Human Machine Learning Symbiosis Kenneth R. Walsh Associate Professor Department of Management and Marketing College of Business, University of New Orleans New Orleans, Louisiana Md Tamjidul Hoque Assistant Professor Department of Computer Science College of Sciences, University of New Orleans New Orleans, Louisiana Kim H. Williams Director and Associate Professor Lester E. Kabacoff School of Hotel, Restaurant and Tourism Administration College of Business, University of New Orleans New Orleans, Louisiana ABSTRACT Human Machine Learning Symbiosis is a cooperative system where both the human learner and the ma- chine learner learn from each other to create an effective and efficient learning environment adapted to the needs of the human learner. Such a system can be used in online learning modules so that the modules adapt to each learner’s learning state both in terms of knowledge and motivation. This paper describes the benefits of such a system and a proposed design that integrates human learning in both the cognitive and affective domains with machine learning which adapts to both. Introduction pend effort to learn which requires motivation. The hu- man teacher is adept at providing motivational input to Learning can be viewed as the transformation from a cur- learners along with the content itself. However, teachers rent state of knowledge and abilities to an improved state come at a cost of the teacher themselves and their infra- of knowledge and abilities. Humans learn through a wide structure. In a society with many potential learners, the variety of artifacts such as computers, books, real-world potential for teaching costs can be extraordinary. interaction and teachers taking them from a given state to In many online learning environments, computer learning another. Effective learning artifacts help take the learner tools are used to augment or replace the teacher in order to from where they are to a new state, however, each human increase the availability or decrease the cost of the learn- exists at a unique state of knowledge and abilities. ing experience. However, as the tool becomes available to Human teachers are exceptional tools for learners because more learners, its design assumptions about the current of their ability to adapt to the state of the learner. A tu- state of the learner may become further off the mark mak- tor helping a single learner can be effective often because ing the tool less effective. Further, as the learner learns they can take the time to understand the individual learn- he or she may outgrow the tool or fall behind the tool. er and what would help them progress. The teacher in a Attempts to make adaptive tool, where based on learner classroom of similar learners can engage the learners in a responses, the tool can present more advanced informa- learning exercise that helps them all. However, learners in tion to those who have mastered certain levels and pres- the same class may be similar, but are not the same and ent remedial material to learners who are not progressing the teacher adapts adjusting the experience or address- widens the range of learners that can be served. The cost ing learners individually. In larger classes, the teacher now shifts to the teacher’s ability to redesign tools with has more of a challenge adapting to individual learners. the many paths that learners may need. Since these de- Complicating the process is the human learner must ex- signs can be applied to many learners, they can be used Journal of Learning in Higher Education 55 Kenneth R. Walsh, Md Tamjidul Hoque, Kim H. Williams Human Machine Learning Symbiosis at a lower cost per learner that the human teacher alone. learning characteristics, content, and environments and mental design where concepts within a subject area have learner will never come back to equilibrium to reach the However, the coarse grained adaptability may make learn- to apply multiple learning theoretical perspectives. been interleaved. For example, Kornell and Bjork (2008) positive state and will become frustrated and may disen- ing less efficient to the individual learners when compared found interleaving helped students learn how to differen- gage. With too simple a task, the learner will not need to The behaviorists helped our understanding of the effect of to direct teacher interaction. Further, such systems are tiate painting from artists with similar styles. Little has leave the equilibrium state to begin with and will have a reward systems on behavior and on how contingencies, or usually weak at motivating the student. been done asking the broader question of how different minimal feeling of accomplishment. Making the applica- partial results, could be used to develop the learner to the subjects should be interleaved and how the timing of tion of gaging learning activities difficult is that the ap- Machine learning is a method of computer problem solv- desired behavioral level (Skinner, 1968). Although later learning activities across subjects should be conducted. A propriate level of difficulty varies widely by the learner. ing whereby the explicit structure of the problem is not theorist criticized the early behaviorist work, much was machine learning approach that uses input across subjects coded by the programmer, but rather is discovered by the still informed by the basic principles of motivation, learn- The balancing of equilibrium through learning activity or academic classes could yield useful information on the machine by analyzing data over time. In complex problem ing, and rewards, the sophistication of these constructs and difficulty is mediated by feedback mechanism. To broader question of interleaving activities. Rohrer (2012) solving, machine learning can be more cost effective than has grown considerably over time. Skinner (1968) noted the extent that feedback is given, it can encourage the notes that the research on interleaving has been limited traditional computer algorithm design because the hu- that although some aspects of human learning appear to learner to exert more effort to complete a more difficult to short term learning activities and simple patterns of in- man programmer spends less time with the details of the have simple stimulus and response relationship that may exercise. Feedback can also reinforce the success so that terleaving. A machine learning approach has the ability problem structure and allows the computer to discover be amendable to straight forward curriculum program- the learner experiences the expected positive affect. Some to consider a wider range of data and may be able to find that structure. This can be a computer intensive process, ming, those stimuli alone would not constitute effective researcher such a Graesser et al. (2008) found confusion larger time scale patterns of interleaving. Further, in a real but with falling computer prices the economics more and teaching. He observed “A good program does lead the stu- to be a similar construct to disequilibrium as a prerequi- world setting, patterns of interleaving may be at the whim more justify letting the computer explore the solution dent step by step, each step within his range, and he usu- site for learning. For example, Graesser et al. (2008) found of student habits and, therefore, data available to machine space over a human programmer explicitly testing the ally understand it before moving on; but programming is that an automated tutoring system could engender confu- learning systems may be biased based on social norms, cul- combinations. much more than this” (Chapter 4, page 3). These obser- sion in the learner when giving hints designed to make the ture, and current practice. vations lead to the need for learning systems to address learner think further and would reduce confusion when Embedding machine learning in online learning modules learners at multiple levels and from multiple theoretical the tutor gave specific facts. Graesser et al. also found that has the potential for modules to adapt to greater degrees perspectives. Effect of Affect positive feedback from the automated tutor when a learn- and more individualistically to learner’s unique character- er grasped a concept increased the students eureka emo- istics than traditional structured learning tools. Further, The affective state of the learner changes their ability to tion. Their research was exploratory but the results seem their lower cost can increase the availability of such tech- Cognitive Learning use, focus upon, and learn from learning systems. Affec- promising and worthy of further research. They also note niques to a wider audience. Developing a learning ma- tive state at any one time may be more or less conducive The spacing between learning activities has been shown the eureka concept is intended to be a major breakthrough chine that is symbiotic with the human learner is at the to learning and may change dramatically for one over to change the effectiveness of learning activities. For ex- in learning although it was measured here as a relatively heart of new learning systems that may greatly accelerate time. The degree to which the learners affective state ef- ample, students “cramming” for a test may show a short small breakthrough and, therefore, the study may have learning while increasing availability and decreasing costs. fect learning also varies by individual. Alternatively, the term effectiveness, but the memory level may decay quick- been more accurately measuring delight, rather than eu- use of a learning system can change the affective state of ly thereafter. In general, dividing study time over multiple reka. Nevertheless, strong correlation between comments the learner. For example, particularly challenging learn- Previous Research session increases its effectiveness (Carpenter et al., 2012). from the automated tutoring system and the learners af- ing task that the learner is not prepared for may disheart- The implication for a machine learning system is that fect were measured. en the learner and reduce their confidence in successfully using only the learning activities and outcomes as input Human Learning mastering such material. On the other hand successfully Bosch and D’Mello (in press) studied an automatic tu- would be insufficient. When and in what order the learner completing a learning exercise can boost confidence for toring system for teaching computer programming and Human learning involves the acquisition of new knowl- participates in the activities effects outcomes and needs to future activities. found the affective states of confusion and frustration fol- edge and skills through effort put forth by the learner. be included as inputs to a machine learning model so that lowing learner errors and those states were lessoned when The effectiveness of learning activities is effected by both the machine learning system can find the best timing and During deep learning experiences, that is when learning the system gave them guidance. Shute et al (2015) studied the current state of knowledge and skills of the learner combination of activities. about something novel and difficult, learners are put in the video game Physics Playground, designed to support and learner’s motivation to put forth effort to change to a state of disequilibrium whereby what they understand How learning activities are interleaved can dramatically physics education and found frustration lead to higher improve those states. A number of paradigms on learn- does not match well with the new material being present- change their impact on the learner (Rohrer, 2012). For ex- performance in the game and ultimately to higher post ing research have emerged historically and have been fo- ed. The state of disequilibrium leads to the negative affec- ample, most learning environments will have more than test scores in the subject matter. DiMello et al. (2014) sug- cused through scientific methods, particularly in the last tive states of “confusion, frustration, boredom, curiosity, one activity around a learning objective. A learning design gest that confusion, when introduced properly and when century and a half. Each new family of research was able and anxiety” (p. 14) and may be a necessary part of learn- where several learning activities around one concept are resolved properly, can have a beneficial effect on learning. to ask new questions and put the human learner in new ing (Graesser and D’Mello 2011). Positive affect is usually completed sequentially, before moving on to another con- light showing another characteristic of the complex way not felt by learners until they have moved back into a state A learner’s level of disequilibrium or frustration in the cept may not be as effective as alternating learning activi- in which humans learn. What is interesting is that each of equilibrium relative to the learning material and have moment of an education experience influences outcome; ties between the two concepts. Concepts that are similar new wave tended to add new knowledge by creating new overcome a hurdle or succeeded in an objective (Graesser however, general student traits that are more persistent can easily be confused and can be difficult to differenti- methods and perspectives that added depth to our under- and D’Mello 2011). Graesser and D’Mello interpret the over time, also influence a learner’s ability to engage in a ate between and such an interleaved learning approach standing of learning. However, applying multiple perspec- Csikzentmihalyi concept of flow as situation where many learning activity. Galla et al. (2014) developed the Aca- can help the learner understand the differences. The im- tive to activities has been a challenge since so many factors cycles of disequilibrium and equilibrium follow together. demic Diligence Task as a measure of self-control in an plication for a machine learning system is that input on need to be considered in real time. In fact, the complex de- A consequence of this interpretation of flow is that the academic setting. They found it “demonstrated incre- past student performance should be known temporally cision making of the human teacher is still one of the most right level of disequilibrium needs to be introduced so mental predictive validity for objectively measured GPA, so that ordering effects can be learned by the system. The important learning tools. A machine learning approach is that the learner can have the positive experience of return- standardized math and reading achievement test scores, interleaving that has been studied has focused on experi- a step toward systematically considering a wide range of ing to the equilibrium state. With too difficult a task, the 56 Spring 2017 (Volume 13 Issue 1) Journal of Learning in Higher Education 57 Kenneth R. Walsh, Md Tamjidul Hoque, Kim H. Williams Human Machine Learning Symbiosis high school graduation, and college enrollment, over and and the behavioral and social elements include participa- full course can be delivered to a student in an interactive to place additional information to bridge the gap if con- beyond demographics and intelligence” (p. 2). tion by current and potential customers, both within and and intelligent manner. necting steps are missing for a student to go to the next outside of exchange situations (Vivek et al. 2012). Poten- level of challenging questions. EM will also include tests Gamification is the process of using game-like elements tial or current customers build experience-based relation- Teachers’ Perspective and quizzes. such as points, badges, challenges, and levels of difficulty ships through intense participation with the brand by way to encourage people to act and boost customer participa- A teacher or an instructor will be able to transfer his/her Architecture of the IIEDS: The engine of the IIEDS will of unique experiences they have with the offerings and tion. Its significance has become increasingly important teaching material in IIEDS’s required format. Once the be built based on Machine Learning (ML) techniques and activities of the organization (Vivek et al. 2012). in the corporate sector, and it is forecasted to be a sub- input is given, then in the absence of the teacher, IIEDS will incorporate the following features: stantial portion of social media marketing budgets in As aforementioned, gamification is a tool that organiza- will guide and engage a student learn and help solve an ▶ Based on the collection of the behavioral entries the future (Findlay and Alberts 2011). Gamification has tions may use to promote customer engagement. Because exercise effectively. and response-features such as various mouse-clicks come to involve studying and identifying natural human CE involves eliciting cognitive, affective, social and be- and responses, amount of time to get to a particular tendencies and employing game-like mechanisms to give havioral responses from consumers, effective gamification Modules of IIEDS level, lesson delivery pattern and timing: inter- customers a sense that they are having fun while working efforts must be successful at engendering these same reac- The IIEDS will have two (02) modules: (a) Lecture Deliv- leaved or non-interleaved delivery of the similar toward a rewards-based goal. An example of gamification tions. Vivek et al. (2012) suggested that participation and ery Module (LDM) and (b) Exercise Module (EM). These topics (as discussed in the Cognitive Learning sec- would include Nike Plus, an online community that mo- involvement are key requisites to CE. Implicit in partici- methods are described below. tion of this article), and success and failure rate per tivates individuals to exercise more by enabling players to pation and involvement are cognitive, affective, social and questions per level etc. will be recorded and use as earn points and set goals. Gamification lessons are anoth- behavioral components. Thus, this research suggests that Lecture Delivery Module (LDM) features in the proposed ML approach. er way to understand the feedback mechanisms that could both participation and involvement are essential com- be used by a machine learning system as a feedback tool. ponents to successful gamification initiatives. Further, it To deliver, lecture-slides will be readout by the software ▶ Based on the computed (using Extra-Tree classifier proposes that gamification tools can not only be affective for the students. Student should be able to pause, repeat, (Geurts and Wehenkel, 2006) and/or TensonFlow In a business context, the potential value of gamification is at engaging consumers in the business environment, but and fast-forward as well as will be able to click the high- (Abadi, et al., 2015) effective feature-sets will be an increased level of customer engagement. Customer en- such tools can also be effective at creating student engage- lighted terms and jargon to check the related information determined. The feature-selection step will not gagement facilitates repeated interactions that strengthen ment in the classroom. The study that follows investigates for further details, as needed – which could be supplied only help the next steps of ML but also will help us the emotional, psychological and physical investment a the efficacy of two instructional methods in creating stu- beforehand or, can be supplied from Internet (links and identify the key features involved in the student’s customer has in a product offering or brand (Brodie et al. dent engagement, one in which gamification techniques readouts) to be explored by the interested students. learning. 2011). This research proposes that the same principles of were employed and the other in which a traditional lec- gamification and customer engagement used in industry The module will record the behavior of the student, sug- ▶ Based on the (effective) feature-set, a classifier ture format was enlisted. The details regarding the design can be applied to the classroom setting, particularly with gest further reading and information and will ask ques- will be built which will classify student’s current of the study, along with its findings, are discussed next. respect to student engagement. Student engagement has tions to raise intuition of the student. Student may skip performance level per lesson – we may define 10 been used to depict students’ willingness to participate or answer. For correct answers, student will be encour- different levels of performance score or grades, for in classroom room activities, including attending classes, Online Learning aged and will be asked next (deeper) questions. For wrong example. An efficient classifier such as support- submitting required work, and participating in classroom answers, the theory behind the question will be readout vector-machine (SVM) (Hsu, Chang et al. 2010) Bowen et al. (2012) found that machine-guided instruc- discussions (Natriello 1984). Students who are engaged again. If it is still wrong, the link of related information or, deep artificial-neural-net (ANN) based Tenson- tion used in a hybrid course could be used with one hour show sustained behavioral involvement in learning activi- from Internet could be provided. For having repeated Flow could be applied for multi-class classification of weekly face-to-face instruction and achieve equal learn- ties accompanied by a positive emotional tone. They select wrong answers, the instructor should be notified by the to rank the performing students appropriately. ing outcomes to a traditional course employing three tasks which cognitively challenge them, initiate action system. All these behaviors will be recorded including the hours of weekly face-to-face instruction. Bowen’s example ▶ The IIEDS itself will be a reinforced learner with a when given the opportunity, and make concerted efforts solution provided by the instructor to overcome the failing shows an increase in learning efficiency within the con- goal: what information needs to provide and when, as they participate in learning tasks (Skinner and Belmont situation. This will form the foundation of reinforcement text of students having complete certain prerequisites in how to provide better pathways to a student to help 1993; Chapman 2003). learning (Dogan and Olmez, 2015; Kaelbling and Litt- a relative homogenous educational environment and still the student become the top ranker based on game- man, 1996) (Sutton and Barto 2016), (Szepesvari 2013) Customer engagement (CE) has been defined as the “in- replies on the support of the human teacher, although at theoretic approach (Tomlin, Lygeros, and Sastry, implemented via machine learning techniques (Rashid et tensity of customer participation with both representa- a reduced level. These results beg the question how can 2000) as well as reinforcement learning based al., 2015; Iqbal and Hoque, 2015) for both IIEDS and the tives of the organization and with other customers in a such learning opportunities become more effective and approaches. Top-ranking target can be defined by students. collaborative knowledge exchange process” (Wagner and less costly. setting the goal to score ≥ 90%, for example. Majchrzak 2007, p. 20). CE manifests in an individual’s Exercise Module (EM) participation in and connection with an organization’s of- Toward a Symbiotic Model of Training of IIEDS ferings and activities (Van Doorn et al. 2010; Vivek et a;. This EM module will be invoked or, independently start- Human and Machine Learning To train IIEDS, it will simply need to be used by students 2012). Bowden (2009) viewed customer engagement as a ed at the end of each section of the lecture. Here, ques- – the more it is used, the more it will obtain the experi- psychological process comprising cognitive and emotional tions and solutions will be delivered in the order from easy ences and will be able to provide effective as well as need- aspects. Further, Bowden proposed that CE is an iterative Proposed Machine Learning based Learning to hard or, as predicted by the software based on the expe- based-variable pathways or suggestions to the students process, beginning with customer satisfaction and culmi- rience (generated from the Machine Learning technique Tools based on their individual feature-parameter values. nating in customer loyalty. ran in the background) – the behavior of the students Our proposed Interactive and Intelligent Education De- such as how fast he is answering what level of questions, CE may be manifested cognitively, affectively, behavior- livery System (IIEDS) is a software-tool, through which a correctness and how he is slowing down, etc. will be re- ally, or socially. The cognitive and affective elements of coded. Necessary steps will to be taken by the instructor CE incorporate the experiences and feelings of customers, 58 Spring 2017 (Volume 13 Issue 1) Journal of Learning in Higher Education 59 Kenneth R. Walsh, Md Tamjidul Hoque, Kim H. Williams Human Machine Learning Symbiosis Utilization of the IIEDS Tool allowing for both proof of concept to test feasibility of Equation 2 Conclusion IIEDS can be used in both synchronous and asynchro- technology and behavior measures to measure efficacy of Course Design Cost Efficiency nous modes. It will be interesting to see what different system on outcomes (Nunamaker, 1991). The methodol- “New ideas about ways to facilitate learning—and about experience IIEDS can get from the synchronous versus ogy is important to this study both because we will be cre- who is most capable of learning—can powerfully affect asynchronous users – which can also help justify better ating new never tried environments and because the fast (cid:1832)(cid:1829)(cid:3400)(cid:1832)(cid:1830)(cid:1834)(cid:3047)(cid:3030) (cid:1832)(cid:1829)(cid:3400)(cid:1832)(cid:1830)(cid:1834)(cid:3030)(cid:3047)(cid:3039)(cid:3397)(cid:1830)(cid:1829)(cid:3400)(cid:1830)(cid:1830)(cid:1834)(cid:3030)(cid:3047)(cid:3039) the quality of people’s lives” (NRC, 2000, p. 5). Achieving mode. Train IIEDS using synchronous users to generate pace of technology change can be taken advantage of in On t(cid:1830)h(cid:1857)e(cid:1871)(cid:1861) (cid:1859)o(cid:1866)t(cid:3)h(cid:1857)(cid:1858)e(cid:1858)r (cid:1861)h(cid:1855)(cid:1857)a(cid:1866)n(cid:1855)(cid:1877)d,(cid:3404) t(cid:3)he effi(cid:1829) cien(cid:3)(cid:1874)c(cid:1871)y(cid:3) of the stu(cid:1829)dent balancing human computer symbiosis has the potential to drastically and capture intelligent moves and then allow asynchro- iterations of the test cycle. school, work, and family is important as well. A challenge change availability and efficiency of advanced education. nous user to use the mature IIEDS, for example, and this with traditional teaching formats for students is the time The machine learning approach allows for the consider- can turn into an effective learning approach. Efficiency Outcome Measures commitment of meeting at a particular time and place ation of many more variables simultaneously in the both for class. Students must therefore consider both cost of the design of learning systems and the design of research Expectation from IIEDS One measure of efficiency is course design efficiency tuition ad time. Time can be divided into the two catego- on such systems. Since human learning is influenced by a which is the cost of course design with the value. A num- ries, time spent on synchronous activities and time spent wide range of competing factors, this approach may find IIEDS is a learner, and being a learner IIEDS will cap- ber of related measures can be developed as a comparison on asynchronous activities. Time spent on synchronous new interactions between factors leading to richer learn- ture effective and intelligent moves by the users – thus, between traditional course design approaches, faculty in- activities can be divided into time spent on same place ing environments. IIEDS will be an excellent tool to store the collective ef- tensive online course design, and Connected Thinking synchronous activities and different place synchronous forts which can keep growing richer by the usage – and Furthering science in human computer symbiosis will re- Lab design approaches. The Connected Thinking Lab activities. Synchronous same place time is often the most in return, IIEDS can deliver most suitable pathways for quire multi-disciplinary approaches to better understand design approach pairs a course designer with a faculty expensive time for students because they must forgo time a student based on the student’s need determined by the the human learning process and how artifacts such as ma- member in the design of multimedia content, student as- at work or with family and must travel to the location. performance and feature-parameter values. Eventually, chine learning impact the human learner. For the whole sessment, and collaborative exercises. If done well, faculty Synchronous distance classes reduce travel cost, but still IIEDS can be regarded as a personal teacher, standing by system to work in concert, theories from the cognitive sci- will make better use of their time contributing as subject have opportunity costs while asynchronous activities al- the student to provide encouragement as well as assistance ences, education, and computer sciences need to be inte- matter experts as course designers efficiently craft arti- low students to schedule learning activities around work as needed. grated and evaluated concurrently. facts. The hope would be that time and cost saved of the and family commitments. faculty member is greater than that of the course designer. Student Costs: Enhancement of the Intelligence of IIEDS Equation 1 shows the time efficiency of course design us- References The IIEDS can be made more powerful by enhancing its ing traditional methods vs Connected Think Lab meth- ▶ Tuition (T) intelligent and capacity to scale. Primarily, IIDES will ods. Equation 2 shows the cost efficiency of course de- Abadi, M., et al., TensorFlow:Large-Scale Machine Learn- collect several optimal sequences of actions via reinforce- sign using traditional methods vs Connected Think Lab ▶ Student time in asynchronous learning activities ing on Heterogeneous Distributed Systems. 2015, Google (STA) ment learning that helped students achieve higher score. methods. This model measures the efficiency of methods Bosch, N., & D’Mello, S. K. (in press). The Affective Expe- The dataset will be invaluable in generating more cre- in two ways. First, the study will compare design times ▶ Student time in synchronous distant learning rience of Novice Computer Programmers. International ative pathways from the samples. Utilizing short schema of new methods to traditional methods. Secondly, it will activities (STSD) Journal of Artificial Intelligence in Education. (Hoque, Chetty et al. 2007) or, short action-steps from compare how new methods design efficiency changes over ▶ Student time in synchronous face-to-face learning Bowen, William G., Chingos, Mathew M., Lack, Kelly A. the collected successful action-sequences, novel and in- time to capture the likely learning curve effective of ap- activities (STSF) and Nygren, Thomas I. 2012. “Interactive Learning On- teresting pathways can be generated fast and intelligently plication of refined design processes. line at Public Universities: Evidence from Randomized using our effective evolutionary algorithm (Hoque and Where the magnitude of the costs can be ordered base on Trials,” ITHAKA. Measures that can contribute to efficiency calculation in- Iqbal 2015). These pathways can then be cross-validated the early discussion as: clude: Brodie, R.J., Hollebeek, L.D., Juric, B. and Ilic, A. (2011), using IIEDS again. Classroom modules then will be redesigned to either in- “Customer engagement: Conceptual domain, funda- ▶ Faculty design hours in a traditional course As the feature-space of IIEDS is expected to be very high, mental propositions, and implications for research,” R3 naturally scalability can be a concern while enhancing the (FDHtc) Equation 3 submitted to the Journal of Service Research. intelligence of IIEDS. Fortunately, we have already devel- ▶ Faculty design hours in Connect Thinking Lab Relative Costs of Student Time Carpenter, Shana K., Cepeda, Nicholas J., Rohrer, Doug, oped novel approach, named hGRGA (Iqbal and Hoque course (FDHctl) Kang, Sean H. K., and Pashler, Harold (2012). “Using 2016), to handle such scalability issues within our evolu- Spacing to Enhance Diverse Forms of Learning: Review ▶ Course designer design hours in Connected Think- tionary approach. The idea will be transformed for this crease the efficienc(cid:1845)y(cid:1846) o(cid:1827)f (cid:3407)a s(cid:1845)t(cid:1846)u(cid:1845)d(cid:1830)en(cid:3407)t t(cid:1845)i(cid:1846)m(cid:1845)e(cid:1832) or shift the activity of Recent Research and Implications for Instruction,” ing Lab course (DDHctl) IIEDS application. Thus, this overall recurrent approach to a lower cost time period. Educational Psychology Review, 24:369–378. can make the IIEDS grow its intelligence effectively. ▶ Course (C) Other efficiency measures in the assessment include Chapman, E. (2003). Alternative approaches to assessing ▶ Faculty cost (FC) course delivery time efficiency, course delivery cost effi- student engagement rates Practical Assessment, Research & Evaluation, 8(13). Retrieved 7/2/07 from http:// A Build and Learn Methodology ciency, which can be measured from both the university ▶ Course designer cost (DC) PAREonline.net/getvn.asp?v=8&n=13 and student perspective, as well as design and delivery ef- Understanding levels of affect in real time and adapting ficiency normalized on a per student basis. D’Mello, S. K., Lehman, B. Pekrun, R., & Graesser, A. appropriately has the potential to greatly improve the ef- Equation 1 C. (2014). Confusion Can be Beneficial For Learning, fectiveness learning environments. Learning & Instruction, 29(1), 153-170. Course Design Time Efficiency The build and learn; evaluate and learn methodology in- Doğan, B. and T. Ölmez, A novel state space representation tegrates systems development with the scientific method for the solution of 2D-HP proteinfolding problem using (cid:1832)(cid:1830)(cid:1834)(cid:3047)(cid:3030) (cid:1832)(cid:1830)(cid:1834)(cid:3030)(cid:3047)(cid:3039)(cid:3397)(cid:1830)(cid:1830)(cid:1834)(cid:3030)(cid:3047)(cid:3039) (cid:1830)(cid:1857)(cid:1871)(cid:1861)(cid:1859)(cid:1866)(cid:3)(cid:1857)(cid:1858)(cid:1858)(cid:1861)(cid:1855)(cid:1857)(cid:1866)(cid:1855)(cid:1877)(cid:3404)(cid:3) (cid:3)(cid:1874)(cid:1871)(cid:3) (cid:1829) (cid:1829) 60 Spring 2017 (Volume 13 Issue 1) Journal of Learning in Higher Education 61 Kenneth R. Walsh, Md Tamjidul Hoque, Kim H. Williams reinforcement learning methods. Applied Soft Comput- schools. Journal of Research and Development in Edu- ing, Elsevier, 2015. 26: p. 213–223. cation, 17, 14-24. Findlay, K. and Alberts, K. (2011) Gamification: How Nunamaker Jr., J. F. 1992. “Build and learn, evaluate and Effective is It? Proceedings of ESOMAR Congress 2011 learn,” Informatica, 1(1), 1-6 conference. Rashid, M.A., et al., An Enhanced Genetic Algorithm for Galla, B., Plummer, B., White, R., Meketon, D., D’Mello, Ab initio Protein Structure Prediction. IEEE Transac- S. K., & Duckworth, A. (2014). The Academic Diligence tions on Evolutionary Computation, 2015. PP(99). Task (ADT): Assessing Individual Differences in Effort Rohrer, Doug, 2012. “Interleaving Helps Students Distin- on Tedious but Important Schoolwork, Contemporary guish among Similar Concepts,” Educational Psychology Educational Psychology, 39(4), 314-325. Review, 24:355–367. Graesser, A. and D’Mello, S. K. (2011). “Theoretical Per- Shute, V. D’Mello, S. K., Baker, R. Cho, K., Bosch, N., spective and Affect and Deep Learning,” in New Perspec- Ocumpaugh, J. Ventura, M., Almeda, V. (2015). Model- tives on Affect and Learning Technologies, R. A. Calvo ing how incoming knowledge, persistence, affective states, and S. K. D’Meloo editors. Springer: New York. and in-game progress influence student learning from an Graesser, A. C., D’Mello, S. K., Craig, S. D., Witherspoon educational game, Computers & Education, 86, 224- A., Sullins J., McDaniel B., Gholson, B., (2008). The 235. Relationship between Affective States and Dialog Pat- Skinner, B. F. (1968). The Technology of Teaching. Mer- terns during Interactions with AutoTutor. Journal of In- edith Corporation. teractive Learning Research, 19(2), 293-312. Skinner, E. A., & Belmont, M. J. (1993). Motivation in the Geurts, P., D. Ernst, and L. Wehenkel, Extremely random- classroom: Reciprocal effects of teacher behavior and stu- ized trees. Machine Learning 2006. 63: p. 3–42. dent engagement across the school year. Journal of Edu- Hoque, M. T., M. Chetty and L. S. Dooley (2007), “Gener- cational Psychology, 85, 571-581. alized Schemata Theorem Incorporating Twin Removal Sutton, R. S. and A. G. Barto (2016), “Reinforcement for Protein Structure Prediction,” Pattern Recognition Learning: An Introduction,” https://www.dropbox. in Bioinformatics, Singapore, Springer. com/s/d6fyn4a5ag3atzk/bookdraft2016aug.pdf?dl=0, Hoque, M. T. and S. Iqbal (2016), “Genetic Algorithm MIT Press. based Improved Sampling for Protein Structure Predic- Syed, U. and R.E. Schapire (2008), “A Game-Theoretic Ap- tion,” International Journal of Bio-Inspired Computa- proach to Apprenticeship Learning.” tion (Accepted), http://cs.uno.edu/~tamjid/TechReport/ Sampling_TR20154.pdf. Szepesvari, C. (2013), “Algorithms for Reinforcement Learning,” Morgan & Claypool Publishers, https://sites. Hsu, C.-w., C.-c. Chang and C.-j. Lin. (2010), “A practical ualberta.ca/~szepesva/papers/RLAlgsInMDPs.pdf. guide to SVM classification” Tomlin, C.J., J. Lygeros, and S.S. Sastry, A game theoretic Iqbal, S. and M. T. Hoque (2015), Dispredict: A Predic- approach to controller design for hybrid systems. IEEE tor of Disordered Protein using Optimized RBF Kernel. 2000. 88(7): p. 949-970 PLoS One. van Doorn, Jenny, Katherine N. Lemon, Vikas Mittal, Iqbal, S. and M. T. Hoque (2016), “hGRGA: A Scalable Stephan Nass, Doreén Pick, Peter Pirner, and Peter Genetic Algorithm using Homologous Gene Schema Re- C. Verhoef (2010), “Customer Engagement Behavior: placement”, Swarm and Evolutionary Computation, El- Theoretical Foundations and Research Directions,” sevier (revision submitted), http://cs.uno.edu/~tamjid/ Journal of Service Research, 13 (3), 253–266. TechReport/hGRGA_TR20161.pdf. Vivek SD, Beatty SE, Morgan RM. (2012) Consumer en- Kaelbling, L.P., M.L. Littman, and A.W. Moore, Rein- gagement: Exploring customerrelationships beyond pur- forcement Learning: A survey. Journal of Articial Intel- chase. Marketing Theory and Practice, 20, 2, 127–145. ligence Research, 1996. 4: p. 237-285. Wagner, Christian, and Ann Majchrzak (2007), “En- Kornell, N., & Bjork, R. A. (2008). “Learning concepts and abling Customer-Centricity Using Wiki the Wiki Way,” categories: Is spacing the “enemy of induction?” Psycho- Journal of Management Information Systems, 23 (3), logical Science, 19, 585–592. 17–43. National Research Council (NRC), 2000. How People Learn: Brain, Mind, Experience, and School: Expanded Edition. Natriello, G. (1984). Problems in the evaluation of stu- dents and student disengagement from secondary 62 Spring 2017 (Volume 13 Issue 1)