SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.12982 · NeurIPS · 2020

Disentangling by Subspace Diffusion

Deepmind London, Sébastien Racanière, David Pfau, Irina Higgins, Aleksandar Botev

arXiv · PDF · Open in the Atlas

Code that ran

6 of 9 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 6 failures of our instrument; 3 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2006.12982.json · badge: /api/ran/2006.12982.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
deepmind/deepmind-research marked official in Papers with Code
Papers with Code archive
6 ran · 3 no recorded run
FunctionStatusWhere it lives
cluster_subspaces Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("796ea6e47b7de60f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import scipy

# Set seed for reproducibility
np.random.seed(0)

# Create a small test case for omega
# omega should be shape (k, n, n) where k is number of matrices, n is dimension
# Looking at the function, it's used as:
# - omega[i] is indexed, so first dim is batch
# - omega[i] @ w @ w.T is computed, so it's a square matrix
# - omega.shape[:2] is used to get (k, n)

k = 3  # number of matrices
n = 4  # dimension of each square matrix

# Create symmetric positive semi-definite matrices for each batch
omega = np.zeros((k, n, n))
for i in range(k):
    # Create a random matrix and make it symmetric positive semi-definite
    A = np.random.randn(n, n)
    omega[i] = A @ A.T + np.eye(n)

ARGS = {
    "omega": omega,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
absl, numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 3, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ffdiag Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("b64fa2f6404a8f75")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import scipy.linalg

np.random.seed(0)

# The function expects:
# - data: shape (k, n, n) - a batch of k square matrices of size n×n
# - lr: learning rate (float)
# - tol: tolerance (float)
# - verbose: boolean flag
# - eig_init: boolean flag

k = 2  # batch size
n = 4  # matrix dimension

# Create random symmetric positive semi-definite matrices
# by generating random matrices and symmetrizing them
data_list = []
for _ in range(k):
    A = np.random.randn(n, n)
    # Make it symmetric and positive semi-definite
    sym = A @ A.T
    data_list.append(sym)

data = np.array(data_list)

ARGS = {
    "data": data,
    "lr": 1.0,
    "tol": 1e-10,
    "verbose": False,
    "eig_init": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
absl, numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (4, 4) — shape (4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_connection Ran
ran; our contract draft was wrong, not the code
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("366974755f039210")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from tqdm import tqdm

class NeighborGraph:
    def __init__(self):
        self.rows = {
            0: [1, 2],
            1: [0, 2],
            2: [0, 1]
        }

# Create tangent matrices: shape (3, 4, 4) means 3 tangent matrices of size 4x4
tangents = np.random.randn(3, 4, 4).astype(np.float32)

# Create neighbor graph
neighbor_graph = NeighborGraph()

ARGS = {
    "tangents": tangents,
    "neighbor_graph": neighbor_graph
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
absl, numpy, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_nearest_neighbors_graph Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("cfc00c11b004f03a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned lil_matrix. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import scipy.sparse

# Set random seed for reproducibility
np.random.seed(0)

# Create small test data
# data shape should be (num_points, num_features)
# Make it divisible by n (which we'll set to 2)
num_points = 4
num_features = 3
data = np.random.randn(num_points, num_features).astype(np.float32)

# Set k (number of nearest neighbors)
# k should be less than num_points - 1
k = 2

# Set n (number of neighbors to compute simultaneously)
# Must divide num_points evenly
n = 2

ARGS = {
    "data": data,
    "k": k,
    "n": n,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
absl, numpy, scipy, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type lil_matrix — a lil_matrix output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a lil_matrix output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sym_op Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("a10eba5982f8b00a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (3, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a small symmetric matrix as input
x = np.array([[1.0, 0.5], 
              [0.5, 2.0]], dtype=np.float64)

ARGS = {
    "x": x,
    "zero_trace": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 3) — shape (3, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
vec_to_sym Ran
ran; our contract draft was wrong, not the code
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("eedf089341d1d425")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Analyze the function:
# - x: a 1D array of floats (vector to be converted to symmetric matrix)
# - n: an integer (size of the output square matrix)
# - zero_trace: a boolean flag (whether to enforce zero trace)

# The function reconstructs a symmetric matrix from lower triangular elements
# For an n×n symmetric matrix, the lower triangular part (including diagonal)
# has n*(n+1)/2 elements
# When zero_trace=True, it appends one more element to x, so x should have
# n*(n+1)/2 - 1 elements before appending

n = 4
# Lower triangular indices count: n*(n+1)/2 = 4*5/2 = 10
# If zero_trace=False, x should have 10 elements
# If zero_trace=True, x should have 9 elements (one will be appended)

x_no_trace = np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0])

ARGS = {
    "x": x_no_trace,
    "n": n,
    "zero_trace": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (4, 4) — shape (4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
fit Not yet run
unverified (harvested, no recorded run)
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("ab9c21314cb23449")
How it ran

no invocation recorded.

make_laplacian Not yet run
unverified (harvested, no recorded run)
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("b406722434735411")
How it ran

no invocation recorded.

make_tangents Not yet run
unverified (harvested, no recorded run)
deepmind/deepmind-research/geomancer/geomancer.py
code served (permissive licence) · get_code("99327d6bf1cf0f06")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 9 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We present a novel nonparametric algorithm for symmetry-based disentangling of data manifolds, the Geometric Manifold Component Estimator (GEOMANCER). GEOMANCER provides a partial answer to the question posed by Higgins et al. (2018): is it possible to learn how to factorize a Lie group solely from observations of the orbit of an object it acts on? We show that fully unsupervised factorization of a data manifold is possible if the true metric of the manifold is known and each factor manifold has nontrivial holonomy -for example, rotation in 3D. Our algorithm works by estimating the subspaces that are invariant under random walk diffusion, giving an approximation to the de Rham decomposition from differential geometry. We demonstrate the efficacy of GEOMANCER on several complex synthetic manifolds. Our work reduces the question of whether unsupervised disentangling is possible to the question of whether unsupervised metric learning is possible, providing a unifying insight into the geometric nature of representation learning.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2006.12982")
get_code_for_paper("2006.12982")
have("2006.12982")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2006.12982.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2006.12982.svg)](https://syntology.ai/paper/2006.12982)

Connect an agent — have() is free.