SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2302.12480 · 2023

Robust Weight Signatures: Gaining Robustness as Easy as Patching Weights?

arXiv · PDF · Open in the Atlas

Code that ran

11 of 19 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, 1 violated it and 7 have no contract verdict; 3 failures of our instrument; 8 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/2302.12480.json · badge: /api/ran/2302.12480.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
VITA-Group/Robust_Weight_Signatures canonical (classifier label)
classifier basis not recorded on the link
11 ran · 8 no recorded run
FunctionStatusWhere it lives
affine_quant Ran
ran on a synthesized input
VITA-Group/Robust_Weight_Signatures/quantize_utils/quant_utils.py
code served (permissive licence) · get_code("87ed711467d719d3")
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 a small input tensor
x = torch.randn(2, 4, requires_grad=True)

# Create scale and zero_point tensors
scale = torch.tensor(0.5)
zero_point = torch.tensor(0.1)

# inplace flag
inplace = False

ARGS = {
    "x": x,
    "scale": scale,
    "zero_point": zero_point,
    "inplace": inplace,
}
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
attack_pgd Ran
ran on a synthesized input
VITA-Group/Robust_Weight_Signatures/trainer/engine_pgd.py
code served (permissive licence) · get_code("28cadfa8650befa7")
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 4, dtype family float, shape (2, 4, 4, 4) as a later run recorded it. 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, 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
import torch.nn.functional as F

torch.manual_seed(0)

# Define a minimal model that takes input and outputs logits
class SimpleModel(nn.Module):
    def __init__(self, num_classes=3):
        super().__init__()
        self.fc = nn.Linear(4 * 4 * 4, num_classes)
    
    def forward(self, x):
        # Flatten and pass through linear layer
        x = x.view(x.size(0), -1)
        return self.fc(x)

# Create model
model = SimpleModel(num_classes=3)

# Create input tensor: batch_size=2, channels=4, height=4, width=4
X = torch.randn(2, 4, 4, 4)

# Create labels: batch_size=2, values in range [0, 2]
y = torch.tensor([0, 1], dtype=torch.long)

# Define attack parameters
epsilon = 0.3
alpha = 0.1
attack_iters = 3
norm = 'l_inf'
early_stop = False
restarts = 2

ARGS = {
    'model': model,
    'X': X,
    'y': y,
    'epsilon': epsilon,
    'alpha': alpha,
    'attack_iters': attack_iters,
    'norm': norm,
    'early_stop': early_stop,
    'restarts': restarts
}
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 4, dtype family float, shape (2, 4, 4, 4) — shape (2, 4, 4, 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
check_bn Ran
ran, violated the contract we drafted
VITA-Group/Robust_Weight_Signatures/tools.py
pointer only (licence: MIT); the stored copy get_code serves, from JiancongXiao/Adaptive-Smoothness-Weighted-AT, is pointer only (licence: NONE) · get_code("6e59364a7cf6008f")
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 JiancongXiao/Adaptive-Smoothness-Weighted-AT, 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.

conv1x1 Ran
ran; our contract draft was wrong, not the code
VITA-Group/Robust_Weight_Signatures/models/resnet.py
code served (permissive licence) · get_code("d9def42110729a85")
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) and returned Conv2d. 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)

# The function conv1x1 takes in_planes, out_planes, and stride as parameters
# It returns an nn.Conv2d layer configured for 1x1 convolution
# We need to provide valid integer arguments for these parameters

ARGS = {
    "in_planes": 3,
    "out_planes": 64,
    "stride": 1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d 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
conv3x3 Ran
ran; our contract draft was wrong, not the code
VITA-Group/Robust_Weight_Signatures/models/resnet.py
code served (permissive licence) · get_code("160bb14bd76201b4")
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) and returned Conv2d. 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 = {
    "in_planes": 3,
    "out_planes": 64,
    "stride": 1,
    "groups": 1,
    "dilation": 1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d 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
get_dirs Ran
ran on a synthesized input
VITA-Group/Robust_Weight_Signatures/tools.py
code served (permissive licence) · get_code("3aea45f74b2c92f2")
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 os
import tempfile
import shutil

# Create a temporary directory structure for testing
temp_root = tempfile.mkdtemp()

# Create subdirectories with corruption types in their names
corruption_dirs = ['gaussian_blur', 'motion_blur', 'brightness']
for corr_dir in corruption_dirs:
    corr_path = os.path.join(temp_root, corr_dir)
    os.makedirs(corr_path, exist_ok=True)
    # Create a dummy model file
    model_file = os.path.join(corr_path, 'model_RA_best.pth.tar')
    with open(model_file, 'w') as f:
        f.write('dummy model')

# Define arguments
ARGS = {
    'root_dir': temp_root,
    'corruption': ['gaussian_blur', 'motion_blur', 'brightness']
}
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 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
get_projection Ran
ran on a synthesized input
VITA-Group/Robust_Weight_Signatures/tools.py
code served (permissive licence) · get_code("5c47b10f35d34594")
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 a 1D tensor for parameter 'a'
a = torch.randn(4)

# Create a 2D tensor for parameter 'param'
# Shape: (2, 4) - 2 samples, 4 dimensions
param = torch.randn(2, 4)

ARGS = {
    "a": a,
    "param": param,
}
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
make_layers Ran
ran; our contract draft was wrong, not the code
VITA-Group/Robust_Weight_Signatures/models/vgg.py
pointer only (licence: MIT); the stored copy get_code serves, from s-huu/TurningWeaknessIntoStrength, is pointer only (licence: NONE) · get_code("eba3f5bcc6a06d36")
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 s-huu/TurningWeaknessIntoStrength, 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.

signer Ran
ran on a synthesized input
VITA-Group/Robust_Weight_Signatures/quantize_utils/quant_utils.py
code served (permissive licence) · get_code("8357390e4d4ad484")
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 a small tensor for testing
x = torch.randn(2, 4, requires_grad=True)

# The function takes two parameters: x (tensor) and inplace (bool)
ARGS = {
    'x': x,
    'inplace': False
}
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
tinyimagenet_c_testloaders Ran
ran on a synthesized input
VITA-Group/Robust_Weight_Signatures/datasets/tiny_imagenet.py
code served (permissive licence) · get_code("5f6874a404cdccdc")
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 DataLoader. 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 os
import tempfile
import torch
import torchvision.datasets as datasets
import torch.utils.data as data
import torchvision.transforms as transforms
from PIL import Image
import numpy as np

torch.manual_seed(0)

# Create a temporary directory structure that mimics tiny-imagenet-c layout
temp_dir = tempfile.mkdtemp()
val_dir = os.path.join(temp_dir, 'val')
corruption_dir = os.path.join(val_dir, 'gaussian_noise')
severity_dir = os.path.join(corruption_dir, '1')
os.makedirs(severity_dir, exist_ok=True)

# Create some dummy image files for the ImageFolder
class_dir = os.path.join(severity_dir, 'class_0')
os.makedirs(class_dir, exist_ok=True)

# Create a few dummy images
for i in range(2):
    img_array = np.random.randint(0, 256, (64, 64, 3), dtype=np.uint8)
    img = Image.fromarray(img_array)
    img_path = os.path.join(class_dir, f'img_{i}.jpg')
    img.save(img_path)

ARGS = {
    'name': 'gaussian_noise',
    'serverity': '1',
    'batch_size': 2,
    'data_dir': temp_dir,
    'num_workers': 0,
    'shuffle': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataLoader — a DataLoader output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataLoader 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
tinyimagenet_dataloaders Ran
ran on a synthesized input
VITA-Group/Robust_Weight_Signatures/datasets/tiny_imagenet.py
code served (permissive licence) · get_code("36e9a7bb8e92f93c")
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 torch.nn as nn
import torchvision.transforms as transforms
from torch.utils.data import Dataset, DataLoader
import os
import tempfile
from PIL import Image
import numpy as np

torch.manual_seed(0)

# Create a minimal stand-in dataset that mimics ImageFolder behavior
class MinimalImageDataset(Dataset):
    def __init__(self, num_samples=4, transform=None):
        self.num_samples = num_samples
        self.transform = transform
    
    def __len__(self):
        return self.num_samples
    
    def __getitem__(self, idx):
        # Create a tiny random PIL image (4x3 RGB)
        img_array = np.random.randint(0, 256, (4, 3, 3), dtype=np.uint8)
        img = Image.fromarray(img_array)
        label = idx % 2
        
        if self.transform:
            img = self.transform(img)
        
        return img, label

# Create temporary directory structure
temp_dir = tempfile.mkdtemp()
train_dir = os.path.join(temp_dir, 'train')
val_dir = os.path.join(temp_dir, 'val')
os.makedirs(os.path.join(train_dir, 'class_0'), exist_ok=True)
os.makedirs(os.path.join(train_dir, 'class_1'), exist_ok=True)
os.makedirs(os.path.join(val_dir, 'class_0'), exist_ok=True)
os.makedirs(os.path.join(val_dir, 'class_1'), exist_ok=True)

# Create minimal dummy images
for split_dir in [train_dir, val_dir]:
    for class_id in ['class_0', 'class_1']:
        class_path = os.path.join(split_dir, class_id)
        for i in range(2):
            img_array = np.random.randint(0, 256, (4, 3, 3), dtype=np.uint8)
            img = Image.fromarray(img_array)
            img.save(os.path.join(class_path, f'img_{i}.png'))

ARGS = {
    'batch_size': 2,
    'data_dir': temp_dir,
    'num_workers': 0,
    'shuffle': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch, torchvision — 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
cifar100_dataloaders Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/datasets/cifar.py
code served (permissive licence) · get_code("2401e51688b0f4c4")
How it ran

no invocation recorded.

cifar10_dataloaders Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/datasets/cifar.py
code served (permissive licence) · get_code("2d5dd53192074c7a")
How it ran

no invocation recorded.

cifar10c_dataloaders Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/datasets/cifar_c.py
code served (permissive licence) · get_code("6a51cdac8ae07f36")
How it ran

no invocation recorded.

clamp Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/trainer/engine_pgd.py
code served (permissive licence) · get_code("8a93e041134b597a")
How it ran

no invocation recorded.

linear_quant Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/quantize_utils/func.py
code served (permissive licence) · get_code("9970e7f2fc273427")
How it ran

no invocation recorded.

resnet18 Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/models/resnet.py
code served (permissive licence) · get_code("31501fc7e813d642")
How it ran

no invocation recorded.

tinyimagenet_c_trainloaders Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/datasets/tiny_imagenet.py
code served (permissive licence) · get_code("9b92236376542dbd")
How it ran

no invocation recorded.

train_epoch_adv Not yet run
unverified (harvested, no recorded run)
VITA-Group/Robust_Weight_Signatures/trainer/engine_pgd.py
code served (permissive licence) · get_code("22d38cfde2b164df")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 distinct functions with an identity, 6 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 3 the same month, 6 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
conv1x1 431 other papers
240 dated earlier, 5 the same month, 177 later, 8 same year, order not determinable, 1 undated · 38 only as an assembly brick
1409.1556 (2014-09)
Very Deep Convolutional Networks for Large-Scale Image Recognition
1506.01186 (2015-06)
Cyclical Learning Rates for Training Neural Networks
1507.05717 (2015-07)
An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition
and 428 more not named here
conv3x3 312 other papers
155 dated earlier, 3 the same month, 146 later, 7 same year, order not determinable, 1 undated · 31 only as an assembly brick
1409.1556 (2014-09)
Very Deep Convolutional Networks for Large-Scale Image Recognition
1507.05717 (2015-07)
An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition
1512.03385 (2015-12)
Deep Residual Learning for Image Recognition
and 309 more not named here
resnet18 95 other papers
47 dated earlier, 46 later, 1 same year, order not determinable, 1 undated
1805.00907 (2018-05)
Glow: Graph Lowering Compiler Techniques for Neural Networks
1902.08153 (2019-02)
Learned Step Size Quantization
1904.01355 (2019-04)
FCOS: Fully Convolutional One-Stage Object Detection
and 92 more not named here
make_layers 59 other papers
37 dated earlier, 21 later, 1 same year, order not determinable
1411.4038 (2014-11)
Fully Convolutional Networks for Semantic Segmentation
1511.00561 (2015-11)
SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation
1511.05950 (2015-11)
Staleness-aware Async-SGD for Distributed Deep Learning
and 56 more not named here
clamp 33 other papers
19 dated earlier, 1 the same month, 12 later, 1 same year, order not determinable · 1 only as an assembly brick
1706.06083 (2017-06)
Towards Deep Learning Models Resistant to Adversarial Attacks
2001.03994 (2020-01)
Fast is better than free: Revisiting adversarial training
2002.11569 (2020-02)
Overfitting in adversarially robust deep learning
and 30 more not named here
check_bn 8 other papers
5 dated earlier, 2 later, 1 same year, order not determinable
1803.05407 (2018-03)
Averaging Weights Leads to Wider Optima and Better Generalization
1905.02244 (2019-05)
Searching for MobileNetV3
1910.09777 (2019-10)
Self-Correction for Human Parsing
and 5 more not named here

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

Given a robust model trained to be resilient to one or multiple types of distribution shifts (e.g., natural image corruptions), how is that "robustness" encoded in the model weights, and how easily can it be disentangled and/or "zero-shot" transferred to some other models? This paper empirically suggests a surprisingly simple answer: linearly - by straightforward model weight arithmetic! We start by drawing several key observations: (1)assuming that we train the same model architecture on both a clean dataset and its corrupted version, resultant weights mostly differ in shallow layers; (2)the weight difference after projection, which we call "Robust Weight Signature" (RWS), appears to be discriminative and indicative of different corruption types; (3)for the same corruption type, the RWSs obtained by one model architecture are highly consistent and transferable across different datasets. We propose a minimalistic model robustness "patching" framework that carries a model trained on clean data together with its pre-extracted RWSs. In this way, injecting certain robustness to the model is reduced to directly adding the corresponding RWS to its weight. We verify our proposed framework to be remarkably (1)lightweight. since RWSs concentrate on the shallowest few layers and we further show they can be painlessly quantized, storing an RWS is up to 13 x more compact than storing the full weight copy; (2)in-situ adjustable. RWSs can be appended as needed and later taken off to restore the intact clean model. We further demonstrate one can linearly re-scale the RWS to control the patched robustness strength; (3)composable. Multiple RWSs can be added simultaneously to patch more comprehensive robustness at once; and (4)transferable. Even when the clean model backbone is continually adapted or updated, RWSs remain as effective patches due to their outstanding cross-dataset transferability.

For agents

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

get_harvested_code_for_paper("2302.12480")
get_code_for_paper("2302.12480")
have("2302.12480")

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

curl https://syntology.ai/api/ran/2302.12480.json

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

[![Syntology run record](https://syntology.ai/api/ran/2302.12480.svg)](https://syntology.ai/paper/2302.12480)

Connect an agent — have() is free.