---
title: "Sakana AI proposes backpropagation alternative for very deep nets"
url: https://www.parallelquant.com/posts/sakana-ai-proposes-backpropagation-alternative-for-very-deep-nets-4bb8ef
source_name: "MarkTechPost"
source_url: https://www.marktechpost.com/2026/09/14/sakana-ai-researchers-introduce-pc-alm-a-layer-local-alternative-to-backpropagation-that-trains-1000-layer-networks/
published: 2026-09-14T20:46:16.000Z
topics: ["research", "open source"]
publisher: "Parallel Quant"
---

# Sakana AI proposes backpropagation alternative for very deep nets

*2026-09-14 · Source: [MarkTechPost](https://www.marktechpost.com/2026/09/14/sakana-ai-researchers-introduce-pc-alm-a-layer-local-alternative-to-backpropagation-that-trains-1000-layer-networks/)*

Sakana AI researchers introduced PC-ALM (Augmented Lagrangian Predictive Coding), a layer-local training method that recovers backpropagation-level gradients without a global backward pass. It matches standard backprop across network widths and depths from 8 to 128 layers, and trains 1,000-layer residual networks within about 2 percentage points of backprop accuracy on MNIST. The team released MIT-licensed JAX code.

**Why it matters:** Backpropagation's need for global, sequential gradient computation is a real bottleneck for scaling and parallelizing training; a workable layer-local alternative could eventually enable more efficient or hardware-friendly training pipelines. Sakana has been prolific this year, including its recent Fugu multi-agent models, positioning itself as a research-heavy lab distinct from the big labs' product race.

**Topics:** research, open source

---
Read the original: https://www.marktechpost.com/2026/09/14/sakana-ai-researchers-introduce-pc-alm-a-layer-local-alternative-to-backpropagation-that-trains-1000-layer-networks/
Canonical: https://www.parallelquant.com/posts/sakana-ai-proposes-backpropagation-alternative-for-very-deep-nets-4bb8ef
Published by Parallel Quant — https://www.parallelquant.com