (AVERAGE) (PEAK) TRANSFER
HTML 0.848 0 10,090 6 11 73 KB/s 130 KB/s 111 MB
a
Other 0.341 0 92,782 59 103 2,464 KB/s 4,258 KB/s 4 GB
learners to master the concepts taught in each segment. Each practice course. The course curriculum was divided into weekly modules that
quiz was worth 10 points, and completion of all practice quizzes included 2 hours of lectures presented in 12 modules of 10 minutes
comprised 20% of a student’s overall course score. Practice quizzes each. The 10-minute modules were video recordings including in-
could be taken multiple times until a satisfactory score was achieved, structor lectures, slides, text, graphics, and presentation of the virtual
with the highest score being counted toward the overall course score. molecular dynamics laboratory. Video segments were captioned by
the Liberate Learning Consortium in a two-step process involving the
Adaptive feedback. At the end of each quiz attempt, the learner use of voice recognition software followed by hand editing accom-
was provided immediate feedback through the adaptive learning plished by staff caption-writing editors.
system in the course. The instructor highly recommended MOOC The converted and edited video caption text was then used by the
participants take all segment practice quizzes and score more than instructional design team members to create the concepts and sub-
80% on practice quizzes before taking the weekly graded assessment. concepts that became the core aMOOC content. The course content
thus derived for each 10-minute video segment was then transliter-
Weekly graded assessment. In addition to the practice quizzes, a ated to provide five learning strategies, namely, apprentice, inci-
weekly assessment was made available at the beginning of each dental, inductive, deductive, and discovery. The details of the
week. Each weekly assessment was worth 20 points. The combined learning strategies and the differentiated instructional model for the
Downloaded by 106.209.186.92 from www.liebertpub.com at 03/08/19. For personal use only.
score of all weekly assessment accounted for 80% of the grade. brain-based adaptive learning can be found elsewhere.
Weekly assessments could not be retaken. The production was successfully completed by a video production
and instructional design team from the College of Advancing and Pro-
Hands-on simulations experience. Learners used virtual molecular fessional Studies (CAPS) at the University of Massachusetts and by the
dynamics lab (VMDL) and MATLAB to derive the answers of the weekly adaptive learning development team of the Synaptic Global Learning.
questions. Software packages used in this course were made available to
registered participants at no cost for the duration of the course. A Scalable AWS Cloud Architecture
The adaptive learning module was deployed on the AMOL plat-
TEXTBOOKS form SaaS tools. An existing patented AMOL platform was modified
There weren’t any required textbooks for the course. Several books to accommodate the adaptive learning pedagogical framework and to
were suggested for reference in the course. provide ample social interaction. The AMOL software technology
platform leverages EC2, S3, and RDX web services on the LAMP stack
CERTIFICATION for the deployment of online programs worldwide.8,9 The course
Students who completed the course successfully were provided was deployed on the Amazon Web Services cloud computing with
with one of three types of certificates. The type of certificate granted auto-scaling option.9
will depend on student’s scores in the weekly assessments: The primary architectural consideration for a mission-critical ap-
. Certificate of Success: An average of 80% or higher achieved plication is to make the system completely fault tolerant. The fault
on weekly assessments with all assessments attempted and tolerance allows systems to function even if there are components in the
submitted on time.
. Certificate of Completion: An
average of 50% or higher (but
less than 80%) achieved on
weekly assessments with all as-
sessments attempted and sub-
mitted on time.
. Certificate of Participation: An
average of 35% or higher (but
less than 50%) achieved on
weekly assessments with all as-
sessments attempted and sub-
mitted on time.
Development of Adaptive
MOOC Weekly Modules
The most ambitious task for this
project was to create the adaptive
learning modules for the 12-week Fig. 4. The loading chart for AMOL aMOOC server.
26 MOOCs Forum
THE FIRST ADAPTIVE MOOC
Table 1 indicates the response time in seconds, which is less than 1 The author and his research team are currently analyzing the learner
second with 0% errors in responses made as results of request per behavior data collected during this case study, which will be pre-
second during the run from 5 users to 200 users with average requests sented in Part II of this research article.
per second as 6.
As indicated in the chart of Figure 4, the load test simulation shows Acknowledgment
a linear increase in the user login, which also has linear increases in The authors would like to thank Philip DiSalvio, Dean of Center for
the requests per second made to the AMOL server and the throughput. Advancing and Professional Studies (CAPS), and Dennis Maxey,
The trend indicates a very healthy growth in throughput response from Associate Dean of CAPS, for providing partial support for the design
the servers in response to the requests made. and development of the first aMOOC. The dedicated team from the
Figure 5 shows that the performance of the AMOL server remains University of Massachusetts under the leadership of Alan Girelli, the
error-free even with the request growing from 5 rps to 115 rps, re- team at Synaptic Global Learning under the leadership of Mamta
presenting over 6000 requests made to the server with responses Sonwalkar and Eric McDonald, and the team at the Liberated
within an average of 1 second and few requests taking longer. Learning Consortium under the leadership of Keith Bain are ac-
This load test data indicates excellent performance of the AMOL knowledged for their respective contributions to the project. Special
server as the errors to the request response remains at 0% even when thanks to Allan Girelli for providing editorial feedback, revisions, and
Downloaded by 106.209.186.92 from www.liebertpub.com at 03/08/19. For personal use only.
the load is increased from 5 concurrent users to 200 concurrent users. contributions to this article.
LOAD TEST MATRIX FOR LARGE CONCURRENT USERS Author Disclosure Statement
Numerous load tests were performed with over 25 test conditions No competing financial interests exist.
and scenarios. Load test data for 10,000 to 300,000 concurrent users
indicates following trends: REFERENCES
As shown in Table 2, the auto scaling feature of AMOL cloud 1. Sonwalkar N. ‘ Adaptive MOOC: A New Frontier in Open Education.’’ Research
architecture scales very well with the need for large number of Presentation, USDLA Annual Conference, April 2013 (www.youtube.com/sonwalkar7).
concurrent users. The EC2 instances grow from a base line of 5 to over 2. Sonwalkar N. Molecular dynamics for computational discoveries in science,
9 to accommodate concurrent users from 10,000 to 150,000. The EC2 author, designer and instructor of the first adaptive massive open online course
instances are also accompanied by database replications and load (aMOOC) for department of physics, University of Massachusetts, Boston, MA,
March 25, 2013, through June 7, 2013 (http://umb.sgleducation.com/
balancing, which ensures that the response time remains within an AdaptiveMOOC/NishMD/#.Uh4KZhv_KSo).
acceptable limit of 5 seconds and the errors remain within 3.3% level. 3. Sonwalkar N, Agarwal A. ‘‘Sustainability of Massive Open Online Courses
It is important to note that auto-scaling does not create spikes when (MOOCs) in Higher Education.’’ Keynote Presentation, Symposium at University
reaching the desired threshold, which prevents hick-ups in the response of Massachusetts, Boston, December 10, 2012 (www.youtube.com/sonwalkar7).
time. The replication is made available before the peak loads are reached, 4. Pappano L. The Year of the MOOC. Education Life, New York Times, November 2,
2012 (www.nytimes.com/2012/11/04/education/edlife/massive-open-online-
ensuring smooth transitions of users to higher number of instances.9
courses-are-multiplying-at-a-rapid-pace.html?pagewanted=all&_r=0).
5. What you need to know about MOOCs. The Chronicle of Higher Education, August
Conclusions 14, 2013 (http://chronicle.com/article/What-You-Need-to-Know-About/133475/).
The adaptive MOOC platform developed for the deployment of
6. Sonwalkar N. Massive open online courses (MOOC): Current trends. Webinar,
the first adaptive MOOC course in the areas of molecular dynamics National Distance Learning Week, USDLA, November 2012 (www.usdla.org).
for the computational discoveries in science was successful in han- 7. Sonwalkar N, Flores JG, Gardner M. ‘‘Designing for Quality and Measuring
dling the large loads of the Massive Open Online Course and con- Quality in Online Learning.’’ USDLA Research Report (prepared for Bill and
current user stress. Following conclusions can be drawn from the first Melinda Gates Foundation), March 2010.
part of this case study: 8. The State of Cloud Computing: An Infoworld Special Report. Infoworld,
February 24, 2012 (www.infoworld.com/d/cloud-computing/cloud-computing-
in-2012-infoworld-special-report-187077?source=fssr).
1. The pedagogical instructional design was modeled to accom-
9. Varia J, Mathew S. Overview of Amazon Web Services. Amazon Web Services,
modate five learning strategies. 2012 (http://d36cz9buwru1tt.cloudfront.net/AWS_Overview.pdf).
2. The AMOL tools were effective and adequate for the rapid
10. Minsky ML, Papert SA. Perceptrons (4th ed.). Cambridge, Mass.: MIT Press, 1990.
production of the adaptive MOOC course in a very difficult
11. Winston PH. Artificial Intelligence (3rd ed.). Boston: Addison-Wesley, 1992.
subject matter.
12. Sonwalkar N. Technology for adaptive learning: from one size-fits-all to
3. The cloud architecture was necessary to accommodate ex-
adaptive individualization, educause center for applied research (ECAR).
pected large loads for a MOOC, where the number of concur- Research Bulletin 2005, 7 (April 5).
rent users is a cause of concern for acceptable throughput and 13. Skinner BF. Science and Human Behavior. New York: John Wiley & Sons, 1953.
page return time for a dynamic rendering of content.
14. Kolb DA. The Learning Style Inventory Technical Manual. Boston: McBer, 1976.
4. The pedagogy and technology developed for the adaptive MOOC
15. Witkin, H.A., Moore, C.A., Goodenough, D.R., and Cox, P.W. ‘‘Field dependent and
shows great promise for the future creation and conversion of field independent cognitive styles and their educational implications’’, Review
the one-size-fits-all MOOC into effective adaptive MOOC. of Educational Research (47:1), Winter 1977, pp 1–64.
28 MOOCs Forum
THE FIRST ADAPTIVE MOOC
16. Dunn R, Dunn K. Teaching Students Through Their Individual Learning Styles: A 24. Romano A, Li J, Yip S ‘‘Atomistic Simulation of Matter Under Stress: Crossover
Practical Approach. Reston, Va.: Reston Publishing, 1978. from Hard to Soft Materials,’’ Physica A 304;11 (2002).
17. Dewey J. How We Think. New York: Dover Publications, 1997. 25. Sonwalkar N. ‘‘Molecular Dynamics of Energy Materials and Interfaces,’’
Invited Presentation, Energy Science and Engineering, IIT Bombay, September
18. Sonwalkar N. ‘ Changing the Interface of Education with Revolutionary Learning
24, 2009.
Technologies: An Effective Guide for Infusing Technology Enabled Education for
Universities and Corporations.’’ Presentation at iUniverse, New York, September 2004. 26. Shon Y, Sonwalkar N, Saka N. ‘‘Molecular Dynamics simulation of
boundary lubricated surfaces’’ J. Computer-Aided Materials Design,
19. Sonwalkar N. The sharp edge of the cube: pedagogically driven
Vol. 6, 69–80, 1999.
instructional design for online education. Syllabus December 2001, pp. 12–16
(www.campus-technology.com/article.asp?id = 5858); Sonwalkar N. 27. Sonwalkar N, Yip S, Sunder SS. A Combined Molecular-Dynamics and Raman
Changing the Interface of Education with Revolutionary Learning Spectroscopy Approach for Designing Ice-Metal Interface. Computer-Aided
Technologies. Syllabus November 2001, pp. 10–13, (http://www Materials Design 1995;2:77–100.
.campus-technology.com/article.asp?id = 5663).
20. Sonwalkar N. The Paradigm Shift for Adult Education: From Educational Slavery
to Learning Freedom of Human Brain with Synaptic Learning. In Kidd T, eds.
Online Education and Adult Learning: New Frontiers for Teaching Practice, IGI Address correspondence to:
Global Publishing, Hershey, Pa., September 2009. Nishikant Sonwalkar
21. Sonwalkar N. Adaptive individualization: the next generation of online Synaptic Global Learning
Downloaded by 106.209.186.92 from www.liebertpub.com at 03/08/19. For personal use only.