← Research
IROS · 2025 · Shared first author

Efficient online learning with predictive coding networks

PCN-TA, a predictive-coding online learner that keeps latent states across consecutive frames of a sensory stream, cutting the cost that usually makes predictive coding expensive.

  • 10%fewer weight updates than backprop
  • 50%fewer inference steps than baseline PC
Schematic: latent state carried across frames Top row, standard predictive coding: the latent state is reset for every frame and inference runs six steps. Bottom row, PCN-TA: the latent state is passed from one frame to the next and inference runs three steps. frame tframe t+1frame t+2 Standard PC latent reset each frame 0 z 0 z 0 z PCN-TA latent carried forward z z z one inference step (counts illustrate the reported ~50% reduction)
Schematic of the idea, drawn for this page (not a figure from the paper).

Problem

Robots at the edge need online learning from streaming sensory data. Predictive coding offers local, Hebbian-like updates that suit neuromorphic hardware, but it pays for them with many inference iterations per training step.

Approach

PCN-TA preserves latent states across temporal frames, so inference on each new frame starts from the previous frame's state instead of from scratch. It exploits the temporal correlation that a robot's input stream has anyway.

Evaluation

Online learning on the COIL-20 robotic perception dataset, against backpropagation and baseline predictive-coding networks.

Result

Accuracy comparable to backpropagation with 10% fewer weight updates, and 50% fewer inference steps than baseline predictive-coding networks.

Paper

Efficient Online Learning with Predictive Coding Networks: Exploiting Temporal Correlations

Darius Masoum Zadeh-Jousdani*, Elvin Hajizada*, Eyke Hüllermeier  ·  * equal contribution

IROS 2025