Representation Network (RN)

Novel construct introduced by this paper: a hypothetical graph embedded in the time series of LLM representations, where each dimension is a node and latent connections are edges.

Neighborhood — ranked by edge-count

Methods (4)

method

Principal components analysis (PCA)
uses
Statistical method used to analyze neural activity data.
Attended Response Representations (ARR)
uses
Time series of response representations contextualized by applying dot-product attention to the corresponding stimulus representations.
Contextually Attended Response Representations (CARR)
uses
Extension of ARR where attention is directed specifically to linguistic spans (complement syntax or mental state verbs) within the stimulus.
Spatio-Permutation Control
about
Control procedure that randomly permutes the embedding dimension order of (C)ARR to investigate latent nodal relationships in the RN; repeated 10 times.

Concepts (3)

concept

Can 'Consciousness' Be Observed from Large Language Model (LLM) Internal States? Dissecting LLM Representations Obtained from Theory of Mind Test with Integrated Information Theory and Span Representation Analysis
introduces
The primary paper being extracted — applies IIT 3.0 and 4.0 to LLM representation sequences derived from ToM test data to investigate whether consciousness phenomena can be observed.
LLM Internal Representations
about
High-dimensional vectors produced at each transformer layer for each input token; the primary substrate analyzed in this study.
Transition Probability Matrix (TPM)
uses
Matrix encoding Markovian micro-dynamics; foundational for EI computation.

Artifacts (1)

artifact

Supplementary code and data at doi.org/10.1016/j.nlp.2025.100163
about
Code, labeled linguistic spans, and augmented responses made publicly available as supplementary material.

Hypotheses (1)

hypothesis

We hypothesize that a Representation Network (RN) emerges from LLM representations, where each dimension is a node and latent connections exist between nodes or clusters of nodes.
grounds
Core methodological hypothesis enabling the application of IIT to LLM representation sequences.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

"Representation geometry is a window into the inner world of neural networks."quote0.739
The paper's concluding summary statement asserting the deep interpretive significance of representation geometry.
Gene Regulatory Networks (GRNs)concept0.735
Systems of molecular regulation exhibiting associative learning and downward causation; example of misplaced mechanistic assumptions.
Recurrent networksconcept0.732
Networks with loop connections that can maintain internal state and exhibit dynamical attractors.
Distributed Neural Representationsconcept0.727
Representations where individual neurons play multiple conceptual roles; patterns consisting of linear combinations of unit vectors.
representation manifoldconcept0.726
One-dimensional curved surface in internal activation space; the paper demonstrates alignment with behavior manifold.
Representational Alignmentconcept0.726
Measure of similarity between the similarity structures (kernels) induced by two different representations
Networks encode structured geometric concepts that reflect external reality.claim0.723
Core claim of the paper: the right level of description for neural representations is geometric structure mirroring the world.
Representation-based Pathconcept0.723
The path in activation space derived by fitting the representation manifold, used to steer along the geometric structure of internal representations.