PHATE
Algorithms AIMech:PHATE DRAFT
Aliases: Potential of Heat Diffusion for Affinity-based Transition Embedding
A dimensionality-reduction algorithm that computes diffusion transition probabilities, measures Euclidean distances between their negative-log rows, and uses metric multidimensional scaling to embed those potential distances.
Identity Notes & Concept Scope
PHATE uses potential distances between transformed diffusion distributions followed by stress fitting; AIMech:DiffusionMaps uses time-scaled Markov eigenvectors. PHATE is not generic MDS, a generative diffusion model, a branch-detection procedure or the DEMaP evaluation criterion. The core algorithm is distinct from its landmark approximation and multislice extensions. All 442 pinned AIO labels/scoped synonyms and 114 preceding native identities were compared, together with hidden/ignored-inclusive searches of both checkouts; source citations and pending frontier mentions were distinguished from existing concept identities.
Claims & Grounded Evidence
1 Claim
Mechanism
#method-scope
Build symmetric local affinities with an adaptive bandwidth and an alpha-decaying kernel, row-normalize them to a Markov operator P, and take a selected diffusion power P^t. For positive transition probabilities, form potential vectors −log(P^t_i) and their Euclidean pairwise distances. Initialize with classical MDS and optimize metric-MDS stress in the requested output dimension. The paper proposes an entropy-knee heuristic for choosing diffusion time and a separate landmark approximation for scaling.
⚠️ Curated Limitations & Negative Results
- Neighborhood affinities, bandwidth, decay exponent and diffusion time determine the geometry; excessive diffusion can remove signal as well as noise.
- Negative logarithms require positive entries. Disconnected or numerically thresholded transitions need explicit zero-handling; the paper’s formula alone does not specify a universal remedy.
- A two- or three-dimensional stress fit can distort potential distances and does not guarantee global optimization, preservation of every manifold property or biological trajectories.
- The entropy discussion assumes a nonnegative spectrum for its probability normalization. This record does not certify that assumption for every adaptive kernel or treat the knee heuristic as an optimal-time theorem.
- The inspected PDF is marked corrected; the correction notice and access limitations are tracked in the discovery report.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:DimensionalityReduction
Dimensionality Reduction
|
related_match | A specific algorithm for constructing lower-dimensional representations, associated with the pinned dimensionality-reduction task. The target inherits UnsupervisedLearning under MachineLearningTask; related_match avoids treating the algorithm as identical to or a subclass of that task. Shared embedding components do not make this concept identical to PCA, MDS or t-SNE. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Kevin R. Moon, David van Dijk, Zheng Wang, Scott Gigante, Daniel B. Burkhardt, William S. Chen, Kristina Yim, Antonia van den Elzen, Matthew J. Hirn, Ronald R. Coifman, Natalia B. Ivanova, Guy Wolf, Smita Krishnaswamy
|
primary paper
full_text
|
ID: doi:10.1038/s41587-019-0336-3
sha256:3f56978e04a3…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T05:21:17Z
Updated: 2026-10-11T05:22:06Z
Source File:
data/curated/PHATE.yaml