SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · aaai_28047 · AAAI · 2024

Rethinking Robustness of Model Attributions

Sandesh Kamath, Amit Deshpande, Vineeth Balasubramanian, Sankalp Mittal

Open in the Atlas

Code that ran

11 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 10 have no contract verdict; 1 failure of our instrument; 4 no recorded run), measured 2026-11-22; 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/aaai_28047.json · badge: /api/ran/aaai_28047.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
ksandeshk/LENS canonical (classifier label)
classifier basis: census
11 ran · 4 no recorded run
FunctionStatusWhere it lives
pad_sequences Ran
ran on a synthesized input
ksandeshk/LENS/Flower/data_utils.py
code served (permissive licence) · get_code("8d57ffc220b82a2b")
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 int, shape (4, 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 numpy as np

# Create test sequences of varying lengths
sequences = [
    [1, 2, 3],
    [4, 5],
    [6, 7, 8, 9, 10],
    [11]
]

# Define all arguments for pad_sequences
ARGS = {
    'sequences': sequences,
    'maxlen': 5,
    'dtype': 'int32',
    'padding': 'post',
    'truncating': 'post',
    'value': 0.0
}
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 int, shape (4, 5) — shape (4, 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
per_image_standardization Ran
ran on a synthesized input
ksandeshk/LENS/Flower/model.py
code served (permissive licence) · get_code("d81479ccd5885af0")
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 a small test image tensor
# Shape: [batch, height, width, channels]
# Using distinct dimensions: batch=2, height=4, width=3, channels=5
image_data = np.random.randn(2, 4, 3, 5).astype(np.float32)
image_tensor = tf.convert_to_tensor(image_data)

ARGS = {
    'image': image_tensor
}
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
projection_transform Ran
ran on a synthesized input
ksandeshk/LENS/Flower/generate_data.py
code served (permissive licence) · get_code("52613a2b19ff74b4")
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 3, dtype family float, shape (128, 128, 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
from skimage.transform import warp, ProjectiveTransform

# Create a small test image
# The function expects an image array that can be warped
# Using a small 128x128 RGB image for testing
image = np.random.rand(128, 128, 3).astype(np.float32)

ARGS = {
    "image": image,
    "max_warp": 0.8,
    "height": 128,
    "width": 128,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, skimage — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (128, 128, 3) — shape (128, 128, 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
rotate_image Ran
ran on a synthesized input
ksandeshk/LENS/Flower/generate_data.py
code served (permissive licence) · get_code("185624ea549c1f71")
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 numpy as np
from skimage.transform import rotate

# Create a small test image
# rotate() expects a 2D or 3D array
# Using a simple 4x3 grayscale image
image = np.random.rand(4, 3)

ARGS = {
    "image": image,
    "max_angle": 15
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, skimage — 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
string_to_semi_redundant_sequences Ran
ran on a synthesized input
ksandeshk/LENS/Flower/data_utils.py
code served (permissive licence) · get_code("c1e432f7d04d1dad")
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. 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

# Create a small test string with diverse characters
test_string = "hello world this is a test string"

ARGS = {
    "string": test_string,
    "seq_maxlen": 10,
    "redun_step": 2,
    "char_idx": None
}
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 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
submat Ran
ran on a synthesized input
ksandeshk/LENS/MNIST/eval_with_lens_attribution_attack.py
code served (permissive licence) · get_code("d9542b6396e26477")
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 torch
import numpy as np

torch.manual_seed(0)

# Analyzing the function: submat(x)
# - Takes parameter x
# - Performs arithmetic operations on x (addition, subtraction)
# - Returns a list of results from these operations
# - x needs to support +, - operators
# - Based on the operations, x should be a numeric type or tensor

# Create a small tensor for x
x = torch.tensor([[1.0, 2.0, 3.0, 4.0],
                   [5.0, 6.0, 7.0, 8.0]])

ARGS = {
    'x': x
}
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 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
submat3x3 Ran
ran on a synthesized input
ksandeshk/LENS/Flower/eval_with_lens_attribution_attack.py
code served (permissive licence) · get_code("f30c328273000681")
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 torch
import numpy as np

torch.manual_seed(0)

# Analyze the function: submat3x3(x)
# - Takes parameter x
# - Performs arithmetic operations on x (addition, subtraction)
# - Returns a list of 9 results
# - x should support arithmetic operations, so a tensor works well

# Create a small tensor for x
x = torch.tensor([[1.0, 2.0, 3.0, 4.0],
                   [5.0, 6.0, 7.0, 8.0]])

ARGS = {
    'x': x
}
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 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
submat5x5 Ran
ran on a synthesized input
ksandeshk/LENS/Flower/eval_with_lens_attribution_attack.py
code served (permissive licence) · get_code("bf901c1e2658b8c2")
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 torch
import numpy as np

torch.manual_seed(0)

# The function submat5x5 takes a single parameter x
# It performs arithmetic operations on x (addition, subtraction)
# and returns a list of transformed versions of x
# x should be a numeric type that supports these operations
# A torch tensor is appropriate here

x = torch.tensor([[1.0, 2.0, 3.0, 4.0],
                  [5.0, 6.0, 7.0, 8.0]], dtype=torch.float32)

ARGS = {
    "x": x
}
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 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
submat7x7 Ran
ran on a synthesized input
ksandeshk/LENS/Flower/eval_with_lens_attribution_attack.py
code served (permissive licence) · get_code("aab399cfd20ab1e1")
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 torch

torch.manual_seed(0)

# The function takes a single parameter 'x' and performs arithmetic operations on it
# It returns a list of tensors/values derived from x
# x needs to support addition and subtraction operations

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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 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
to_categorical Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ksandeshk/LENS/Flower/data_utils.py
code served (permissive licence) · get_code("ca9076577f084d46")
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 (6, 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

# Construct arguments for to_categorical function
y = np.array([0, 1, 2, 1, 0, 3])  # class labels from 0 to 3
nb_classes = 4  # total number of classes

ARGS = {
    'y': y,
    'nb_classes': nb_classes
}
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 (6, 4) — shape (6, 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
translate_image Ran
ran on a synthesized input
ksandeshk/LENS/Flower/generate_data.py
code served (permissive licence) · get_code("2996883b55528d1d")
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 3, dtype family int, shape (128, 128, 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
import cv2

# Create a small test image (height=100, width=100, 3 channels for RGB)
test_image = np.random.randint(0, 256, size=(100, 100, 3), dtype=np.uint8)

ARGS = {
    "image": test_image,
    "max_trans": 5,
    "height": 128,
    "width": 128,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (128, 128, 3) — shape (128, 128, 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
addwindowList Not yet run
unverified (harvested, no recorded run)
ksandeshk/LENS/MNIST/eval_with_lens_attribution_attack.py
code served (permissive licence) · get_code("d9e577e2ae36a930")
How it ran

no invocation recorded.

addwindowPx Not yet run
unverified (harvested, no recorded run)
ksandeshk/LENS/MNIST/eval_with_lens_attribution_attack.py
code served (permissive licence) · get_code("e5b5c01239577d25")
How it ran

no invocation recorded.

load_data Not yet run
unverified (harvested, no recorded run)
ksandeshk/LENS/Flower/flower_input.py
code served (permissive licence) · get_code("7cba5a1e5ed21fc0")
How it ran

no invocation recorded.

maybe_download Not yet run
unverified (harvested, no recorded run)
ksandeshk/LENS/Flower/flower_input.py
code served (permissive licence) · get_code("93b364e33723ced3")
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, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 0 later) and 13 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.

FunctionAlso present inEarliest-datedOthers named
to_categorical 3 other papers
3 dated earlier
1606.08415 (2016-06)
Gaussian Error Linear Units (GELUs)
1608.00530 (2016-08)
Early Methods for Detecting Adversarial Images
1610.02136 (2016-10)
A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
per_image_standardization 1 other paper
1 dated earlier
1905.09957 (2019-05)
Robust Attribution Regularization

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

For machine learning models to be reliable and trustworthy, their decisions must be interpretable. As these models find increasing use in safety-critical applications, it is important that not just the model predictions but also their explanations (as feature attributions) be robust to small human-imperceptible input perturbations. Recent works have shown that many attribution methods are fragile and have proposed improvements in either these methods or the model training. We observe two main causes for fragile attributions: first, the existing metrics of robustness (e.g., top-k intersection) overpenalize even reasonable local shifts in attribution, thereby making random perturbations to appear as a strong attack, and second, the attribution can be concentrated in a small region even when there are multiple important parts in an image. To rectify this, we propose simple ways to strengthen existing metrics and attribution methods that incorporate locality of pixels in robustness metrics and diversity of pixel locations in attributions. Towards the role of model training in attributional robustness, we empirically observe that adversarially trained models have more robust attributions on smaller datasets, however, this advantage disappears in larger datasets. Code is made available at https://github.com/ksandeshk/LENS.

For agents

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

get_harvested_code_for_paper("aaai_28047")
get_code_for_paper("aaai_28047")
have("aaai_28047")

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

curl https://syntology.ai/api/ran/aaai_28047.json

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

[![Syntology run record](https://syntology.ai/api/ran/aaai_28047.svg)](https://syntology.ai/paper/aaai_28047)

Connect an agent — have() is free.