🧠 Daily Study Log [2025-09-29]
Studied regression pipelines, blockchain consensus, and cryptography basics. Explored a new idea inspired by claw machines. Continued the Vision Transformer paper (Day 3) with focus on experiments and scaling.

📄 Coursework

• Big Data Analysis & AI Modeling
  • Learned the full pipeline of linear regression: preprocessing → modeling → evaluation (e.g., MAE).
• Blockchain & Cryptocurrency
  • Studied public vs private blockchains.
  • Learned about Proof-of-Stake consensus (EOS as an example), contrasting it with Proof-of-Work.
• Understanding Cryptography
  • Took a short quiz on classical ciphers.
  • Enjoyed applying manual calculations to encryption problems.

💡 Idea

• Proposed Claw Machine Authentic Prize Verifier.
• Motivation: Questioned whether claw machine prizes are authentic → idea to verify authenticity before attempting to grab.

📖 Paper Study – Vision Transformer (ViT)

• Day 3 – Experiments & Scaling (Summary)
  • ViT needs large-scale pretraining (e.g., JFT-300M) to surpass CNNs.
  • Outperforms ResNets on large datasets, but CNNs remain stronger on small datasets.
  • Scaling laws confirmed: larger models + more data → better performance.
  • Ablation studies show patch size, depth, width, and regularization critically influence results.
  • Limitation: Requires huge compute and datasets, less accessible than CNNs.

✅ TL;DR

📍 Learned regression pipelines and evaluation metrics
📍 Compared PoW vs PoS blockchains (EOS as example)
📍 Took a fun quiz on classical cryptography
📍 New idea: claw machine prize authenticity checker
📍 ViT (Day 3): Scaling laws proven, ViT strong with large data but costly