NeuroAI Multimodal Intelligence Workshop at AAAI 2026 · Workshop paper

Decoding Cortical Microcircuits: A Generative Model for Latent Space Exploration and Controlled Synthesis

Xingyu Liu, Yubin Li, Guozhang Chen

A generative latent model compresses mouse cortical microcircuit connectomes to explore structure-function relationships and synthesize bio-plausible networks.

The scientific question

Can high-dimensional cortical microcircuits be represented and synthesized through a compact latent code?

Directly analyzing structure-function links in massive connectome graphs is intractable, though developmental biology suggests circuits arise from compressed genetic blueprints.

Overall model structure for decoding and synthesizing cortical microcircuits.
Figure 1 from the preprint ↗

The core idea

Generative Microcircuit Latent Compression

We train a variational generative model on mouse cortical microcircuit connectivity maps, linking positions in the learned latent space to computational performance on reservoir tasks.

What we found

Controlled Bio-Plausible Circuit Synthesis

The compressed latent space captures essential microcircuit structure, associates encoded topologies with computational functions on reservoir tasks, and enables controllable generation of novel synthetic microcircuits with desired structural features.

Continue the question

Related reading

Explore our research

Cite this work

Xingyu Liu, Yubin Li, Guozhang Chen. Decoding Cortical Microcircuits: A Generative Model for Latent Space Exploration and Controlled Synthesis. NeuroAI Multimodal Intelligence Workshop at AAAI 2026.