TEM-t requires less time per gradient step than TEM

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

• TEM-t requires many fewer data samples than TEM to reach equivalent performance (sample efficiency improvement)finding0.841
  Empirical performance comparison showing TEM-t is a more efficient learner than the original TEM.
• TEM memory retrieval is mathematically equivalent to transformer self-attention without softmaxclaim0.768
  Central theoretical claim: a single step of TEM attractor dynamics equals a dot-product attention, making TEM a special case of transformer.
• TEM-t with linear activations learns grid-cell-like position encoding representations in 2D spatial environmentsfinding0.745
  Empirical result showing TEM-t recapitulates entorhinal grid cell representations with linear post-transition activation.
• TEM-t instantiates hippocampal indexing theory by using memory neurons to bind cortical representations across brain regionsclaim0.741
  Theoretical claim linking the TEM-t architecture to the Teyler-Rudy hippocampal indexing theory.
• TEM-t learns grid cells in hexagonal 6-connected worldsfinding0.737
  Empirical extension showing grid cell learning generalises to non-4-connected spatial environments.
• TEM-t learns band-cell-like position encoding representations resembling Krupic et al. band cellsfinding0.717
  Empirical result showing TEM-t position encodings also recapitulate band cells, not just grid cells.
• The step-by-step approach works. The all-or-nothing approach does not work. This is the secret of biological evolution.quote0.704
  Crisp conclusion from the 30-coin thought experiment, linking adaptation in buildings to evolution.
• TEM-t memory neurons show spatially-tuned firing resembling hippocampal place cells in each environmentfinding0.703
  Empirical result demonstrating that the sparse softmax activation of memory neurons produces place-cell-like spatial tuning.