finding

active

finding:rsa-shows-low-rdm-correlation-on-embedding-layers-for-gru-gru-comparisons-despite-high-within-seed-functional-similarity

RSA shows low RDM correlation on embedding layers for GRU-GRU comparisons, despite high within-seed functional similarity

Demonstrates RSA's sensitivity issue in embedding layers; attributed partly to Spearman rank handling of RDMs with differing relative extrema.

Source paper

extracted_from

Model Alignment Search

(2025) · Satchel Grant

Neighborhood — ranked by edge-count

Claims (1)

claim

Correlative methods like RSA and CKA are insufficient for determining functional similarity between neural systems; causal methods are necessary
supports
Central motivating claim of the paper; supported by empirical comparisons showing RSA/CKA miss Markovian differences detectable by MAS.

Frameworks (1)

framework

Representational Similarity Analysis (RSA)
associated_with
A correlational similarity method compared against MAS; uses RDM correlations between model representations.

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.

CKA and RSA show potentially unintuitive (over-estimated) hidden state similarity for GRU-Transformer comparisons on Multi-Object taskfinding0.817
Prior work shows transformers use anti-Markovian solutions; MAS correctly shows low IIA reflecting this, while RSA/CKA do not detect it.
MAS successfully aligns behavior between Multi-Object GRU models in both embedding and hidden state layers with high IIAfinding0.790
Demonstrates MAS's ability to bidirectionally transfer behavior where RSA shows low embedding correlation.
GRU behavior can be compressed to as few as 4 dimensions using DAS and MAS with comparable IIAsfinding0.759
Shows that behaviorally relevant information is low-dimensional; contrasted with model stitching achieving near-perfect IIA at rank 2.
Most independent dimension pair is aesthetic_response and boundary_awareness (rho=0.553); most correlated is prediction_error and conceptual_crystallization (rho=0.886)finding0.749
Characterizes internal structure of the six scoring dimensions
MAS reveals that numeric representations differ between GRUs trained on Multi-Object, Rounding, and Modulo tasksfinding0.740
Case study showing MAS can compare specific causal information types across models trained on different tasks.
MAS IIA is low for GRU hidden states vs Transformer hidden states on Multi-Object task, consistent with anti-Markovian transformer solutionfinding0.739
Validates MAS as a causal detector of representational differences invisible to correlative methods.
Experimental condition adjective embeddings show mean cosine similarity 0.657 (n=9,591 pairs), significantly higher than history (0.628, t=15.8, p=1.4×10⁻⁵⁵), conceptual (0.587, t=38.5, p<10⁻³⁰⁰), and zero-shot (0.603, t=35.1, p=4.3×10⁻²⁶²)finding0.738
Core result of Experiment 3: cross-model semantic convergence under self-referential processing
Representational Dissimilarity Matrix (RDM)method0.736
Pairwise dissimilarity matrix used in RSA computations; constructed using cosine distance between neural representations.