What Architectural Features Distinguish Networks Capable Of Representing

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

• What Architectural Properties Of Interaction Networks Enable Representationquestion0.845
• Feature Neurons and Memory Neurons Architectureframework0.757
  Biologically plausible two-pool architecture from Krotov & Hopfield (2020) splitting self-attention into feature and memory neuron populations; used to interpret TEM-t place cells.
• Features are connected by weights forming circuits, and these circuits can be rigorously studied and understood as meaningful algorithms.claim0.755
  Second of three speculative claims asserting that subgraphs of neural networks are tractable and meaningful objects of study
• Networks encode structured geometric concepts that reflect external reality.claim0.751
  Core claim of the paper: the right level of description for neural representations is geometric structure mirroring the world.
• Features are the fundamental unit of neural networks; they correspond to directions and can be rigorously studied and understood.claim0.748
  First of three speculative claims forming the foundation of the circuits research agenda
• How do representations differ or converge between architectures, tasks, and modalities?question0.748
  Broader research question MAS is positioned to address, citing multiple recent works.
• Lack of rigorous cross-model comparison demonstrating that specific named features (not just correlated ones) form across architecturesquestion0.747
  Explicitly identified research gap: anecdotal evidence exists but rigorous characterization is absent
• There is a growing similarity in how datapoints are represented in different neural network models, spanning different architectures, training objectives, and data modalitiesclaim0.747
  Primary empirical claim of the paper