SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.02708 · NeurIPS · 2023

Provable Adversarial Robustness for Group Equivariant Tasks: Graphs, Point Clouds, Molecules, and More

Stephan Günnemann, Jan Schuchardt, Yan Scholten

arXiv · PDF · Open in the Atlas

Code that ran

22 of 31 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 7 honoured the contract we drafted, 3 violated it and 2 have no contract verdict; 10 failures of our instrument; 9 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/2312.02708.json · badge: /api/ran/2312.02708.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
RobustGraph/RoboGraph canonical (classifier label)
classifier basis: census
12 ran · 3 no recorded run
jan-schuchardt/invariance-smoothing canonical (classifier label)
classifier basis: census
6 ran · 4 no recorded run
copy not recorded community
not in the paper's declared repositories
2 ran · 0 no recorded run
danielzuegner/robust-gcn community
Papers with Code archive
1 ran · 1 no recorded run
abojchevski/sparse_smoothing community
Papers with Code archive
1 ran · 0 no recorded run
wollschl/uncertainty_for_molecules community
Papers with Code archive
0 ran · 1 no recorded run
FunctionStatusWhere it lives
RobustGCNLayer Ran
ran on a synthesized input · constructed an object; no result was computed
danielzuegner/robust-gcn/robust_gcn/robust_gcn.py
code served (permissive licence) · get_code("a268573b18831a29")
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 an instance of its own class, RobustGCNLayer: the run constructed the object and did not call it, so no result was computed. 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 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
from torch import nn
import scipy.sparse as sp
import numpy as np

torch.manual_seed(0)

# Create a small sparse adjacency matrix
# Using COO format then converting to CSR
rows = np.array([0, 1, 2, 3, 0, 2])
cols = np.array([1, 2, 3, 0, 3, 1])
data = np.ones(len(rows), dtype='float32')
adj = sp.coo_matrix((data, (rows, cols)), shape=(4, 4)).tocsr()

# Define dimensions: (input_dim, output_dim)
dims = (5, 3)

ARGS = {
    "adj": adj,
    "dims": dims,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type RobustGCNLayer — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a RobustGCNLayer 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
SPG Ran
ran; our contract draft was wrong, not the code
RobustGraph/RoboGraph/robograph/attack/SPG.py
code served (permissive licence) · get_code("9073926e56ac1c34")
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

class SPGOptions:
    def __init__(self):
        self.verbose = 0
        self.numdiff = 0
        self.testOpt = False
        self.maxIter = 100
        self.useSpectral = True
        self.bbType = 1
        self.curvilinear = False
        self.progTol = 1e-5
        self.memory = 10
        self.suffDec = 1e-4
        self.interp = 2

def simple_funObj(x):
    f = np.sum(x**2)
    g = 2 * x
    return (f, g)

def simple_funProj(x):
    return np.clip(x, -10, 10)

x_init = np.array([1.0, 2.0, 3.0, 4.0])
options = SPGOptions()

ARGS = {
    'funObj0': simple_funObj,
    'funProj': simple_funProj,
    'x': x_init,
    'options': options
}
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
add_gaussian_noise Ran
ran; our contract draft was wrong, not the code
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/training.py
code served (permissive licence) · get_code("fee01cc7d1d5c624")
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)

x = torch.randn(2, 4)
std = 0.5
device = torch.device('cpu')

ARGS = {
    'x': x,
    'std': std,
    'device': device,
}
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
calculate_budget Ran
ran, honoured the contract we drafted
RobustGraph/RoboGraph/robograph/attack/genetic_attack.py
code served (permissive licence) · get_code("637b58808cd09ffa")
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 int, ndim 0, dtype family int, shape () 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 two small numpy arrays representing adjacency matrices or similar structures
# A_org is the original array
# A_perturbed is the perturbed version with some differences
A_org = np.array([[1, 0, 1, 0],
                   [0, 1, 0, 1]], dtype=np.int32)

A_perturbed = np.array([[1, 1, 1, 0],
                         [0, 1, 0, 0]], dtype=np.int32)

ARGS = {
    "A_org": A_org,
    "A_perturbed": A_perturbed
}
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 int, ndim 0, dtype family int, shape () — shape () 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_local Ran
ran, violated the contract we drafted
RobustGraph/RoboGraph/robograph/attack/genetic_attack.py
code served (permissive licence) · get_code("313a20ad80cd5b01")
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 int, ndim 0, dtype family int, shape () 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 small numpy arrays for testing
# A_org_row: original row
A_org_row = np.array([1.0, 2.5, 3.2, 4.1])

# A_perturbed_row: perturbed version of the row
A_perturbed_row = np.array([1.1, 2.3, 3.5, 4.0])

ARGS = {
    "A_org_row": A_org_row,
    "A_perturbed_row": A_perturbed_row,
}
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 int, ndim 0, dtype family int, shape () — shape () 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
closed_form_B Ran
ran, honoured the contract we drafted
RobustGraph/RoboGraph/robograph/attack/admm.py
code served (permissive licence) · get_code("a01aa6ebe170c0dc")
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, 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 closed_form_B function
# The function expects:
# - lamb: a scalar (float)
# - A_pert: a 2D numpy array (nG x nG)
# - mu: a scalar (float)

nG = 4  # number of nodes/groups

# Create a small perturbation matrix
A_pert = np.array([
    [0.1, 0.2, 0.15, 0.05],
    [0.2, 0.1, 0.1, 0.2],
    [0.15, 0.1, 0.1, 0.25],
    [0.05, 0.2, 0.25, 0.1]
])

# Create scalar parameters
lamb = 0.5
mu = 0.1

ARGS = {
    "lamb": lamb,
    "A_pert": A_pert,
    "mu": mu
}
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 (4, 4) — shape (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
compute_rho Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
abojchevski/sparse_smoothing/sparse_smoothing/cert.py
pointer only (licence: NONE) · get_code("13a44cb31287e83c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, abojchevski/sparse_smoothing 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.

dict_to_dot Ran
ran; our contract draft was wrong, not the code
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/utils.py
code served (permissive licence) · get_code("2d5870b6c4fe5b59")
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 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 torch

torch.manual_seed(0)

# The function dict_to_dot takes a nested dictionary and converts it to a flat list of tuples
# with dot notation keys. It recursively processes nested dicts and flattens them.
# The function only requires:
# - x: a dict (can be nested or flat)

# Create test cases with various nesting levels
nested_dict = {
    'a': 1,
    'b': 2,
    'c': {
        'd': 3,
        'e': 4,
        'f': {
            'g': 5,
            'h': 6
        }
    },
    'i': {
        'j': 7
    }
}

ARGS = {
    'x': nested_dict
}
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 2, 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
eval Ran
ran, honoured the contract we drafted
RobustGraph/RoboGraph/robograph/model/gnn.py
code served (permissive licence) · get_code("4bae3496a287a478")
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 float, ndim 0, dtype family float, shape () 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
import torch.nn as nn
from torch_geometric.data import Data, DataLoader

torch.manual_seed(0)

# Create a minimal GC_NET-like model
class GC_NET(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, data):
        # Expect data to have x and y attributes
        output = self.linear(data.x)
        return output

# Create minimal graph data objects with required attributes
data_list = []
for _ in range(2):
    x = torch.randn(2, 4)  # 2 nodes, 4 features
    y = torch.tensor([0, 1])  # 2 node labels
    data = Data(x=x, y=y)
    data_list.append(data)

# Create a DataLoader
loader = DataLoader(data_list, batch_size=2)

# Create model instance
model = GC_NET()
model.eval()

ARGS = {
    'model': model,
    'loader': loader,
    'testing': False,
    'save_path': None,
    'robust': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torch_geometric — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () 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
forward_cert_cohen Ran
ran, honoured the contract we drafted
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/certificates/blackbox.py
code served (permissive licence) · get_code("de80f2bd3a451839")
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 torch
import numpy as np
from scipy.stats import norm

torch.manual_seed(0)

ARGS = {
    "sigma": 0.5,
    "norm_delta": 0.3,
    "p_lower": 0.8,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — 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
inverse_cert_cohen Ran
ran, honoured the contract we drafted
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/certificates/blackbox.py
code served (permissive licence) · get_code("e2bcee4d8b42f4b1")
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.

from scipy.stats import norm

ARGS = {
    "sigma": 0.5,
    "norm_delta": 0.3,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
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
isLegal Ran
ran, violated the contract we drafted
RobustGraph/RoboGraph/robograph/attack/SPG.py
code served (permissive licence) · get_code("02d6a2dca6a6dc32")
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 bool, ndim 0, dtype family float, shape () 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 valid numpy array with real (non-complex) values, no NaN, no inf
v = np.array([1.0, 2.5, 3.7, 4.2], dtype=np.float64)

ARGS = {
    "v": v
}
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 bool, ndim 0, dtype family float, shape () — shape () 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
load_graph Ran
ran; our contract draft was wrong, not the code
RobustGraph/RoboGraph/robograph/utils.py
code served (permissive licence) · get_code("b6d7cd2f3295bedf")
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 pickle as pkl
import tempfile
import os

# Create a temporary directory and pickle file
temp_dir = tempfile.mkdtemp()
test_data = {
    'graph': {'nodes': 5, 'edges': 10},
    'parameters': {'weight': 0.5, 'bias': 0.1}
}

pickle_file = os.path.join(temp_dir, 'test_graph.pickle')
with open(pickle_file, 'wb') as f:
    pkl.dump(test_data, f)

ARGS = {
    'filename': pickle_file
}
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
local_solver Ran
ran on a synthesized input
RobustGraph/RoboGraph/robograph/attack/dp_old_version.py
code served (permissive licence) · get_code("dc92a95e2bacf927")
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

# Analyze the function to understand parameter requirements:
# - A_org: indexed as A_org[i-1, :], so needs shape (nG, something)
#   Used with A_i = A_org[i-1, :], A_i[x] checks, np.dot(A_i + canonical, XWU)
# - XWU: used with np.argsort(XWU), XWU[x] indexed, np.dot(..., XWU)
# - nG: used as range(1, nG+1), creating arrays of size nG
# - delta_l: used as range(0, delta_l+1)

# Create a small example:
# nG = 3 means 3 nodes/groups
# We need A_org to be a matrix where each row has nG entries (adjacency/feature matrix)
# XWU should be a vector of length nG

nG = 3
delta_l = 2

# A_org: (nG, nG) matrix - adjacency or edge matrix for nG nodes
# Initialize with some structure - a sparse adjacency matrix
A_org = np.array([
    [0, 1, 0],  # node 0: connected to node 1
    [1, 0, 1],  # node 1: connected to nodes 0 and 2
    [0, 1, 0]   # node 2: connected to node 1
], dtype=float)

# XWU: vector of length nG with some values to score/rank nodes
XWU = np.array([0.5, 2.1, 1.3], dtype=float)

ARGS = {
    "A_org": A_org,
    "XWU": XWU,
    "nG": nG,
    "delta_l": delta_l
}
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
normalize_adj Ran
ran; our contract draft was wrong, not the code
RobustGraph/RoboGraph/robograph/model/gcn_conv.py
code served (permissive licence) · get_code("c55badf69fe079ef")
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 (4, 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 adjacency matrix (4x4)
# The function expects a square matrix since it adds eye(A.shape[0])
A = torch.tensor([
    [0.0, 1.0, 0.0, 1.0],
    [1.0, 0.0, 1.0, 0.0],
    [0.0, 1.0, 0.0, 1.0],
    [1.0, 0.0, 1.0, 0.0]
], dtype=torch.float32)

ARGS = {
    "A": A,
    "c": 1
}
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 (4, 4) — shape (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
polyinterp2 Ran
ran, violated the contract we drafted
RobustGraph/RoboGraph/robograph/attack/SPG.py
code served (permissive licence) · get_code("4d997c194830837b")
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

# The function polyinterp2 takes a single parameter 'points'
# which is used as a numpy array with shape (n, 3)
# It accesses points[:, 0], points[:, 1], points[:, 2]
# and uses argmin, so we need at least 2 rows to make notMinPos work

# Create a simple points array with 2 rows and 3 columns
# Column 0: x-coordinates (will use argmin on this)
# Column 1: y-coordinates
# Column 2: derivatives or slopes
points = np.array([
    [1.5, 2.0, 0.5],
    [0.3, 1.0, -0.2]
], dtype=np.float64)

ARGS = {
    "points": points
}
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 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
preprocess_adj Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from danielzuegner/robust-gcn
pointer only · get_code("a3cccfaa8bd4c86c")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. 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.

process_data Ran
ran; our contract draft was wrong, not the code
RobustGraph/RoboGraph/robograph/utils.py
code served (permissive licence) · get_code("802848e9f51394d1")
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 torch
import numpy as np
import scipy.sparse as sp

torch.manual_seed(0)

# Create a minimal torch_geometric Data object
class MinimalData:
    def __init__(self):
        # edge_index: shape [2, num_edges] - 5 edges in this case
        self.edge_index = torch.tensor([
            [0, 1, 2, 1, 3],
            [1, 2, 3, 0, 2]
        ], dtype=torch.long)
        
        # x: node features, shape [num_nodes, num_features]
        # 4 nodes, 3 features each
        self.x = torch.randn(4, 3, dtype=torch.float32)
        
        # y: label
        self.y = 1

data = MinimalData()

ARGS = {
    "data": data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — 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
randomly_scale Ran
ran; our contract draft was wrong, not the code
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/training.py
code served (permissive licence) · get_code("be9f96362f6c27b7")
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 3, dtype family float, shape (2, 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 torch

torch.manual_seed(0)

# Create a small batch of point clouds
# Shape: (batch_size=2, num_points=4, features=3)
x = torch.randn(2, 4, 3)

scale_min = 0.5
scale_max = 1.5

device = torch.device('cpu')

ARGS = {
    'x': x,
    'scale_min': scale_min,
    'scale_max': scale_max,
    'device': device
}
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 3, dtype family float, shape (2, 4, 3) — shape (2, 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
sample_single_axis_likelihood_ratio_logs Ran
ran, honoured the contract we drafted
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/certificates/rotation_elemental.py
code served (permissive licence) · get_code("1daf26cd6d81ba89")
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 1, dtype family float, shape (100,) 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

ARGS = {
    "clean": True,
    "sigma": 0.5,
    "norm_delta": 0.3,
    "n_samples": 100
}
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 1, dtype family float, shape (100,) — shape (100,) 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
sparse_tensor Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from danielzuegner/robust-gcn
pointer only · get_code("c25e88689b0251b5")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. 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.

train Ran
ran, honoured the contract we drafted
RobustGraph/RoboGraph/robograph/model/gnn.py
code served (permissive licence) · get_code("3df5b2fc12d5a4ec")
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 float, ndim 0, dtype family float, shape () 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
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Adam

torch.manual_seed(0)

class SimpleGCNet(nn.Module):
    def __init__(self, input_dim=4, hidden_dim=8, output_dim=3):
        super().__init__()
        self.linear1 = nn.Linear(input_dim, hidden_dim)
        self.linear2 = nn.Linear(hidden_dim, output_dim)
    
    def forward(self, data):
        x = data.x
        x = F.relu(self.linear1(x))
        output = self.linear2(x)
        return output

class MockData:
    def __init__(self, x, y, num_graphs=1):
        self.x = x
        self.y = y
        self.num_graphs = num_graphs
    
    def to(self, device):
        self.x = self.x.to(device)
        self.y = self.y.to(device)
        return self

class MockDataLoader:
    def __init__(self, data_list):
        self.data_list = data_list
        self.dataset = data_list
    
    def __iter__(self):
        return iter(self.data_list)
    
    def __len__(self):
        return len(self.data_list)

# Create model
model = SimpleGCNet(input_dim=4, hidden_dim=8, output_dim=3)

# Create mock data samples
data1 = MockData(
    x=torch.randn(2, 4),
    y=torch.tensor([0, 1], dtype=torch.long),
    num_graphs=2
)

data2 = MockData(
    x=torch.randn(2, 4),
    y=torch.tensor([2, 0], dtype=torch.long),
    num_graphs=2
)

# Create loaders
loader = MockDataLoader([data1, data2])
adv_loader = MockDataLoader([data1, data2])

ARGS = {
    "model": model,
    "loader": loader,
    "robust": False,
    "adv_loader": adv_loader,
    "lamb": 0.1
}
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 float, ndim 0, dtype family float, shape () — shape () 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
RobustGCNModel Not yet run
unverified (harvested, no recorded run)
danielzuegner/robust-gcn/robust_gcn/robust_gcn.py
code served (permissive licence) · get_code("a3af207d6d21af5d")
How it ran

no invocation recorded.

attack_global Not yet run
unverified (harvested, no recorded run)
RobustGraph/RoboGraph/robograph/utils.py
code served (permissive licence) · get_code("98474b68028dbdbe")
How it ran

no invocation recorded.

dp_solver Not yet run
unverified (harvested, no recorded run)
RobustGraph/RoboGraph/robograph/attack/dp_old_version.py
code served (permissive licence) · get_code("714d5d63e4a9394b")
How it ran

no invocation recorded.

forward_elemental_rotation_cert Not yet run
unverified (harvested, no recorded run)
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/certificates/rotation_elemental.py
code served (permissive licence) · get_code("f60179fa8fb21d2b")
How it ran

no invocation recorded.

inverse_elemental_rotation_cert Not yet run
unverified (harvested, no recorded run)
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/certificates/rotation_elemental.py
code served (permissive licence) · get_code("015c7f55ac954727")
How it ran

no invocation recorded.

local_solver_linear_term Not yet run
unverified (harvested, no recorded run)
RobustGraph/RoboGraph/robograph/attack/dp_old_version.py
code served (permissive licence) · get_code("5ad2414d2eb09c07")
How it ran

no invocation recorded.

sample_votes_datapoint Not yet run
unverified (harvested, no recorded run)
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/smoothing.py
code served (permissive licence) · get_code("97b718aa6a264cde")
How it ran

no invocation recorded.

sample_votes_dataset Not yet run
unverified (harvested, no recorded run)
jan-schuchardt/invariance-smoothing/pointcloud_invariance_smoothing/smoothing.py
code served (permissive licence) · get_code("c42e1cd047ffdb2c")
How it ran

no invocation recorded.

wrap_prediction_model Not yet run
unverified (harvested, no recorded run)
wollschl/uncertainty_for_molecules/ue4mol/models/utils.py
pointer only (licence: NONE) · get_code("7510b92771842954")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, wollschl/uncertainty_for_molecules 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. No invocation recorded.

Also present in other papers' code

Of 31 samples lifted from this paper's linked repositories, 31 carry a function identity and 0 do not. Of the 31 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 29 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
preprocess_adj
brick of this paper's assembly: data_transform
1 other paper
1 dated earlier
1906.12269 (2019-06)
Certifiable Robustness and Robust Training for Graph Convolutional Networks
sparse_tensor
brick of this paper's assembly: helper
1 other paper
1 dated earlier
1906.12269 (2019-06)
Certifiable Robustness and Robust Training for Graph Convolutional Networks

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

A machine learning model is traditionally considered robust if its prediction remains (almost) constant under input perturbations with small norm. However, real-world tasks like molecular property prediction or point cloud segmentation have inherent equivariances, such as rotation or permutation equivariance. In such tasks, even perturbations with large norm do not necessarily change an input's semantic content. Furthermore, there are perturbations for which a model's prediction explicitly needs to change. For the first time, we propose a sound notion of adversarial robustness that accounts for task equivariance. We then demonstrate that provable robustness can be achieved by (1) choosing a model that matches the task's equivariances (2) certifying traditional adversarial robustness. Certification methods are, however, unavailable for many models, such as those with continuous equivariances. We close this gap by developing the framework of equivariance-preserving randomized smoothing, which enables architecture-agnostic certification. We additionally derive the first architecture-specific graph edit distance certificates, i.e. sound robustness guarantees for isomorphism equivariant tasks like node classification. Overall, a sound notion of robustness is an important prerequisite for future work at the intersection of robust and geometric machine learning. decreases robustness to ℓ p perturbations and vice-versa [21][22][23][24][25]. Schuchardt and Günnemann [26] used knowledge about the invariances of point cloud classifiers to prove that they are constant within larger regions than could be shown using previous approaches. Group invariant distances. Recently, stability results for graph classifiers under isomorphism invariant optimal transport distances have been derived [27][28][29]. For point cloud classifiers, using the permutation invariant Chamfer or Hausdorff distance to craft attacks has been proposed [30,31]. These works only focus on invariance and specific domains and do not consider that distances should be task-dependent: A rotation invariant distance for images may be desirable when segmenting cell nuclei, but not when classifying hand-written digits, since it would fail to distinguish 6 and 9. String edit distance. In concurrent work, Huang et al. [32] use randomized smoothing to prove robustness of classifiers w.r.t. string edit distance, i.e., the number of substitutions that are needed to convert one string from alphabet Σ ∪ {⊥} into another, up to insertion of alignment tokens ⊥. Their work further emphasizes the need for invariant distance functions in domains with symmetries, and the usefulness of randomized smoothing for proving robustness w.r.t. such distances. Robustness of models with equivariances. Aside from work that studies invariance and adversarial robustness jointly, there is a rich literature investigating the robustness of models that happen to have equivariances. This includes convolutions [5][6][7]33], transformers [34][35][36][37], point cloud models [30,31,[38][39][40][41][42][43][44][45][46] and graph neural networks [8][9][10][11][12][13][14][15][16][17][47][48][49][50][51][52][53][54][55]. The models are however treated as a series of matrix multiplications and nonlinearities, without accounting for their equivariances or the equivariances of the tasks they are used for. Nevertheless, many methods can actually be reused for proving (non-)robustness under our proposed notion of adversarial robustness (see Section 5). Transformation-specific robustness. A subfield of robust machine learning focuses on robustness to unnoticeable parametric transformations (e.g. small rotations) [42,46,[56][57][58][59][60][61][62][63][64][65]. These works implicitly assume that large transformations lead to easily identifiable out-of-distribution samples. This is not the case with equivariant tasks: For instance, a molecule rotated by 180 • is still the same geometric object. Furthermore, they do not consider unstructured perturbations. Nevertheless, transformation-specific robustness can be framed as a special case of our proposed notion (see Appendix J). Semantics-aware robustness. Our work is closely related to different proposals to include ground truth labels in the definition of adversarial robustness [22,52,[66][67][68][69][70][71]. A problem is that the ground truth is usually unknown, which limits experimental evaluation to simple data generating distributions [52,70] or using human study participants [22,67,71]. Geisler et al. [72] overcome this problem in the context of neural combinatorial optimization by using adversarial perturbations that are known to change the ground truth of a decision problem. Group equivariant tasks admit a similar approach, since we know how the ground truth changes for specific input transformations.

For agents

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

get_harvested_code_for_paper("2312.02708")
get_code_for_paper("2312.02708")
have("2312.02708")

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

curl https://syntology.ai/api/ran/2312.02708.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.02708.svg)](https://syntology.ai/paper/2312.02708)

Connect an agent — have() is free.