Tensor Product Representation Probes Reveal Shared Structure Across Linear Directions

• URL: http://arxiv.org/abs/2605.09967v1
• Date: Mon, 11 May 2026 04:18:43 GMT
• Title: Tensor Product Representation Probes Reveal Shared Structure Across Linear Directions
• Abstract summary: We study a model with known linear representations, but trained in a highly structured domain.<n>We train TPR probes to recover shared structure amongst the linear probes.<n>Our findings suggest that directional representations may be projections of more structured underlying representations.
• Score: 48.32734639771152
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: While researchers are finding concepts represented as linear directions in language models, a bag of linear directions fails to capture relational structure. To better understand this dichotomy, we study a model with known linear representations, but trained in a highly structured domain -- the board game Othello. While the model's internal board-state representation is linearly decodable, we find additional structure in the form of tensor product representations (TPRs). We train TPR probes to recover shared structure amongst the linear probes, yielding a factorization into square-embeddings, color-embeddings, and a binding matrix that composes them to construct the model's board-state representation. We find geometric signatures within the weights of our TPR probe that align with the structure of the board, but perhaps more importantly, that the linear probes can be recovered directly from the parameters of our TPR probe. Our findings suggest that directional representations may be projections of more structured underlying representations.

Related papers

• Are Common Substructures Transferable? Riemannian Graph Foundation Model with Neural Vector Bundles [53.65187206795911]
  We develop a graph intrinsic geometry learning framework called Neural Vector Bundle.<n>Building on this, we design GAUGE, a pretrainable neural architecture that constructs the vector bundle and flattens geometrically compatible local coordinates.<n>We empirically validate its superior expressiveness in challenging tasks including zero-shot link prediction and graph isomorphism.
  arXiv  Detail & Related papers  (2026-06-02T07:35:42Z)
• Sheaf Neural Networks on SPD Manifolds: Second-Order Geometric Representation Learning [9.973930984631547]
  We develop the first sheaf neural network operating on the SPD manifold.<n>We show that SPD-valued sheaves are strictly more expressive than Euclidean sheaves.<n>Our dual-stream architecture achieves SOTA on 6/7 MoleculeNet benchmarks, with the sheaf framework providing consistent depth robustness.
  arXiv  Detail & Related papers  (2026-04-22T08:09:04Z)
• The Lattice Representation Hypothesis of Large Language Models [18.00499182102749]
  We show that linear attribute directions with separating thresholds induce a concept lattice via half-space intersections.<n>This geometry enables symbolic reasoning through geometric meet (intersection) and join (union) operations.
  arXiv  Detail & Related papers  (2026-03-01T18:42:59Z)
• Structural Disentanglement in Bilinear MLPs via Architectural Inductive Bias [0.0]
  We argue that failures arise from how models structure their internal representations during training.<n>We show analytically that bilinear parameterizations possess a non-mixing' property under gradient flow conditions.<n>Unlike pointwise nonlinear networks, multiplicative architectures are able to recover true operators aligned with the underlying algebraic structure.
  arXiv  Detail & Related papers  (2026-02-05T13:14:01Z)
• Cross-Model Semantics in Representation Learning [1.2064681974642195]
  We show that structural regularities induce representational geometry that is more stable under architectural variation.<n>This suggests that certain forms of inductive bias not only support generalization within a model, but also improve the interoperability of learned features across models.
  arXiv  Detail & Related papers  (2025-08-05T16:57:24Z)
• Linear Representation Transferability Hypothesis: Leveraging Small Models to Steer Large Models [6.390475802910619]
  We show that representations learned across models trained on the same data can be expressed as linear combinations of a emphuniversal set of basis features.<n>These basis features underlie the learning task itself and remain consistent across models, regardless of scale.
  arXiv  Detail & Related papers  (2025-05-31T17:45:18Z)
• Compositional Structures in Neural Embedding and Interaction Decompositions [101.40245125955306]
  We describe a basic correspondence between linear algebraic structures within vector embeddings in artificial neural networks. We introduce a characterization of compositional structures in terms of "interaction decompositions" We establish necessary and sufficient conditions for the presence of such structures within the representations of a model.
  arXiv  Detail & Related papers  (2024-07-12T02:39:50Z)
• On the Origins of Linear Representations in Large Language Models [51.88404605700344]
  We introduce a simple latent variable model to formalize the concept dynamics of the next token prediction. Experiments show that linear representations emerge when learning from data matching the latent variable model. We additionally confirm some predictions of the theory using the LLaMA-2 large language model.
  arXiv  Detail & Related papers  (2024-03-06T17:17:36Z)
• Linear Spaces of Meanings: Compositional Structures in Vision-Language Models [110.00434385712786]
  We investigate compositional structures in data embeddings from pre-trained vision-language models (VLMs) We first present a framework for understanding compositional structures from a geometric perspective. We then explain what these structures entail probabilistically in the case of VLM embeddings, providing intuitions for why they arise in practice.
  arXiv  Detail & Related papers  (2023-02-28T08:11:56Z)
• Learning Bijective Feature Maps for Linear ICA [73.85904548374575]
  We show that existing probabilistic deep generative models (DGMs) which are tailor-made for image data, underperform on non-linear ICA tasks. To address this, we propose a DGM which combines bijective feature maps with a linear ICA model to learn interpretable latent structures for high-dimensional data. We create models that converge quickly, are easy to train, and achieve better unsupervised latent factor discovery than flow-based models, linear ICA, and Variational Autoencoders on images.
  arXiv  Detail & Related papers  (2020-02-18T17:58:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.