Skip to main content

Programming instead of Neural Networks

Programming instead of Neural Networks

Overview

Exploring aProject pivot discussion for the CS411 Digital Education project (Group: Kyoto baby) exploring switching the focus from AI-assisted learning of neural networks to teaching programming concepts./ algorithmic logic to novice learners.


Upcoming Meeting AgendaNotes & Decisions

Discussion on the benefits, disadvantages, and trade-offs of switching the research domain to programming.

1. PotentialScope & Course Constraints

    TA Validation Check (Urgent): Confirm with the TA (Dominique) whether the programming scope is sufficiently technical and acceptable. (Precedent: an AI project teaching foreign languages was previously rejected for being "too broad"). Pre-Validation Strategy: Submit an early draft to the TA via email ahead of the official deadline to get early feedback/green light, as the group feels ~2 weeks behind.

    2. Key Benefits

    • AutomatedSignificantly &easier Fine-Grained Evaluation:recruitment: Code execution, unit test pass rates, syntax vs. logic error categorization provide objective benchmarks.
    Student Familiarity & Recruitment: Programming tasks generally align with aVastly larger pool of EPFL/UNILprospective studentsnovice comparedparticipants toacross specializedEPFL neuraland network training intuition.UNIL. GranularFunding Telemetry:& Publication Pathway: EasyOpportunity for external project funding from a business project; professor is on board to trackturn keystrokethis cadence,into compilera errors,research AIpublication prompt(targeting queries,a andconference revisionin history.Kyoto).

    2.3. Potential DisadvantagesRisks & ChallengesPedagogical Focus

    • PriorScope Experience Confound:Management: LargeMust variance in baseline coding skills among students may confoundconstrain the measuredlearning impact of AI assistance.
    Scope & Open-Endedness: Debugging and code generation can become overly broad compared to constrained parameter tuning in a visual playground. Separating AI Help from Syntax Fixes: Risk of students relyingobjective—focusing on AIintuitive asalgorithmic a simple syntax autocompletelogic rather than engagingheavy syntax overhead. Baseline Confound: High variability in conceptualself-reported learning.programming skills.

    Experimental Protocol Design

      Study Format: Between-subjects design. Target Audience: Strictly novice / beginner programmers (screened). Pre-Test / Screening: Include an objective baseline screening test/survey prior to the experiment (do not rely on subjective self-reporting). Participant Session Budget:
        ~1h15m – 1h30m per participant (1 hour task/training + 15–30 min pre-test, post-test questionnaire & debrief). Cohort Size: ~20 participants total. Time Commitment: ~30 hours total (~2 full testing days per team member, accounting for cancellations/rescheduling). Remote logging remains a secondary backup option.

        Participant Recruitment Pipeline

          EPFL Bachelor Students: Request Mauro Cherubini to send study call to students in Éléments de programmation. UNIL Cohort: Reach out via 3 friendly UNIL professors + Yash to broadcast to students.

          Deadlines & Next Steps

          • AlignThis withWeek: teamPrepare a concise 1-page project proposal (Karo,hypotheses, Isna,independent/dependent Stergios)variables, duringmeasurement metrics) for Mauro and the meeting on whether to pivot or maintain the NN playground setup.TA.
          • DefineTA targetedPre-check: programmingEmail conceptsthe 1-pager to Dominique / exercisesTA iffor pivotearly isclearance approved.before the milestone.
          • ReviewMilestone recruitment1 requirementsSubmission: October 8th (MichalisOfficial /validation Mauroof studentproject cohorts)idea).
          • UpdateDesign experimentalScreening protocolInstrument: (GroupCreate Athe vsshort Grouppre-test Bquiz intervention).to filter true novices.