Today for AI
HOT RADAR
industryHEAT 11.9°

AI CLUSTERED EVENT · 10/15/2026

KAN Creator Liu Ziming's MetaCircle Raises Hundreds of Millions to Build Meta Models for Predictive Architecture Design

1 reports archived1 independent sourcesupdated 10/15/2026, 10:39:41
Synthesis & Latest Updates
1 Sources Cross-Validated

MetaCircle, founded by KAN co-author and Tsinghua professor Liu Ziming, has raised hundreds of millions in seed rounds within six months. The company is developing 'Meta Models' that predict neural network performance from architecture configurations before full training, aiming to streamline AutoML and reduce computational trial-and-error costs.

LATEST/MetaCircle, founded by KAN co-author and Tsinghua professor Liu Ziming, has raised hundreds of millions in seed rounds within six months. The company is developing 'Meta Models' that predict neural network performance from architecture configurations before full training, aiming to streamline AutoML and reduce computational trial-and-error costs.

HEAT TRENDHourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. InfoQ 中文T2·68 pts
    • Founder Liu Ziming is the first author of Kolmogorov-Arnold Networks (KAN) with a background spanning Peking University, MIT, and Stanford.
    • The core product, 'Meta Model', does not perform tasks directly but learns the mapping between architecture configurations and experimental outcomes.
    • This approach aims to filter candidate architectures via pre-training performance prediction, differing from traditional NAS/AutoML methods that rely on extensive actual runs.