Feature Neurons and Memory Neurons Architecture

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

• Memory Management Neurons/Headsconcept0.777
  MLP neurons and attention heads hypothesized to delete information from the residual stream by writing the negative of what they read
• Features are the fundamental unit of neural networks; they correspond to directions and can be rigorously studied and understood.claim0.775
  First of three speculative claims forming the foundation of the circuits research agenda
• Feature (neural network)concept0.768
  Scalar function of the input corresponding to a direction in the vector space of neuron activations; claimed to be the fundamental unit of neural networks
• would place-like representations emerge in memory neurons for activation functions other than softmax?question0.765
  Open empirical question left for future work about robustness of place cell emergence.
• What Architectural Features Distinguish Networks Capable Of Representingquestion0.757
• Geometry unifies diverse neural architectures in machine learning systems.claim0.752
• Neural Representations of Location Composed of Spatially Periodic Bands (Krupic et al., 2012)concept0.750
  Discovery of band cells; TEM-t also recapitulates these representations.
• There is a growing similarity in how datapoints are represented in different neural network models, spanning different architectures, training objectives, and data modalitiesclaim0.748
  Primary empirical claim of the paper