Tolman-Eichenbaum Machine (TEM)

Neuroscience model of hippocampal formation that the paper shows is mathematically equivalent to a transformer with recurrent position encodings.

Neighborhood — ranked by edge-count

Thinkers (1)

thinker

James C.R. Whittington
introduces

Methods (1)

method

Hebbian Learning
implements
Principle that correlations strengthen connections; implements distributed learning in connectionist networks without centralized supervision.

Concepts (2)

concept

Outer Product Conjunctive Representations
implements
TEM's method of binding location g and sensory x representations by computing their outer product; corresponds to place cells.
Path Integration
implements
Neural mechanism for tracking location through accumulation of self-movement vectors; shown to play the role of position encodings in TEM.

Frameworks (2)

framework

TEM-Transformer (TEM-t)
extends
The transformer version directly analogous to TEM, introduced in this paper, offering dramatic performance improvements.
Hopfield Network
implements
A recurrent connectionist architecture that implements associative memory and distributed computation through symmetric weighted connections and Hebbian learning rules. The network converges to stable states through recurrent dynamics, enabling both memory retrieval and combinatorial problem-solving in a fully distributed manner.

Conceptual bridges

2-hop · via this framework's ideas

Where ideas in this framework connect to the rest of the corpus — the same concept, an analogy, or a restatement elsewhere.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

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

The Tolman-Eichenbaum Machine: Unifying Space and Relational Memory through Generalization in the Hippocampal Formation (Whittington et al., 2020)concept0.787
Prior paper by the same lead author introducing TEM; the neuroscience model whose relationship to transformers is the central topic.
TEM memory retrieval is mathematically equivalent to transformer self-attention without softmaxclaim0.737
Central theoretical claim: a single step of TEM attractor dynamics equals a dot-product attention, making TEM a special case of transformer.
TEM-t requires many fewer data samples than TEM to reach equivalent performance (sample efficiency improvement)finding0.717
Empirical performance comparison showing TEM-t is a more efficient learner than the original TEM.
Turing Machineconcept0.714
Theoretical construct establishing classical demarcation between machine and environment via input/output channels.
TEM-t instantiates hippocampal indexing theory by using memory neurons to bind cortical representations across brain regionsclaim0.697
Theoretical claim linking the TEM-t architecture to the Teyler-Rudy hippocampal indexing theory.
Experience Sampling Method (ESM)method0.693
Human psychology method for repeated in-situ self-report; methodological inspiration for the paper's approach
TEM-t requires less time per gradient step than TEMfinding0.687
Empirical computational efficiency result comparing TEM-t to the original TEM implementation.
TEM-t learns grid cells in hexagonal 6-connected worldsfinding0.686
Empirical extension showing grid cell learning generalises to non-4-connected spatial environments.