Deep Residual Learning for Image Recognition

Details

Photo

—

Name

—

Field

Machine learning / computer vision

Year

2015

Lead author

Kaiming He

Core idea

Introduces residual networks (ResNets), using skip connections to train very deep neural nets effectively.

Why it hooked me

The trick is conceptually simple but unlocked a whole new performance tier—my favorite kind of hack.

Best for

Getting a feel for how modern deep nets are actually architected, not just buzzworded.

Length feel

Conference‑paper length; figures and results tables do heavy lifting.

Vibe

Efficient, empirical, no drama—just “here’s a clever idea and the leaderboard receipts.”

Nerdy flex

Referencing ResNets when someone says “AI is basically magic” is deeply therapeutic.