SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.11267 · NeurIPS · 2020

Fast Matrix Square Roots with Applications to Gaussian Processes and Bayesian Optimization

Geoff Pleiss, Anil Damle, Martin Jankowiak, David Eriksson, Jacob Gardner

arXiv · PDF · Open in the Atlas

Code that ran

12 of 15 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 12 have no contract verdict; 0 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/2006.11267.json · badge: /api/ran/2006.11267.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
gpleiss/ciq_experiments canonical (classifier label)
classifier basis: census
12 ran · 3 no recorded run
FunctionStatusWhere it lives
Phi Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/crps.py
code served (permissive licence) · get_code("af00b91406dbc454")
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 Tensor, ndim 2, dtype family float, shape (2, 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 torch
import torch.nn as nn

torch.manual_seed(0)

# x: input tensor (will be used in (x - mu) / sigma calculation)
x = torch.randn(2, 4)

# mu: mean parameter (will be used in (x - mu) / sigma calculation)
mu = torch.randn(2, 4)

# sigma: standard deviation parameter (will be used in (x - mu) / sigma calculation)
sigma = torch.randn(2, 4) + 1.0  # Add 1.0 to ensure positive values

ARGS = {
    "x": x,
    "mu": mu,
    "sigma": sigma,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
crps Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/crps.py
code served (permissive licence) · get_code("58983a88cbdb1cad")
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 float, ndim 0, dtype family float, shape () 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 torch
import math

torch.manual_seed(0)

# Create small tensors for the function parameters
# x: observation value(s) - shape (2,) for batch of 2
x = torch.randn(2)

# mu: mean parameter - shape (2,) for batch of 2
mu = torch.randn(2)

# sigma: standard deviation parameter - shape (2,) for batch of 2
sigma = torch.abs(torch.randn(2)) + 0.1  # ensure positive

ARGS = {
    "x": x,
    "mu": mu,
    "sigma": sigma,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () 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
downsample_img Ran
ran on a synthesized input
gpleiss/ciq_experiments/super_resolution/sr.py
code served (permissive licence) · get_code("99a0fb32c2e3d86f")
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, 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 torch
import numpy as np

torch.manual_seed(0)

# Create a small test image as a 2D numpy array
# Using shape 8x6 so downsampling by factor 2 gives clean results
img = np.arange(48, dtype=np.float32).reshape(8, 6)

ARGS = {
    "img": img,
    "binary": True,
    "factor": 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — 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, 3) — shape (4, 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
from_unit_cube Ran
ran on a synthesized input
gpleiss/ciq_experiments/bayesopt/ciq_bo/utils.py
code served (permissive licence) · get_code("4882830652616cef")
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 Tensor, ndim 2, dtype family float, shape (2, 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 torch

torch.manual_seed(0)

# Create small tensors for the function
# x: shape (batch=2, dim=4), values in [0, 1]
x = torch.rand(2, 4)

# lb: shape (4,), lower bounds
lb = torch.tensor([-1.0, -2.0, -3.0, -4.0])

# ub: shape (4,), upper bounds (must be > lb)
ub = torch.tensor([1.0, 2.0, 3.0, 4.0])

ARGS = {
    "x": x,
    "lb": lb,
    "ub": ub,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
load_airline_data Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/load_uci_data.py
code served (permissive licence) · get_code("19ac6cfc2910162d")
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 tuple. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch
import numpy as np
import pandas as pd
from io import BytesIO
import pickle

torch.manual_seed(0)

# Create a minimal mock airline dataset
mock_data = {
    'Month': np.random.randint(1, 5, 100),
    'DayofMonth': np.random.randint(1, 29, 100),
    'DayOfWeek': np.random.randint(1, 8, 100),
    'plane_age': np.random.randint(1, 30, 100),
    'AirTime': np.random.randint(30, 500, 100),
    'Distance': np.random.randint(100, 2000, 100),
    'ArrTime': np.random.randint(600, 2300, 100),
    'DepTime': np.random.randint(600, 2300, 100),
    'ArrDelay': np.random.randint(-50, 200, 100)
}

df = pd.DataFrame(mock_data)

# Create a temporary directory-like object that stores the pickle data
class MockDataDir:
    def __init__(self, data):
        self.data = data
    
    def __add__(self, filename):
        return self

class MockPath:
    def __init__(self, data):
        self.data = data
    
    def __fspath__(self):
        # Return a string path, but we'll use read_pickle override
        return "mock"

# Create a minimal mock data_dir - just use a temporary pickle file approach
import tempfile
import os

temp_dir = tempfile.mkdtemp()
pickle_path = os.path.join(temp_dir, 'airline.pickle')
df.to_pickle(pickle_path)

# Update pandas read_pickle to handle our mock
data_dir = temp_dir + '/'

ARGS = {
    'data_dir': data_dir,
    'seed': 42
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
load_covtype_data Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/load_uci_data.py
code served (permissive licence) · get_code("d36530514ec97b1a")
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 tuple. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch
import pandas as pd
import numpy as np
from tempfile import TemporaryDirectory
import os

torch.manual_seed(0)

# Create a temporary directory and mock covtype.csv data
temp_dir = TemporaryDirectory()
data_dir = temp_dir.name + '/'

# Create mock covtype.csv with minimal data
# 54 features + 1 target column (to match covtype structure)
# We'll create 100 samples to have enough for train/valid/test splits
np.random.seed(0)
n_samples = 100
n_features = 54
mock_data = np.random.randn(n_samples, n_features).astype(np.float32)
# Add target column (values 1-7, we'll create some 2s and 3s)
target = np.random.randint(1, 8, n_samples)
target[:20] = 2  # Ensure some 2s
target[20:40] = 3  # Ensure some 3s
mock_data = np.column_stack([mock_data, target])

df = pd.DataFrame(mock_data)
df.to_csv(data_dir + 'covtype.csv', index=False)

ARGS = {
    'data_dir': data_dir,
    'seed': 42
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
load_robopush_data Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/load_uci_data.py
code served (permissive licence) · get_code("8a1f1a7cf2984a2e")
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 tuple. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch
from pathlib import Path
import tempfile
import os

torch.manual_seed(0)

# Create a temporary directory and mock robopush data
temp_dir = tempfile.mkdtemp()
data_dir = temp_dir + '/'

# Create mock robopush.pt file
# The function expects: X_rand, X_turbo, fX_rand, fX_turbo
# X shapes: (batch, features)
# fX shapes: (batch,)
X_rand = torch.randn(4, 5)  # 4 samples, 5 features
X_turbo = torch.randn(3, 5)  # 3 samples, 5 features
fX_rand = torch.randn(4)  # 4 samples
fX_turbo = torch.randn(3)  # 3 samples

torch.save((X_rand, X_turbo, fX_rand, fX_turbo), data_dir + 'robopush.pt')

ARGS = {
    'data_dir': data_dir,
    'seed': 42,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
output_class Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/util.py
code served (permissive licence) · get_code("085f08870c397e8d")
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 type. 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 torch
from collections import namedtuple

torch.manual_seed(0)

def output_class(fields):
    class Output(namedtuple('Output', fields)):
        def __repr__(self):
            res = 'Outputs:\n'
            fieldstrs = []
            for key in self._fields:
                fieldstrs.append('  - %s: %s' % (key, repr(getattr(self, key).size())))
            res = res + '\n'.join(fieldstrs)
            return res
    return Output

# Create a simple Output class with sample fields
fields = ['logits', 'loss', 'embeddings']

# Create instance to test
Output = output_class(fields)
instance = Output(
    logits=torch.randn(2, 4),
    loss=torch.tensor(0.5),
    embeddings=torch.randn(2, 3)
)

# Test the __repr__ method
repr_output = repr(instance)

ARGS = {
    'fields': ['logits', 'loss', 'embeddings']
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type type — a type output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a type 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
phi Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/crps.py
code served (permissive licence) · get_code("400cde6d40f64792")
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 Tensor, ndim 2, dtype family float, shape (2, 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 torch
import torch.nn as nn

torch.manual_seed(0)

# The function phi takes three parameters: x, mu, sigma
# All are used in arithmetic operations: (x - mu) / sigma
# They should be tensors that support broadcasting and arithmetic

x = torch.randn(2, 4)
mu = torch.randn(2, 4)
sigma = torch.abs(torch.randn(2, 4)) + 0.1  # ensure positive values for sigma

ARGS = {
    "x": x,
    "mu": mu,
    "sigma": sigma
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
result_class Ran
ran on a synthesized input
gpleiss/ciq_experiments/svgp/util.py
code served (permissive licence) · get_code("8c6692f79520aea4")
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 type. 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 torch
from collections import namedtuple

torch.manual_seed(0)

# Analyze the function: result_class takes 'fields' parameter
# which should be a sequence of field names (strings)
# The function creates a namedtuple-based Result class with those fields
# The Result class has methods: items(), to_str(), __repr__()
# These methods iterate over _fields and access attributes via getattr()

# Valid test case: fields should be a list/tuple of strings
fields = ['accuracy', 'loss', 'epoch']

ARGS = {
    'fields': fields
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type type — a type output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a type 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
standardize Ran
ran on a synthesized input
gpleiss/ciq_experiments/bayesopt/ciq_bo/utils.py
code served (permissive licence) · get_code("675f960778d15b16")
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 Tensor, ndim 1, dtype family float, shape (2,) 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 torch

torch.manual_seed(0)

# Create a 1-dimensional tensor with small size
y = torch.randn(2)

ARGS = {
    "y": y
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) 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
to_unit_cube Ran
ran on a synthesized input
gpleiss/ciq_experiments/bayesopt/ciq_bo/utils.py
code served (permissive licence) · get_code("d4f3607c98798b5f")
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 Tensor, ndim 2, dtype family float, shape (2, 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 torch

torch.manual_seed(0)

# Construct arguments for to_unit_cube
# x: 2D tensor of shape (batch_size, dim)
# lb: 1D tensor of lower bounds
# ub: 1D tensor of upper bounds

batch_size = 2
dim = 4

x = torch.randn(batch_size, dim) * 10  # Random values in a wider range
lb = torch.tensor([-5.0, -3.0, -2.0, -1.0])  # Lower bounds, shape (dim,)
ub = torch.tensor([5.0, 3.0, 2.0, 1.0])  # Upper bounds, shape (dim,)

ARGS = {
    "x": x,
    "lb": lb,
    "ub": ub,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
get_logger Not yet run
unverified (harvested, no recorded run)
gpleiss/ciq_experiments/svgp/logger.py
code served (permissive licence) · get_code("fb5b4b75bd55fd1c")
How it ran

no invocation recorded.

inv_matmul Not yet run
unverified (harvested, no recorded run)
gpleiss/ciq_experiments/super_resolution/sr.py
code served (permissive licence) · get_code("d0f69205712e19ce")
How it ran

no invocation recorded.

mvn_sample Not yet run
unverified (harvested, no recorded run)
gpleiss/ciq_experiments/super_resolution/sr.py
code served (permissive licence) · get_code("64c69a6908432123")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 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 15 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

Matrix square roots and their inverses arise frequently in machine learning, e.g., when sampling from high-dimensional Gaussians N (0, K) or "whitening" a vector b against covariance matrix K. While existing methods typically require O(N 3 ) computation, we introduce a highly-efficient quadratic-time algorithm for computing K 1/2 b, K -1/2 b, and their derivatives through matrix-vector multiplication (MVMs). Our method combines Krylov subspace methods with a rational approximation and typically achieves 4 decimal places of accuracy with fewer than 100 MVMs. Moreover, the backward pass requires little additional computation. We demonstrate our method's applicability on matrices as large as 50,000 × 50,000well beyond traditional methods-with little approximation error. Applying this increased scalability to variational Gaussian processes, Bayesian optimization, and Gibbs sampling results in more powerful models with higher accuracy. In particular, we perform variational GP inference with up to 10,000 inducing points and perform Gibbs sampling on a 25,000-dimensional problem. * This work was conducted while David Eriksson was at Uber AI. 34th Conference on Neural Information Processing Systems (NeurIPS 2020),

For agents

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

get_harvested_code_for_paper("2006.11267")
get_code_for_paper("2006.11267")
have("2006.11267")

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

curl https://syntology.ai/api/ran/2006.11267.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.11267.svg)](https://syntology.ai/paper/2006.11267)

Connect an agent — have() is free.