Hierarchical temporal memory¶

Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta.

Version: v1 · 2026-09-28 · History
Domain-specific #: 9860
Domain group: Applied Sciences & Engineering
Origin domain: Computer Science & Software Engineering
Subdomains: Machine Learning, Biologically Inspired AI → Computer Science & Software Engineering

Core Idea¶

Hierarchical temporal memory is treated here as the recurring computational neuroscience identity summarized by this source-grounded definition: Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta.

Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta. Originally described in the 2004 book On Intelligence by Jeff Hawkins with Sandra Blakeslee, HTM is primarily used today for anomaly detection in streaming data. The technology is based on neuroscience and the physiology and interaction of pyramidal neurons in the neocortex of the mammalian (in particular, human) brain.

At the core of HTM are learning algorithms that can store, learn, infer, and recall high-order sequences. Unlike most other machine learning methods, HTM constantly learns (in an unsupervised process) time-based patterns in unlabeled data. HTM is robust to noise, and has high capacity (it can learn multiple patterns simultaneously).

For Hierarchical temporal memory, the abstraction is narrower than the article's general subject matter: a positive case must preserve Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta. Retaining only the name, a familiar example, or a downstream effect is insufficient. The specialist roles and tests remain anchored in computational neuroscience, which is why this identity is domain-specific rather than prime.

Structural Signature¶

Sig role-phrases:

Defining carrier — Since resolution in space and time is lost in each node as described above, beliefs formed by higher-level nodes represent an even larger range of space and time.
Constitutive relation — The learning process consists of two stages.
Operating condition — Higher hierarchy levels can reuse patterns learned at the lower levels by combining them to memorize more complex patterns.
Recognition evidence — Each HTM region learns by identifying and memorizing spatial patterns—combinations of input bits that often occur at the same time.
Admissible variation — (In a simplified implementation, node's belief consists of only one winning group).
Characteristic consequence — This is meant to reflect the organisation of the physical world as it is perceived by the human brain.
Failure boundary — Each HTM layer (not to be confused with an HTM level of an HTM hierarchy, as described above) consists of a number of highly interconnected minicolumns.

What It Is Not¶

Not the whole field of computational neuroscience. The node requires the specific identity stated by Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta.
Not an over-broad reading. The new findings do not necessarily invalidate the previous parts of the model, so ideas from one generation are not necessarily excluded in its successive one.
Not an over-broad reading. Temporal pooling is not yet well understood, and its meaning has changed over time (as the HTM algorithms evolved).
Not an over-broad reading. "Unexpected" patterns to the node do not have a dominant probability of belonging to any one temporal group but have nearly equal probabilities of belonging to several of the groups.
Not automatically Associative Memory. Retrieval proximity does not establish equivalence; the two identities must be compared by carrier, operation, and failure boundary.

Scope of Application¶

Hierarchical temporal memory applies literally inside computational neuroscience wherever the source-defined carrier and relation can be established. Its documented habitats include:

Structure and algorithms. A typical HTM network is a tree-shaped hierarchy of levels (not to be confused with the "layers" of the neocortex, as described below).
Inference. More details about the functioning of Zeta 1 HTM can be found in Numenta's old documentation.
Inference. Each HTM layer (not to be confused with an HTM level of an HTM hierarchy, as described above) consists of a number of highly interconnected minicolumns.
Spatial pooling. The amount of memory used by each layer can be increased to learn more complex spatial patterns or decreased to learn simpler patterns.
Inference and online learning. This allows each HTM layer to be constantly predicting the likely continuation of the recognized sequences.
Comparing HTM and neocortex. HTM attempts to implement the functionality that is characteristic of a hierarchically related group of cortical regions in the neocortex.

Outside computational neuroscience, the name should be retained only when these same operational conditions survive; otherwise the comparison belongs to the broader parent Theory or should be marked as analogy.

Clarity¶

A clear use of Hierarchical temporal memory names the carrier, the operative relation, and the conditions under which the source treats the identity as present. The minimal definition is Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta. The strongest recognition evidence in the frozen account is: Each HTM region learns by identifying and memorizing spatial patterns—combinations of input bits that often occur at the same time. A report should distinguish that evidence from a proxy, consequence, or common implementation. It should also state the qualification The new findings do not necessarily invalidate the previous parts of the model, so ideas from one generation are not necessarily excluded in its successive one. so that a reader can reproduce the classification rather than infer it from topical resemblance.

Manages Complexity¶

Hierarchical temporal memory compresses multiple computational neuroscience details into a stable diagnostic relation. The source shows both the central mechanism—the learning process consists of two stages.—and the practical consequence—this is meant to reflect the organisation of the physical world as it is perceived by the human brain. This compression makes cases comparable while leaving parameters, conventions, exceptions, and evidential quality explicit. It is lossy by design: local history and implementation details may be omitted only when they do not alter the defining relation.

Abstract Reasoning¶

Type the carrier. Identify the computational neuroscience entities to which the claim applies.
State the relation. Use the source-grounded identity: Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta.
Check operation and conditions. Higher hierarchy levels can reuse patterns learned at the lower levels by combining them to memorize more complex patterns.
Demand recognition evidence. Each HTM region learns by identifying and memorizing spatial patterns—combinations of input bits that often occur at the same time.
Test variation. Change an implementation or setting while preserving (In a simplified implementation, node's belief consists of only one winning group).
Run the collapse test. Remove the defining operation; if the label still seems equally apt, only a topic or correlate was retained.
Reduce cautiously. When the specialist conditions cannot be carried, route the residual comparison to Theory.

Knowledge Transfer¶

Within the home domain. Knowledge about Hierarchical temporal memory transfers literally when a new case preserves the same carrier type, relation, and recognition test. A typical HTM network is a tree-shaped hierarchy of levels (not to be confused with the "layers" of the neocortex, as described below). More details about the functioning of Zeta 1 HTM can be found in Numenta's old documentation.

Beyond the home domain. No canonical parent is asserted for Hierarchical temporal memory. An outside case receives the specialist name only when the same typed roles and rejection conditions can be filled literally; otherwise the comparison remains an analogy pending later graph densification.

Examples¶

Canonical¶

In learning and inference modes, sensory data (e.g. data from the eyes) comes into bottom-level regions. This case is canonical because it supplies a concrete carrier and lets the defining relation be checked rather than merely named.

Mapped back: carrier → the entities in the documented case; operation → Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta; recognition evidence → Each HTM region learns by identifying and memorizing spatial patterns—combinations of input bits that often occur at the same time

Applied / In Practice¶

Larger concepts (e.g. causes, actions, and objects) are perceived to change more slowly and consist of smaller concepts that change more quickly. The applied case shows how the identity is used under a second setting or qualification while keeping the same operative relation.

Mapped back: changed setting → Inference; invariant → Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta; boundary → the case exits the class when the new findings do not necessarily invalidate the previous parts of the model, so ideas from one generation are not necessarily excluded in its successive one

Structural Tensions¶

T1 — Stable identity versus admissible variation. The new findings do not necessarily invalidate the previous parts of the model, so ideas from one generation are not necessarily excluded in its successive one. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Which changes preserve the defining relation, and which replace it?

T2 — Recognition versus proxy. Temporal pooling is not yet well understood, and its meaning has changed over time (as the HTM algorithms evolved). The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Does the cited evidence establish the identity or only a correlated sign?

T3 — Definition versus implementation. "Unexpected" patterns to the node do not have a dominant probability of belonging to any one temporal group but have nearly equal probabilities of belonging to several of the groups. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Is the observed implementation constitutive, optional, or merely common?

T4 — Scope versus overextension. If sequences of patterns are similar to the training sequences, then the assigned probabilities to the groups will not change as often as patterns are received. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Can every claimed application fill the same typed roles without metaphor?

T5 — Transfer versus domain accent. Since resolution in space and time is lost in each node as described above, beliefs formed by higher-level nodes represent an even larger range of space and time. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Does the receiving case instantiate Hierarchical temporal memory literally, co-instantiate Theory, or only resemble it?

T6 — Autonomy versus reduction. The learning process consists of two stages. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: What does Hierarchical temporal memory distinguish that the broader parent Theory leaves together?

Structural–Framed Character¶

Hierarchical temporal memory is mixed or framed-leaning. Its structural side is the repeatable organization summarized by Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta. Its framed side is the computational neuroscience vocabulary that fixes the carrier, evidence, exceptions, and admissible transformations.

Evaluative weight: the identity can be stated descriptively even when applications carry practical stakes. Human-practice dependence: the source-grounded carrier determines whether the relation exists independently or is constituted by a practice. Institutional origin: disciplinary conventions stabilize the name and test. Vocabulary portability: Higher hierarchy levels can reuse patterns learned at the lower levels by combining them to memorize more complex patterns. Import versus recognition: literal transfer requires the same mechanism; shape alone is analogy.

Its portable skeleton is Theory. Its character: a recurring specialist identity whose thin organization can be abstracted, while its operational meaning remains domain-bound.

Structural Core vs. Domain Accent¶

What is skeletal. Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta. The stable skeleton is the typed relation expressed in that definition and the entry's recognition and collapse tests. The source identifies these operative conditions: Since resolution in space and time is lost in each node as described above, beliefs formed by higher-level nodes represent an even larger range of space and time. The learning process consists of two stages. It further constrains recognition and variation through: Higher hierarchy levels can reuse patterns learned at the lower levels by combining them to memorize more complex patterns. Each HTM region learns by identifying and memorizing spatial patterns—combinations of input bits that often occur at the same time.

What is domain-bound. computational neuroscience supplies the operative entities, technical vocabulary, warrants, and exceptions that make Hierarchical temporal memory literal. Its documented scope includes the condition that A typical HTM network is a tree-shaped hierarchy of levels (not to be confused with the "layers" of the neocortex, as described below). Another bounded application condition is that More details about the functioning of Zeta 1 HTM can be found in Numenta's old documentation. These are not decorative examples; they determine which carrier and evidence can fill the abstraction's roles.

Why no parent is asserted. Removing those specialist details does not currently yield one live catalog node that is a necessary genus for every instance. The entry is therefore approved as unparented rather than attached by topical resemblance. Its collapse evidence remains specific—(In a simplified implementation, node's belief consists of only one winning group).—and future graph densification may discover a defensible relation only if it preserves that boundary.

This entry under conditions is a kind of Machine-Learning Model.

Approved unparented node. No current live node supplies a defensible necessary genus or structural prerequisite for Hierarchical temporal memory. The reviewed identity is: Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta. The accelerated suggestion was declined because topical or lexical similarity does not establish hierarchy; the node is admitted without a parent pending later graph densification.
Related reasoning operations. Evidence, representation, comparison, classification, transformation, or evaluation may participate in particular cases, but participation does not make any one of them a necessary parent of every instance.

Relationships to Other Abstractions¶

Current abstraction Hierarchical temporal memory Domain-specific

Parents (1) — more general patterns this builds on

Hierarchical temporal memory is a kind of, conditional Machine-Learning Model Domain-specific

It supplies learned models, though the term can also name a technology or algorithm family.

Condition / exception It supplies learned models, though the term can also name a technology or algorithm family.

Hierarchy path (1) — routes to 1 parentless root

Hierarchical temporal memory → Machine-Learning Model

Neighborhood in Abstraction Space¶

Hierarchical temporal memory sits in a moderately populated region (60^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Neural & Cognitive Representation Models (11 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Theory. The parent omits the specialist differentia. Tell: Can the case establish Hierarchical temporal memory (HTM) is a biologically constrained machine intelligence technology developed by Numenta?
Associative Memory. Content-addressable storage where a cue retrieves linked content. Tell: Which entry's carrier, operation, and failure condition are satisfied?
Attribute Hierarchy Method. Diagnose learners' mastery by arranging cognitive attributes in a prerequisite hierarchy, deriving feasible response patterns, and matching observed item responses to those patterns. Tell: Which entry's carrier, operation, and failure condition are satisfied?
Model of hierarchical complexity. A content-neutral developmental scoring framework that assigns tasks an order based on how nonarbitrarily they coordinate actions from the immediately lower order into a new hierarchical organization. Tell: Which entry's carrier, operation, and failure condition are satisfied?
A measurement, proxy, or consequence. Those may provide evidence without being the identity. Tell: Would Hierarchical temporal memory remain present if the detector or downstream effect changed?
A metaphorical analogue. A similar shape outside computational neuroscience lacks the specialist mechanism. Tell: Do the native roles transfer literally, or only the parent Theory?

References¶

Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Hierarchical_temporal_memory (revision 1358202218).
Preserved source candidate: https://discourse.numenta.org/t/preliminary-details-about-new-theory-work-on-sensory-motor-inference/697
Preserved source candidate: https://web.archive.org/web/20090527174304/http://numenta.com/for-developers/education/general-overview-htm.php
Preserved source candidate: http://numenta.com/for-developers/education/general-overview-htm.php
Preserved source candidate: http://grokstream.com/product/
Preserved source candidate: https://web.archive.org/web/20190426092617/https://www.grokstream.com/product/
Preserved source candidate: https://ai.stanford.edu/~joni/papers/LasersonXRDS2011.pdf
Preserved source candidate: https://www.eurekalert.org/pub_releases/2019-01/kta-np011119.php
Preserved source candidate: https://web.archive.org/web/20201108011941/https://www.eurekalert.org/pub_releases/2019-01/kta-np011119.php

The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.