✳ NEURIPS 2026 / NEUROAI

Neural
Structural
Reasoner.

A brain-inspired architecture for
reasoning over structured knowledge.

What if the structure of a network
was the structure of its knowledge?

A NETWORK WITH SOMETHING ON ITS MIND
a thought, taking shape
structure activitydrag to wander ↔
Zixing Jia1,2,3,4 *Yuhang Pan1,2,3 *Ni Ji1,2,3 †* Equal contribution · † Corresponding author
Affiliations
  1. Beijing Institute for Brain Research, CAMS & PUMC
  2. Chinese Institute for Brain Research, Beijing
  3. Beijing Key Laboratory of Brain Science and Brain-Machine Interface
  4. Sun Yat-sen University
FOLLOW THE CONNECTIONS ↓
01 / THE QUESTIONA DIFFERENT STARTING POINT

The world has structure.
Where does it go?

Paris is in France. France is in Europe. The connection tells us something no entity can tell us alone.

Embeddings encode these relationships in numerical representations. NSR asks: what if we keep the relational structure explicit in the network itself?

✳ A SMALL THOUGHT EXPERIMENTinteractive illustration
Vector representations

Follow the arrows: Paris → France → Europe. The path is right there.

A visual analogy, not an embedding algorithm. Vector models can encode relations implicitly; the distinction here is whether the relational topology remains directly inspectable.

✳

Knowledge in the connections.
Reasoning in the activity.

That’s the idea
behind NSR.
02 / THE ARCHITECTUREINSPIRED BY THE BRAIN

Four populations.
One connected idea.

Different kinds of knowledge get different neural populations. Inspired by complementary brain functions, these populations bind entities to relations and learn how relations fit together.

FROM BIOLOGICAL INSPIRATION …
Brain regions inspiring NSR: prefrontal cortex for relational integration; anterior temporal lobe for concepts; entorhinal cortex for relational maps; hippocampus for relational memory.
E HPCZ ECR ATLC PFC
… TO A COMPUTATIONAL ARCHITECTURE
Between populationsWithin a population
E

HIPPOCAMPUS → ENTITY POPULATION

Give every entity
a place to belong.

A dedicated neuron represents each entity. Stable identities let the network refer to the same thing across different facts and queries.

EParis · EFrance · EEurope

The brain correspondences motivate the design. They are not a claim of biological equivalence.

See a triple become a circuit ↓
FROM A FACT TO A CIRCUIT

Where does a triple live?

A fact becomes a pattern of connections. Give NSR a head, a relation and a tail—then watch where each part belongs.

TRY A FACTor edit the triple above
ONE FACT, MADE EXPLICIT
Select a neuron to inspect its role.
Fixed E–Z connectionsFact-encoded connections · weight 1
01 / 04

Keep the entities.

The head and tail each have an entity neuron. E–Z connections are already fixed: each entity connects to its own relation-indexed binding neurons.

THE SAME FACT, IN REVERSE

r⁻¹ is the inverse direction of r.

CCompositions are learned across facts. Single-triple encoding leaves C unchanged.

Illustrative encoding, following §3.2 / Eq. 3. Highlights explain roles; they are not recorded activations. Encoding assigns weights directly—the animation reveals the construction in steps. Read the formulation ↗

03 / THE REASONING REAL RECORDED ACTIVATIONS

Don’t just get an answer.
Watch it happen.

A query sets the network in motion. Follow the activity from a relation, through an entity–relation binding, to a candidate answer.

These are five recorded states from the Nations model—the same data behind our prototype animation.

ENTITY READOUT?
Between populationsWithin a population
Hover a neuron to inspect its activity
R → Z → E
24 of 110 binding columns shown
01

Cue the relation.

Loading the recorded activation data…

0 → 1
activation

Recorded replay, not live inference. Three illustrative traces selected for a clear, correct top-1 readout; not an aggregate performance sample. Node activity comes from the model. Straight connections run between populations; curved connections stay within a population. The selected pathway is schematic. This E–R–Z example does not display composition-layer activity.

Download trace data ↗
04 / THE LEARNINGLOCAL CHANGES. REUSABLE STRUCTURE.

Connections that learn
to connect the dots.

Local Hebbian updates strengthen relation associations and learn compositions, without backpropagation. The connections that learn are also the connections used to reason.

01 / RELATIONS BECOME RULES
father_of+father_of→grandfather_of

A relation can be a route.

Compositions make multi-step patterns reusable. The example above illustrates the idea; below are composition patterns learned on Nations.

Learned Nations composition routes involving exportbooks, releconomicaid, relexportbooks, economicaid and embassy.
02 / STRUCTURE BECOMES ORGANIZATIONProjection of learned Nations representations shows concept-like groupings associated with historical political blocs.

Patterns emerge from experience.

Learned relational representations organize Nations entities into concept-like groups associated with historical political blocs. These are patterns in this dataset, rather than supplied group labels.

05 / THE EVIDENCEACCURACY MEETS EFFICIENCY

Accuracy matters.
So does learning time.

Competitive link prediction, with a different learning process. NSR learns through local updates, without backpropagation. Explore the accuracy and reported training time together.

LESS TIME. MORE ACCURACY. ↖Click a method to compare

Source: NeurIPS author slides, slides 8–9. Training times use the reported hardware and protocols; ratios describe those runs, not a hardware-controlled speedup. The horizontal axis uses a logarithmic scale. Models without reported times remain in the table. * denotes a baseline value quoted from prior work.

View all reported numbers
06 / THE WHOLE STORYA FEW MINUTES WITH NSR

Let’s walk through it.

Take the slides with you ↗

4 min 28 sec · English · Subtitled · 1.1× original presentation speed

KEEP THE CONNECTIONS GOING

A little structure.
A new way to reason.

IF NSR CONNECTS WITH YOUR RESEARCH
@misc{jia2026neuralstructuralreasoner,
  title={Neural Structural Reasoner: A Brain-inspired Architecture
         for Reasoning over Structured Knowledge},
  author={Zixing Jia and Yuhang Pan and Ni Ji},
  year={2026},
  eprint={2609.36620},
  archivePrefix={arXiv},
  primaryClass={cs.AI},
  url={https://arxiv.org/abs/2609.36620}
}

The original activation trace

Original prototype visualization: Cuba, timesincewar, USA, with five recorded E/R/Z activation states.

The same recorded Nations trajectory, as shown in the original visualization.

Download GIF ↗