SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1905.04172 · 2019

On the Connection Between Adversarial Robustness and Saliency Map Interpretability

arXiv · PDF · Open in the Atlas

Code that ran

6 of 16 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; 10 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/1905.04172.json · badge: /api/ran/1905.04172.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
cetmann/robustness-interpretability canonical (classifier label)
classifier basis: census
6 ran · 10 no recorded run
FunctionStatusWhere it lives
bias_shifted_input Ran
ran on a synthesized input
cetmann/robustness-interpretability/robust_model.py
code served (permissive licence) · get_code("290432e2a93b956c")
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 EagerTensor, ndim 4, dtype family float, shape (2, 4, 3, 5) 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 tensorflow as tf
import numpy as np

# Create small test tensors
batch_size = 2
height = 4
width = 3
channels = 5

# x: input tensor of shape [batch, height, width, channels]
x = tf.constant(np.random.randn(batch_size, height, width, channels), dtype=tf.float32)

# b: bias vector of shape [batch]
b = tf.constant(np.random.randn(batch_size), dtype=tf.float32)

# direction: direction tensor of shape [batch, height, width, channels]
direction = tf.constant(np.random.randn(batch_size, height, width, channels), dtype=tf.float32)

ARGS = {
    'x': x,
    'b': b,
    'direction': direction
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (2, 4, 3, 5) — shape (2, 4, 3, 5) 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
collect_summaries Ran
ran on a synthesized input
cetmann/robustness-interpretability/summary_utils.py
code served (permissive licence) · get_code("73e9801abcef43f3")
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 tensorflow as tf
import numpy as np

# Create a list of summary dictionaries with different keys and tensor values
summary_dict_list = [
    {
        "loss": tf.constant(0.5, dtype=tf.float32),
        "accuracy": tf.constant(0.85, dtype=tf.float32),
        "learning_rate": tf.constant(0.001, dtype=tf.float32),
    },
    {
        "loss": tf.constant(0.4, dtype=tf.float32),
        "accuracy": tf.constant(0.88, dtype=tf.float32),
        "learning_rate": tf.constant(0.001, dtype=tf.float32),
    },
    {
        "loss": tf.constant(0.35, dtype=tf.float32),
        "accuracy": tf.constant(0.90, dtype=tf.float32),
        "learning_rate": tf.constant(0.001, dtype=tf.float32),
    },
]

ARGS = {
    "summary_dict_list": summary_dict_list,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — 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
create_filter_bank Ran
ran on a synthesized input
cetmann/robustness-interpretability/fwt.py
code served (permissive licence) · get_code("2da11163c7dd1ff0")
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 5, 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 tensorflow as tf
import pywt

# The function takes one parameter: wavelet_type
# It should be a string that is compatible with pywt.Wavelet()
# Common valid wavelet types include 'bior2.2', 'db1', 'haar', etc.

ARGS = {
    'wavelet_type': 'bior2.2'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pywt, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 5, 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
fwt Ran
ran on a synthesized input
cetmann/robustness-interpretability/fwt.py
code served (permissive licence) · get_code("00affbb71bdec4df")
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 EagerTensor, ndim 4, dtype family float, shape (2, 8, 8, 1) 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 tensorflow as tf
import numpy as np

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

# Create a small input image tensor
# Shape: [batch, height, width, channels]
# Using batch=2, height=8, width=8, channels=1
f_img = tf.constant(np.random.randn(2, 8, 8, 1).astype(np.float32))

# Create a filter bank tensor
# The function expects shape [filter_height, filter_width, in_channels, out_channels]
# Looking at the code: filter_shape[0] // 2 - 1 and filter_shape[1] // 2 - 1
# So we need even dimensions. Using 4x4 filter with 1 input channel and 4 output channels
# (for a, d_1, d_2, d_3 decomposition)
filter_bank = tf.constant(np.random.randn(4, 4, 1, 4).astype(np.float32))

# Create arguments dictionary
ARGS = {
    'f_img': f_img,
    'filter_bank': filter_bank,
    'scale': 1,
    'pad_mode': 'REFLECT',
    'output_type': 'image',
    'method': 'strided'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (2, 8, 8, 1) — shape (2, 8, 8, 1) 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
interpret_as_image Ran
ran on a synthesized input
cetmann/robustness-interpretability/preprocessing.py
code served (permissive licence) · get_code("3a20ef1704a91774")
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 4, dtype family float, shape (2, 4, 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 NHWC image (batch=2, height=4, width=3, channels=3 for RGB)
image = np.random.randn(2, 4, 3, 3).astype('float32')

ARGS = {
    'image': image,
    'add_by': 0.0,
    'multiply_by': 1.0,
    'colorspace': 'RGB'
}
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 4, dtype family float, shape (2, 4, 3, 3) — shape (2, 4, 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
multi_channel_fwt Ran
ran on a synthesized input
cetmann/robustness-interpretability/fwt.py
code served (permissive licence) · get_code("e66107cb7dc652d4")
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 EagerTensor, ndim 4, dtype family float, shape (2, 8, 8, 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 tensorflow as tf
import numpy as np

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

# Create a small input image: batch_size=2, height=8, width=8, channels=3
f_img = tf.constant(np.random.randn(2, 8, 8, 3).astype(np.float32))

# Create a filter bank for wavelet transform
# Shape should be (filter_height, filter_width, in_channels, out_channels)
# For wavelet transform with 4 output filters (approximation + 3 details)
# Using a Daubechies-like filter bank
filter_bank = tf.constant(
    np.random.randn(4, 4, 1, 4).astype(np.float32) * 0.1
)

ARGS = {
    'f_img': f_img,
    'filter_bank': filter_bank,
    'scale': 1,
    'pad_mode': 'REFLECT',
    'output_type': 'image'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (2, 8, 8, 3) — shape (2, 8, 8, 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
Lop Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/diff_ops.py
code served (permissive licence) · get_code("cd4c53c68d14383e")
How it ran

no invocation recorded.

Rop Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/diff_ops.py
code served (permissive licence) · get_code("c91e29ab9169cee1")
How it ran

no invocation recorded.

conv_block Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/resnet50.py
code served (permissive licence) · get_code("8ceb111da64aa651")
How it ran

no invocation recorded.

create_model Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/smallnet.py
code served (permissive licence) · get_code("eb60f5c054b90a0b")
How it ran

no invocation recorded.

create_summary_op Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/summary_utils.py
code served (permissive licence) · get_code("5dd668c0f769debb")
How it ran

no invocation recorded.

identity_block Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/resnet50.py
code served (permissive licence) · get_code("764560d8341cc583")
How it ran

no invocation recorded.

image_preprocessing Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/preprocessing.py
code served (permissive licence) · get_code("834b1e6f61965aa3")
How it ran

no invocation recorded.

prepare_summaries Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/summary_utils.py
code served (permissive licence) · get_code("a37437994d53bdf4")
How it ran

no invocation recorded.

train_BI Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/mnist_data.py
code served (permissive licence) · get_code("e1d633264f33d554")
How it ran

no invocation recorded.

val_BI Not yet run
unverified (harvested, no recorded run)
cetmann/robustness-interpretability/mnist_data.py
code served (permissive licence) · get_code("d3e664e0da0368a9")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 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 16 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

Recent studies on the adversarial vulnerability of neural networks have shown that models trained to be more robust to adversarial attacks exhibit more interpretable saliency maps than their non-robust counterparts. We aim to quantify this behavior by considering the alignment between input image and saliency map. We hypothesize that as the distance to the decision boundary grows,so does the alignment. This connection is strictly true in the case of linear models. We confirm these theoretical findings with experiments based on models trained with a local Lipschitz regularization and identify where the non-linear nature of neural networks weakens the relation.

For agents

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

get_harvested_code_for_paper("1905.04172")
get_code_for_paper("1905.04172")
have("1905.04172")

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

curl https://syntology.ai/api/ran/1905.04172.json

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

[![Syntology run record](https://syntology.ai/api/ran/1905.04172.svg)](https://syntology.ai/paper/1905.04172)

Connect an agent — have() is free.