SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.07709 · 2020

Auditing Differentially Private Machine Learning: How Private is Private SGD?

arXiv · PDF · Open in the Atlas

Code that ran

6 of 7 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 6 have no contract verdict; 0 failures of our instrument; 1 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/2006.07709.json · badge: /api/ran/2006.07709.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
jagielski/auditing-dpsgd canonical (classifier label)
classifier basis: census
6 ran · 1 no recorded run
FunctionStatusWhere it lives
bkd_find_thresh Ran
ran on a synthesized input
jagielski/auditing-dpsgd/code/bkd_parser.py
code served (permissive licence) · get_code("d314d09dc17a785f")
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 float64, ndim 0, 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
from scipy import optimize, stats

# Construct small numpy arrays for the function
np.random.seed(0)

# nobkd_li: array of scores for non-background events
nobkd_li = np.array([0.1, 0.3, 0.5, 0.7, 0.9])

# bkd_li: array of scores for background events
bkd_li = np.array([0.2, 0.4, 0.6, 0.8])

# use_dkw: boolean flag to use Dvoretzky-Kiefer-Wolfowitz inequality
use_dkw = False

ARGS = {
    "nobkd_li": nobkd_li,
    "bkd_li": bkd_li,
    "use_dkw": use_dkw,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
bkd_get_eps Ran
ran on a synthesized input
jagielski/auditing-dpsgd/code/bkd_parser.py
code served (permissive licence) · get_code("15892840c1537e01")
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, ndim 1, dtype family float; no output shape is stated. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:55:27Z. 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
from scipy import optimize, stats

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

# Create sample data for the function parameters
cfg = [10]  # n_repeat = 10

# Create two arrays of random values between 0 and 1
# These represent some kind of scores or probabilities
nobkd_li = np.random.rand(100)  # 100 samples for no-background
bkd_li = np.random.rand(120)    # 120 samples for background

# Set a threshold value
thresh = 0.5

# use_dkw flag
use_dkw = False

ARGS = {
    "cfg": cfg,
    "nobkd_li": nobkd_li,
    "bkd_li": bkd_li,
    "thresh": thresh,
    "use_dkw": use_dkw
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple 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 syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); 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
build_model Ran
ran on a synthesized input
jagielski/auditing-dpsgd/code/audit.py
code served (permissive licence) · get_code("86f540150b92a2dc")
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 Sequential. 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.

from absl import flags
import tensorflow as tf
import numpy as np

FLAGS = flags.FLAGS

# Define the FLAGS that the function reads
flags.DEFINE_string('dataset', 'fmnist', 'Dataset name')
flags.DEFINE_string('model', 'lr', 'Model type')

# Mark flags as parsed to avoid issues
FLAGS.mark_as_parsed()

# Create small dummy data
x = tf.constant(np.random.randn(2, 4, 5, 3).astype(np.float32))  # batch=2, height=4, width=5, channels=3
y = tf.constant(np.eye(7).astype(np.float32)[:2])  # batch=2, num_classes=7

ARGS = {
    'x': x,
    'y': y,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
absl, numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential 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
clopper_pearson Ran
ran on a synthesized input
jagielski/auditing-dpsgd/code/bkd_parser.py
code served (permissive licence) · get_code("3df71a2c373df03e")
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, ndim 2, 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
from scipy import optimize, stats

ARGS = {
    "count": np.array([5, 10, 15]),
    "trials": np.array([20, 30, 50]),
    "conf": np.array([0.05, 0.01, 0.1])
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, dtype family float — a tuple 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
get_cfg Ran
ran on a synthesized input
jagielski/auditing-dpsgd/code/make_nps.py
code served (permissive licence) · get_code("814bbbf2838c3144")
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, ndim 1, 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 torch

def get_cfg(h5):
    splt = h5.split('-')
    if 'no' in h5:
        return ('no', '.', splt[2], splt[3], splt[4])
    else:
        return ('new', splt[1], splt[2], splt[3], splt[4])

ARGS = {
    'h5': 'prefix-new-config-model-data'
}
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, ndim 1, dtype family float — a tuple 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
parse_name Ran
ran on a synthesized input
jagielski/auditing-dpsgd/code/combine_files.py
code served (permissive licence) · get_code("cb7f2c6a458b26ab")
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, ndim 1, 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 torch
import numpy as np

def parse_name(fname):
    splt = fname.split('-')
    splt[7] = splt[7][:-4]
    return tuple([splt[v] for v in [1, 4, 5, 6, 7]])

# Analyze the function:
# - Takes fname (a filename string)
# - Splits by '-' delimiter
# - Removes last 4 characters from element at index 7
# - Returns a tuple of elements at indices [1, 4, 5, 6, 7]
# So fname needs at least 8 parts when split by '-', and element at index 7 needs at least 4 characters

fname = "part0-part1-part2-part3-part4-part5-part6-part7.txt"

ARGS = {
    "fname": fname
}
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 tuple, ndim 1, dtype family float — a tuple 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
backdoor_train Not yet run
unverified (harvested, no recorded run)
jagielski/auditing-dpsgd/code/audit.py
code served (permissive licence) · get_code("3d53bb923bd2a274")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 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 7 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 investigate whether Differentially Private SGD offers better privacy in practice than what is guaranteed by its state-of-the-art analysis. We do so via novel data poisoning attacks, which we show correspond to realistic privacy attacks. While previous work (Ma et al., arXiv 2019) proposed this connection between differential privacy and data poisoning as a defense against data poisoning, our use as a tool for understanding the privacy of a specific mechanism is new. More generally, our work takes a quantitative, empirical approach to understanding the privacy afforded by specific implementations of differentially private algorithms that we believe has the potential to complement and influence analytical work on differential privacy.

For agents

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

get_harvested_code_for_paper("2006.07709")
get_code_for_paper("2006.07709")
have("2006.07709")

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

curl https://syntology.ai/api/ran/2006.07709.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.07709.svg)](https://syntology.ai/paper/2006.07709)

Connect an agent — have() is free.