Generative Artificial Intelligence: How generative AI works.pdf
ACB4 tec pre - p4 - presenting a technical paper
1. Eng. Nisansa Dilushan de Silva, Dr. Shahani Markus Weerawarana, Dr. Amal Shehan Perera
Presented by–
Nisansa Dilushan de Silva
Department of Computer Science and Engineering
University of Moratuwa
Sri Lanka
Enabling Effective Synoptic
Assessment via Algorithmic
Constitution of Review Panels
IEEE International Conference on Teaching, Assessment, and Learning for Engineering
(TALE 2013)
3. • 5th semester
• Creativity & Software Engineering rigor
• Pre-industrial training (6th semester )
• Is comparable to a mini Capstone project
• Course design straddles several program ILOs [3]
Introduction - Software Engineering
Project (CS 3202) [2]
[2] Weerawarana, S. M., Perera, A. S., Nanayakkara, V. (2012). Promoting Innovation, Creativity and
Engineering Excellence: A Case Study from Sri Lanka, IEEE International Conference on Teaching, Assessment
and Learning for Engineering 2012 (TALE2012), Hong Kong, August 2012.
[3] Anderson, L. & Krathwohl, D. A. (2001). Taxonomy for Learning, Teaching and Assessing: A Revision of
Bloom's Taxonomy of Educational Objectives New York: Longman.
4. • A large number of students (n=101)
• Assessment of the end-of-semester project demonstrations
• Technical standpoint
• Creative standpoint
• Needs a synoptic assessment methodology so that it
“enables students to integrate their experiences,
providing them with important opportunities to
demonstrate their creativity”[4]
Software Engineering Project-
Challenges
[4] Jackson, N. (2003). Nurturing creativity through an imaginative curriculum, Imaginative curriculum Project,
Learning and Teaching Support Network, Higher Education Academy.
5. • Student projects cover a large spectrum of technologies.
How to evaluate all of them fairly?
• Is it enough to use university academic staff only? [2]
• Involve external evaluators from the industry each being an expert of
some of the technologies used by students?
• We went with the second approach by agreeing with
Elton that assessment of creative work should be „viewed
in light of the work‟, highlighting important aspects as,
“the ability of experts to assess work in their own field of
expertise” and “the willingness to employ judgment”. [5]
Software Engineering Project-
Challenges
[2] Weerawarana, S. M., Perera, A. S., Nanayakkara, V. (2012). Promoting Innovation, Creativity and
Engineering Excellence: A Case Study from Sri Lanka, IEEE International Conference on Teaching, Assessment
and Learning for Engineering 2012 (TALE2012), Hong Kong, August 2012.
[5] Elton, L. (2005). Designing assessment for creativity: an imaginative curriculum guide. York: Higher
Education Academy (in offline archive).
6. • Is one evaluator matched according to technological
expertise enough to fairly evaluate a student project?
• Creative projects -> Evaluation is subjective
• Thus might not be a fair judgement
• We decided to use panels of evaluators in accordance to
what Balchin suggests when he states that “consensual
assessment by several judges” can be used to enhance
the reliability of subjective evaluation. [6]
Software Engineering Project-
Challenges
[6] Balachin, T. (2006). Evaluating creativity through consensual assessment, in N. Jackson, M. Oliver, M. Shaw
and J. Wisdom (eds) Developing Creativity in Higher Education: Imaginative Curriculum. Abingdon:
Routledge.
7. • Mapping individual evaluators to student projects
according to the expertise in technology was fairly an
easy task…
Creating Evaluation Panels -
Challenges
8. • But with panels it is not so easy to assign the „best fit‟ panel
of evaluators comprising external industry experts and
internal faculty members considering the technologies
used in the student‟s project.
Creating Evaluation Panels -
Challenges
?
9. • Multiple and often conflicting evaluator availability
constraints
• Balance of external versus internal evaluators in
the panels
• Minimization of the number of panel composition
reshuffles
• Avoidance of the same internal evaluator who
assessed the mid-semester project demonstration
being included in the end-semester evaluation
panel
• Preventing internal evaluators who mentored
specific projects being included in the end-
semester evaluation panel for the same projects
Creating Evaluation Panels -
Challenges
10. Introduce an algorithm to
Automate the Composition and
Scheduling Process for the
Synoptic Assessment Panels
Solution
11. Input : Student technology requests
(Matrix A)
Total Number of Student Requests= 6
13. • Internal evaluator: conflict data (Matrix C)
• Was a mentor
• Was the mid-semester evaluator
More Inputs….
I. Evaluator 1 I. Evaluator 2 ... I. Evaluator 7
Student 1 1 0 ... 0
Student 2 0 0 ... 1
... ... ... ... ...
Student101 1 1 ... 0
14. • Evaluator availability ( Matrix D )
• Some are available throughout the day
• Some arrive late
• Some plan to leave early
More Inputs….
Time slot 1 Time slot 2 ... Time slot 24
I. Evaluator 1 1 1 ... 0
I. Evaluator 2 1 1 ... 1
... ... ... ... ...
I. Evaluator 7 1 1 ... 0
E. Evaluator 1 1 1 ... 1
E. Evaluator 2 0 0 ... 1
... ... ... ... ...
E. Evaluator 9 0 1 ... 0
15. • Other constraints in place to ensure a fair
evaluation being taken place [2]
• Grading calibration
• Minimum number of evaluators in a panel
• At least one internal evaluator in a panel
More Inputs….
[2] Weerawarana, S. M., Perera, A. S., Nanayakkara, V. (2012). Promoting Innovation, Creativity and
Engineering Excellence: A Case Study from Sri Lanka, IEEE International Conference on Teaching, Assessment
and Learning for Engineering 2012 (TALE2012), Hong Kong, August 2012.
16. • Create panel slots according to the evaluator availability
Algorithm: Rough outline
18. • Create a set of panels with a randomization algorithm
• Assign panel slots to created panels
• Apply internal evaluator constraints.
• Calculate the merit value for each panel slot against
each project (Matrix F)
Algorithm: Rough outline
Panel-slot 1 Panel-slot 2 ... Panel-slot 108
Student 1 20 4 ... 0
Student 2 9 32 ... 65
... ... ... ... ...
Student101 0 19 ... 16
19. • Create the The inverted value matrix (Matrix G)
• Run the well-known combinatorial optimization algorithm
known as the “The Hungarian algorithm” [7] on the above
data. ( Results in the Boolean Matrix H)
Algorithm: Rough outline
[7] Kuhn, H. W. (1955). The Hungarian method for the assignment problem, Naval Research Logistics
Quarterly, 2:83–97
20. • The underlying randomness of the base „seed‟
evaluation panels that was fed in to the core
algorithm might introduce a certain degree of
unfairness to the system.
• To fix this we introduced a second layer of
indirection which was placed over the core
algorithm.
• Thus the result was now a pool of panel
assignment schedules instead of a single schedule.
• The best one of which was selected as the winning
schedule.
Algorithm: Removing the randomness
21. Algorithm: Removing the randomness
400
420
440
460
480
500
520
1 11 21 31 41
400
420
440
460
480
500
520
1 251 501 751 1001 1251 1501 1751
Algorithm pool results. (Left: pool
size=50, right: pool size =2000) x-axis:
epochs, y-axis: total value of assignment
schema
22. • The algorithm was able to match 120 of the total
141 feasible requests giving an 85.12% success
rate.
• The average number of requests satisfied per
student was 92.1% among the 83 students whose
requests constituted the 141 feasible requests.
• The average number of requests satisfied per
technology was 71.69% among the 18
technologies.
Results
23. • The advantage of automating the panel composition
process
• When some external evaluators make sudden changes in their
time constraints
• Cancellation of their commitment mere hours prior to the
commencement of the project demonstrations.
• In this critical situation the algorithm facilitated rapid
recalculation that produced an alternate optimal
schedule
Discussion
24. This approach can be considered as a
major improvement over the manual
assignment of panels for Synoptic
Assessment.
Conclusion
25. The future work is to implement an
online application.
Our recommendation is that other
educators could use this application
for a similar purpose.
Future work & Recommendation
26. [1] TALE 2012 Presentation, Weerawarana, S. M., Perera, A.
S., Nanayakkara, V. (2012). Promoting Innovation, Creativity and
Engineering Excellence: A Case Study from Sri Lanka, IEEE International
Conference on Teaching, Assessment and Learning for Engineering 2012
(TALE2012), Hong Kong, August 2012.
[2] Weerawarana, S. M., Perera, A. S., Nanayakkara, V. (2012). Promoting
Innovation, Creativity and Engineering Excellence: A Case Study from Sri
Lanka, IEEE International Conference on Teaching, Assessment and
Learning for Engineering 2012 (TALE2012), Hong Kong, August 2012.
[3] Anderson, L. & Krathwohl, D. A. (2001). Taxonomy for Learning, Teaching
and Assessing: A Revision of Bloom's Taxonomy of Educational
Objectives New York: Longman.
[4] Jackson, N. (2003). Nurturing creativity through an imaginative
curriculum, Imaginative curriculum Project, Learning and Teaching
Support Network, Higher Education Academy.
[5] Elton, L. (2005). Designing assessment for creativity: an imaginative
curriculum guide. York: Higher Education Academy (in offline archive).
[6] Balachin, T. (2006). Evaluating creativity through consensual
assessment, in N. Jackson, M. Oliver, M. Shaw and J. Wisdom (eds)
Developing Creativity in Higher Education: Imaginative Curriculum.
Abingdon: Routledge.
[7] Kuhn, H. W. (1955). The Hungarian method for the assignment
problem, Naval Research Logistics Quarterly, 2:83–97
References
Hinweis der Redaktion
Good EveningI’ll be explaining the research conducted for “Enabling Effective Synoptic Assessment via Algorithmic Constitution of Review Panels” by the “Department of Computer Science and Engineering University of Moratuwa, Sri Lanka
The outline of the presentation is…..
First of all let me give you a brief introduction to University of Moratuwa…
The Department of Computer Science and Engineering is one of the 7 departments in the faculty of Engineering.Department of Computer Science and Engineering offers theHonours Bachelor of Science degree in Engineering.
In the 5th semesterof their studies, the students of the department have to complete a compulsory module called the Software Engineering Project. The module acts as a mini Capstone project for the students who will undergo their industrial training in the following semester. Thus the course straddles several program Intended Learning Outcomes.The students are supposed to complete a project that would enhance their creativity & Software engineering rigor.
In handling this module, as lecturers we face a number of challenges.The first is the sheer number of student projects. Since this is a compulsory module, each year we have roughly 100 student projects to grade. In this particular year in which the described algorithm was introduced, there were 101 students.Secondly since the projects are supposed to be creative, the need to be evaluated in the creative standpoint as well as the technical standpoint. This calls for a synoptic assessment methodology. “Synoptic assessment encourages students to combine elements of their learning from different parts of a programme and to show their accumulated knowledge and understanding of a topic or subject area. A synoptic assessment normally enables students to show their ability to integrate and apply their skills, knowledge and understanding with breadth and depth in the subject. It can help to test a student's capability of applying the knowledge and understanding gained in one part of a programme to increase their understanding in other parts of the programme, or across the programme as a whole. Synoptic assessment can be part of other forms of assessment.”
As explained by prior research on the same course, it is not enough just to have the internal academic staff grading these projects. We decided to involve external experts for the specific technologies that the students have used.
Then the next question is whether one evaluator matched according to technological expertise enough to fairly evaluate a student project?As you might have already guessed, in cases such as this where creativity is involved, the evaluation of anybody, be it an expert of a given technology or not, is going to be subjective. Thus this might not give us the fair judgment that we strived for.Yet again referring to prior research on creative assessment, we decided to involve panels of evaluators so that they can have “consensual assessment”
Say we have a set of student projects;And a set of evaluators;Each project having a set of technologies;Each evaluator being an expert to a set of technologies.We can come up with a mapping relatively easily
But when we have panels, it is not that easy to find the best fit panel.
In creating evaluation panels, we face some more challenges than mapping the bestfit panels to projects.
Total number of student requests (r=244)
To handle the additional requirements explained previously, we have few more inputs.
To handle the additional requirements explained previously, we have few more inputs.
To handle the additional requirements explained previously, we have few more inputs.
2. Some technologies might not have an expert. Thus total number of satisfiable student requests =141Our objective was to comeup with the optimum panel allocation so that the maximum number of these satisfiable student requests are fulfilled.
2. Some technologies might not have an expert. Thus total number of satisfiable student requests =141Our objective was to comeup with the optimum panel allocation so that the maximum number of these satisfiable student requests are fulfilled.
The reason for taking the product for the internal evaluators and the summation for the external evaluators is the fact that the internal evaluator matrix (C) is essentially a binary matrix about existence or nonexistence of constraints and conflicts and the external evaluator matrix (E) on the other hand is an integer matrix with the relevance factor of each external evaluator for each project as elements. 𝑤2- to dampen the net merit value and prevent a single panel gaining above average advantage over the others within the same time-slot category. 𝑤1- The constant weight variable was used so that the calculated gross merit value (VPro,g−pan) does not get completely subdued by the damping variable value w2. Thus the value of w2 is initiated to a variable value which is proportional to the size of the relevant time slot category and is reduced each time a certain panel is reused. 𝑤1 is set to a value which would ensure that the value obtained by the multiplication of the gross merit value and 𝑤1 will be greater than the maximum possible 𝑤2even at the minimum attainable value of the gross merit value.The reason for taking the product for the internal evaluators and the summation for the external evaluators is the fact that the internal evaluator matrix (C) is essentially a binary matrix about existence or nonexistence of constraints and conflicts and the external evaluator matrix (E) on the other hand is an integer matrix with the relevance factor of each external evaluator for each project as elements. 𝑤_2- to dampen the net merit value and prevent a single panel gaining above average advantage over the others within the same time-slot category. 𝑤_1- The constant weight variable was used so that the calculated gross merit value (V_(Pro,g−pan)) does not get completely subdued by the damping variable value〖 w〗_2. Thus the value of w_2 is initiated to a variable value which is proportional to the size of the relevant time slot category and is reduced each time a certain panel is reused. 𝑤_1 is set to a value which would ensure that the value obtained by the multiplication of the gross merit value and 𝑤_1 will be greater than the maximum possible 𝑤_2even at the minimum attainable value of the gross merit value.
2. Some technologies might not have an expert. Thus total number of satisfiable student requests =141Our objective was to come up with the optimum panel allocation so that the maximum number of these satisfiable student requests are fulfilled.