TildAlice – Tech Blog for Developers
Python, AI, Deep Learning, Data Analysis, Computer Vision, Reinforcement Learning โ tips and tutorials
Latest Posts
-
Knowledge Distillation ์ค์ : LLM ์์ถ๊ณผ TinyLlama ์ฌ๋ก
Compress BERT to 40% size with 97% accuracy retained. Knowledge Distillation guide: DistilBERT and TinyLlama cases with PyTorch implementation.
-
World Model ๊ธฐ๋ฐ ๊ฐํํ์ต: Dreamer v3์ MBPO๋ก ์ํ ํจ์จ 10๋ฐฐ ๋์ด๊ธฐ
Dreamer v3 and MBPO boost RL sample efficiency 10ร. World Models learn environment dynamicsโRSSM architecture and ensemble tricks explained.
-
ViT vs Swin vs MaxViT: ์ด๋ฏธ์ง ๋ถ๋ฅ ํธ๋์คํฌ๋จธ ๋น๊ต
ViT vs Swin vs MaxViT: which Vision Transformer wins? Hierarchical attention benchmarks on ImageNet with code for classification and detection.
-
KAN vs MLP: Kolmogorov-Arnold Networks ์ฐจ์ธ๋ ์ ๊ฒฝ๋ง์ ์ํ์ ์๋ฆฌ์ ์ฑ๋ฅ ๋น๊ต
KAN uses 100ร fewer parameters than MLP for function approximation. Kolmogorov-Arnold Networks: learnable activation functions on edges explained.
-
์ฝ๋ฉํ ์คํธ ํ์ ์ ํ: ํฌ ํฌ์ธํฐ(Two Pointers) ์๋ฒฝ ๊ฐ์ด๋ – ์ฐ์ต ๋ฌธ์ ์ ์์ธ ํ์ด
Two Pointers solves sum problems in O(N) vs brute-force O(Nยฒ). Master two-sum, subarray sum, and three-sum with step-by-step Python solutions.
-
Edge AI ๊ธฐ๋ฐ ์ค์๊ฐ ์ง๋ ๋ถ์: Raspberry Pi์ MEMS ์ผ์๋ก ๊ตฌํํ๋ ์ ๋น์ฉ ์์ง๋ณด์ ์์คํ
Build a $70 Edge AI predictive maintenance system with Raspberry Pi and MEMS sensors. Real FFT and TinyML code included.
-
FlashAttention ๋ฆฌ๋ทฐ: Transformer ์ดํ ์ 2~4๋ฐฐ ๊ฐ์
FlashAttention cuts Transformer memory from O(Nยฒ) to O(N) while running 2-4ร faster. Here's the IO-aware tiling trick explained.
-
RetNet ๋ ผ๋ฌธ ๋ฆฌ๋ทฐ: Transformer๋ฅผ ๋์ฒดํ ๋ณ๋ ฌ ํ์ต+์์ฐจ ์ถ๋ก ์ํคํ ์ฒ ์์ ๋ถ์
RetNet achieves 8.4ร faster inference than Transformer with 70% less memory. The parallel training + sequential inference proof inside.