Seminar Participation Tracker 2026

[algomaster99]

Here we track active participation in lectures.

To do this, you record as a comment the question you make to presentations or demos during the seminar.

Also, provide the title of the presentation/demo.

Lecture Participation Stats (Updated on 2026-09-11 13:47:03)

Index	Student Name	Number of Lectures Attended	Lecture(s) attended
1	nalin431	1	S2
2	tctev	1	S2
3	a-runebou	1	S2
4	terahidro2003	1	S2
5	Jeecis	1	S2
6	Hasan-Kalzi	1	S2
7	bog1200	1	S2
8	riccardo-fragale	1	S2
9	Elpa430	1	S2
10	eryngia	1	S2
11	f-r00t	1	S2
12	einhar1	1	S2
13	LeuisKen	1	S2
14	lolindgr	1	S2
15	AlexWigren	1	S2
16	sangeetha-murugesan-sda9	1	S2
17	MiTO-X2	1	S2
18	Kristerooo	1	S2
19	ramikhd	1	S2
20	TheArtemis	1	S2
21	EccirilloM	1	S2
22	derfeshMariush	1	S2

[algomaster99]

See an example from previous year: #2697.

[nalin431]

Question related to FlakyGuard - Scientific Paper - #2935 :

Q: Why does flakyguard use "bfs" as opposed to a different traversal method?

A: Uses BFS because it's collecting information by each layer, as the LLM is deciding which children of each node are relevant enough to keep exploring. BFS is better because it fully explores this context at each level. Could theoretically use DFS but is just more inefficient

[tctev]

Question related to FlakyGuard: Automatically Fixing Flaky Tests at Industry Scale - Scientific Paper - #2935

Q: What happens when the real root cause isn’t present in the dynamic call graph?

A: FlakyGuard may not have enough context to find or fix the issue. This is one reason it only manages to fix around 47% of the cases. At Uber’s scale, even that can still save a lot of engineering time. They also rerun tests many times to verify that the test is actually flaky and that the generated patch fixes it.

[a-runebou]

Question related to: What helped, and what did not? An Evaluation of the Strategies to Improve Continuous Integration - Scientific Paper - #2952

Q: Is this meant for larger projects, or would it work for a newer project as well?

A: There might be some overhead to implement it in newer projects, even if it is possible. It is primarily used by larger CI pipelines like the one Google uses.

[terahidro2003]

Question related to What helped, and what did not? - Scientific Paper - #2952

Q: How exactly selective test execution maps the code to the individual tests and what happens if some integration tests span across multiple related components? (Deeper execution stack trace)?

A: There is a cost calculated for each test that takes into account certain benefit features (e.g., prioritization of tests with minimal execution time) versus risk of not-executed test not covering some important changes.

[Jeecis]

Question related to #2952 - Scientific Paper - #2952

Q: Are there any companies that use your presented hybrid or similar approaches in the production environment?

A: We don't know, there is probably a lot of research going on right now, and considering the good cost saving results of the study some form of them are implemented in some existing pipelines.

[Hasan-Kalzi]

Question related to #2940 Scientific paper

Q: How did the researchers evaluate whether the generated tests were actually good at detecting faults rather than simply increasing code coverage?

A: They mainly evaluated the generated tests using statement and branch coverage and compared the results with existing test-generation approaches. They did not directly measure fault-detection effectiveness, which is a limitation of the evaluation.

[bog1200]

Question related to #2940

Q: The paper discusses the TESTPILOT algorithm, which uses an iterative loop of documentation gathering, processing and test generation. On average how many iterations would it take to get acceptable tests?

A: The number of iterations varied widely between the tested packages, but on average each step fixed about 50% of the errors.

[riccardo-fragale]

Question related to An Empirical Evaluation of Using Large Language Models for Automated Unit Test Generation, week 2 - #2940 Scientific Paper:

Q: Given that in the iteration loop they use LLMs also as judges, does the paper mention possibility or limitations of introducing humans in the loop?

A: The papers mentions that most of the LLM errors when generating tests were very simple and easily identifiable. Moreover, in the future research part of the paper they say that they might try to introduce humans somewhere in the loop a human which can hep the iteration loop to be faster and to improve the whole procedure. As a consequence we believe that introducing a human in the loop will very likely reduce the time needed to reach a good result.

[Elpa430]

Question related to #2952 - Scientific Paper - #2952

Q: Is there a way to combine all tree techniques to give a better outcome?

A: Yes there is a one* to cover all of them but it depends on the project, it is hard to cover all the projects.

*He mentioned the name of it but i did not remember the name

[eryngia]

Presentation: FlakyGuard: Automatically Fixing Flaky Tests at Industry Scale - Scientific Paper #2935
Question: Did the authors discuss any particular contexts in which the depth-first traversal is particularly effective, or in which a different traversal method would be more suitable?
Answer: The authors did not discuss exact contexts in significant detail, however their strategy is much better than previous works when the root cause is many layers deep in the call graph. Furthermore, limiting the number of nodes after searching is very helpful when there are a lot of potentially relevant methods.

[f-r00t]

Presentation: Improving Test Quality with Mutation Testing in a CI Workflow
Q: What is the point of mutation testing?
A: The main idea of mutation testing is too see whether tests are accurate and specific enough to fail even small (potentially breaking) changes to the code.

[Hasan-Kalzi]

Question about #2962 Demo

Q: Given the execution cost of mutation testing, would you run it on every pull request or only periodically, and what would determine that choice?

A: They chose to run it periodically rather than every pull request because the mutation test's cost will increase significantly.