Neural Representations of Location Composed of Spatially Periodic Bands (Krupic et al., 2012)

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

• Linear representation hypothesis: neural networks represent meaningful concepts as directions in their activation spaces.hypothesis0.812
  Foundation for interpreting features as linear directions.
• Smolensky (1986) proposes that viewing a neural representation under a basis that is not aligned with individual neurons can reveal the interpretable distributed structure of the neural representations.quote0.804
  Load-bearing theoretical claim providing the conceptual foundation for DAS.
• Emergence of grid-like representations by training recurrent neural networks to perform spatial localization (Cueva & Wei, 2018)concept0.794
  RNN model recapitulating grid cells; related work category 4.
• Distributed Neural Representationsconcept0.788
  Representations where individual neurons play multiple conceptual roles; patterns consisting of linear combinations of unit vectors.
• "Representation geometry is a window into the inner world of neural networks."quote0.787
  The paper's concluding summary statement asserting the deep interpretive significance of representation geometry.
• Superposition hypothesis: neural networks represent more features than dimensions using almost-orthogonal directions.hypothesis0.782
  Explanation for why dictionary learning can recover many more features than dimensions.
• Neural representation geometry causally shapes behavior; interventions respecting that geometry will yield natural trajectories.hypothesis0.779
  Central hypothesis tested via manifold steering experiments across language models and video world models.
• would place-like representations emerge in memory neurons for activation functions other than softmax?question0.778
  Open empirical question left for future work about robustness of place cell emergence.