🧠 Daily Study Log [2025-07-10]
Wasn’t feeling too great today, so I kept the pace light.
Still managed to stay consistent with TOEIC, read Day 2 of EfficientNet, and pushed a new SCU_Competition submission forward.

📊 SCU_Competition — Submission #78

Focus:

Added new derived features while keeping the winning structure (VotingClassifier 6:2:2)
Minor adjustments to feature importance threshold for pruning
Used same clustering features from best model (#73)

Result:

✅ Local AUC slightly improved
✅ Kaggle AUC still very stable — no major breakthroughs but steady performance

Reflection:

Feature richness is key, not excessive weight tuning
Will try Feature Selection with feature importance next

🧠 TOEIC Study — Daily Routine

Tasks Completed:

Covered the scheduled units from both listening and grammar books
Focused on pacing and answer consistency

Reflection:

Sticking to small daily progress is keeping the pressure low
Plan to finish at least 5 more units this week

📘 Paper Study — EfficientNet (Day 2)

Focus:

Studied architecture details and scaling coefficients (ϕ, α, β, γ)
Understood how EfficientNet scales depth, width, and resolution in sync
Took notes on B0–B7 variants and scaling rationale

Reflection:

The compound scaling logic is elegant and intuitive
Excited to explore the NAS-based model discovery in the next part

❌ Skipped: CV Chapter 2

Unfortunately had to skip today’s CV study due to fatigue.
Planning to resume tomorrow with hands-on experiments from Chapter 2.

🔍 Tomorrow’s Plan

✅ Continue SCU_Competition submission tuning (try Feature Selection)
✅ EfficientNet Day 3: Experiments & Results
✅ Resume CV study — backprop code and basic layer implementations
✅ TOEIC: maintain daily progress

✅ TL;DR

📍 SCU: Submission #78 — stable performance, more feature tuning ahead
📍 TOEIC: Kept up with listening and grammar units
📍 Paper: EfficientNet Day 2 — compound scaling mechanics covered
📍 CV: Skipped today — back on track tomorrow
📍 Tomorrow: Tune, train, and catch up on skipped CV work