Sakana AI Builds a Brain-Like Neural Network That Learns Without Backpropagation
A new paper accepted at ALIFE 2026 shows biologically constrained neural networks can learn competitively — without backpropagation's weight transport trick — across image classification and reinforcement learning.
- What: Researchers at Sakana AI published a paper (ALIFE 2026) showing Dale's-principle-compliant neural networks can learn competitively without backpropagation's weight transport.
- Method: "Diffusing Blame" uses a dual-stream excitatory/inhibitory architecture with four non-negative weight matrices per layer and a new modulo error routing scheme for multi-class problems.
- Classification results: 96.7% on MNIST and 61.7% on CIFAR-10 — the first time Error Diffusion has worked on convolutional networks, though 7.4 points below DFA on CIFAR-10.
- RL results: ED-PPO matches or beats standard backprop-PPO on HalfCheetah and Ant, but trails on Humanoid and Craftax with higher variance.
- Emergent biology: Networks spontaneously develop balanced excitatory/inhibitory ratios and depth-dependent inhibitory gradients mirroring real cortical circuits — with no explicit mechanism enforcing this.
- Key caveat: The dual-stream design requires ~4x more parameters; 37.3% of weights collapse to the non-negative floor, and no public code was released.
Every neuron in your brain is either excitatory or inhibitory, and it never switches sides. This is Dale's principle, a fundamental rule of biological neural circuits. Standard artificial neural networks ignore it entirely, freely mixing positive and negative weights on the same unit. A new paper accepted at ALIFE 2026, authored by Yutaro Yamada, Luca Grillotti, Rujikorn Charakorn, Sebastian Risi, David Ha, and Robert Tjarko Lange (all at Sakana AI), asks a pointed question: can a network that strictly obeys Dale's principle still learn well? Their answer, detailed in Diffusing Blame: Task-Dependent Credit Assignment in Biologically Plausible Dual-Stream Networks, is yes, with some important caveats.
Two biological rules backprop breaks
Standard deep learning simultaneously violates two properties of biological neural circuits.
- Dale's principle: Each neuron releases the same neurotransmitter at all its synapses, making it uniformly excitatory or inhibitory. Artificial networks freely assign positive and negative weights to any connection, which enables efficient credit assignment via backpropagation but has no biological counterpart.
- The weight transport problem: Backpropagation requires the backward pass to use exact transposes of the forward weight matrices. The brain has no known mechanism for copying weight matrices backward through layers.
Prior methods such as Feedback Alignment and Direct Feedback Alignment address weight transport by replacing backward weights with fixed random matrices, but none enforce Dale's principle. The authors set out to fix both problems at once.
The dual-stream architecture
The core idea is to split every layer into two parallel streams: one excitatory (p) and one inhibitory (n). Error Diffusion (ED), originally proposed for this kind of dual-stream architecture, drives learning by routing global error signals to all layers without transporting transposed forward weights or relying on random feedback matrices.
Each layer maintains four non-negative weight matrices: excitatory-to-excitatory (W_pp), inhibitory-to-excitatory (W_np), inhibitory-to-inhibitory (