SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.16262 · ICML · 2024

Layer-Aware Analysis of Catastrophic Overfitting: Revealing the Pseudo-Robust Shortcut Dependency

Hang Su, Bo Han, Tongliang Liu, Runqi Lin, Chaojian Yu

arXiv · PDF · Open in the Atlas

Code that ran

7 of 8 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; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 5 failures of our instrument; 1 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/2405.16262.json · badge: /api/ran/2405.16262.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
tmllab/2024_ICML_LAP canonical (classifier label)
classifier basis: census
7 ran · 1 no recorded run
FunctionStatusWhere it lives
PreActResNet18 Ran
ran; our contract draft was wrong, not the code
tmllab/2024_ICML_LAP/CIFAR10/preact_resnet.py
code served (permissive licence) · get_code("0234f49f9592dec8")
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 PreActResNet. 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
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "num_classes": 10
}
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 PreActResNet — a PreActResNet output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PreActResNet 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
PreActResNet18 Ran
ran; our contract draft was wrong, not the code
tmllab/2024_ICML_LAP/Tiny-imagenet/preact_resnet.py
code served (permissive licence) · get_code("12838ecee21bc861")
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 PreActResNet. 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
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'num_classes': 10
}
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 PreActResNet — a PreActResNet output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PreActResNet 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
diff_in_weights Ran
ran; our contract draft was wrong, not the code
tmllab/2024_ICML_LAP/Tiny-imagenet/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from Alexkael/S2O, is pointer only (licence: NONE) · get_code("4170809219439d13")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from Alexkael/S2O, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

filter_state_dict Ran
ran; our contract draft was wrong, not the code
tmllab/2024_ICML_LAP/CIFAR10/eval_aa.py
pointer only (licence: MIT); the stored copy get_code serves, from inspire-group/variation-regularization, is pointer only (licence: NONE) · get_code("18b797b7c88af64d")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from inspire-group/variation-regularization, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

find_classes Ran
ran; our contract draft was wrong, not the code
tmllab/2024_ICML_LAP/Tiny-imagenet/tiny_imagenet.py
code served (permissive licence) · get_code("403277d0bfbb477b")
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 tempfile
import os

# Create a temporary file with class names
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
temp_file.write("cat\n")
temp_file.write("dog\n")
temp_file.write("bird\n")
temp_file.close()

ARGS = {
    "class_file": temp_file.name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — 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
is_image_file Ran
ran, violated the contract we drafted
tmllab/2024_ICML_LAP/Tiny-imagenet/tiny_imagenet.py
pointer only (licence: MIT); the stored copy get_code serves, from sgoldyaev/DeepFashion.ADGAN, is pointer only (licence: NONE) · get_code("ab4109634b75ef8b")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from sgoldyaev/DeepFashion.ADGAN, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

loadPILImage Ran
ran, honoured the contract we drafted
tmllab/2024_ICML_LAP/Tiny-imagenet/tiny_imagenet.py
code served (permissive licence) · get_code("27e7954cc9b19461")
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 Image. 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
import numpy as np
from PIL import Image
import io

# Create a small test image in memory
img_array = np.random.randint(0, 256, (10, 10, 3), dtype=np.uint8)
test_img = Image.fromarray(img_array, mode='RGB')

# Save to a temporary file-like object path
# Since we can't do real file I/O, we'll create a temporary path string
# and mock the Image.open to work with it
import tempfile
import os

# Create a temporary directory and file
temp_dir = tempfile.gettempdir()
temp_path = os.path.join(temp_dir, "test_image_temp.png")

# Save the test image
test_img.save(temp_path)

ARGS = {
    "path": temp_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
get_loaders Not yet run
unverified (harvested, no recorded run)
tmllab/2024_ICML_LAP/CIFAR10/utils.py
code served (permissive licence) · get_code("816caa026adecbe6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 6 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 1 the same month, 5 later) and 1 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
is_image_file 62 other papers
55 dated earlier, 5 later, 2 same year, order not determinable
1609.04802 (2016-09)
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
1703.07511 (2017-03)
Deep Photo Style Transfer
1710.10916 (2017-10)
StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks
and 59 more not named here
PreActResNet18 22 other papers
18 dated earlier, 1 the same month, 3 later
1710.09412 (2017-10)
mixup: BEYOND EMPIRICAL RISK MINIMIZATION
2003.04887 (2020-03)
ReZero is All You Need: Fast Convergence at Large Depth
2103.14222 (2021-03)
Adversarial Attacks are Reversible with Natural Supervision
and 19 more not named here
diff_in_weights 7 other papers
6 dated earlier, 1 later · 1 only as an assembly brick
1901.08573 (2019-01)
Theoretically Principled Trade-off between Robustness and Accuracy
2208.08270 (2022-08)
On the Privacy Effect of Data Enhancement via the Lens of Memorization
2210.03543 (2022-10)
A2: Efficient Automated Attacker for Boosting Adversarial Training
and 4 more not named here
find_classes 7 other papers
5 dated earlier, 2 later
2103.13027 (2021-03)
AutoMix: Unveiling the Power of Mixup for Stronger Classifiers
2209.04996 (2022-09)
Switchable Online Knowledge Distillation
2310.00258 (2023-10)
NAYER: Noisy Layer Data Generation for Efficient and Effective Data-free Knowledge Distillation
and 4 more not named here
filter_state_dict 5 other papers
5 dated earlier
2103.14347 (2021-03)
Combating Adversaries with Anti-Adversaries
2107.14110 (2021-07)
Enhancing Adversarial Robustness via Test-time Transformation Ensembling
2204.13779 (2022-04)
Formulating Robustness Against Unforeseen Attacks
and 2 more not named here
loadPILImage 3 other papers
2 dated earlier, 1 later
2209.04996 (2022-09)
Switchable Online Knowledge Distillation
2404.08154 (2024-04)
Eliminating Catastrophic Overfitting Via Abnormal Adversarial Examples Regularization
2606.00738 (2026-06)
SORA: Free Second-Order Attacks in Fast Adversarial Training

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

Catastrophic overfitting (CO) presents a significant challenge in single-step adversarial training (AT), manifesting as highly distorted deep neural networks (DNNs) that are vulnerable to multi-step adversarial attacks. However, the underlying factors that lead to the distortion of decision boundaries remain unclear. In this work, we delve into the specific changes within different DNN layers and discover that during CO, the former layers are more susceptible, experiencing earlier and greater distortion, while the latter layers show relative insensitivity. Our analysis further reveals that this increased sensitivity in former layers stems from the formation of pseudo-robust shortcuts, which alone can impeccably defend against singlestep adversarial attacks but bypass genuine-robust learning, resulting in distorted decision boundaries. Eliminating these shortcuts can partially restore robustness in DNNs from the CO state, thereby verifying that dependence on them triggers the occurrence of CO. This understanding motivates us to implement adaptive weight perturbations across different layers to hinder the generation of pseudo-robust shortcuts, consequently mitigating CO. Extensive experiments demonstrate that our proposed method, Layer-Aware Adversarial Weight Perturbation (LAP), can effectively prevent CO and further enhance robustness. Our implementation can be found at https:// github.com/tmllab/2024_ICML_LAP.

For agents

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

get_harvested_code_for_paper("2405.16262")
get_code_for_paper("2405.16262")
have("2405.16262")

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

curl https://syntology.ai/api/ran/2405.16262.json

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

[![Syntology run record](https://syntology.ai/api/ran/2405.16262.svg)](https://syntology.ai/paper/2405.16262)

Connect an agent — have() is free.