SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.08595 · ICML · 2024

Homomorphism Counts for Graph Neural Networks: All About That Basis

Michael Bronstein, İsmail Ceylan, Emily Jin, Matthias Lanzinger

arXiv · PDF · Open in the Atlas

Code that ran

8 of 9 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; 5 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 1 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/2402.08595.json · badge: /api/ran/2402.08595.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
ejin700/hombasis-gnn canonical (classifier label)
classifier basis: classifier_2025
4 ran · 1 no recorded run
icml2024357/hombasis-gnn canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
choose_activation Ran
ran; our contract draft was wrong, not the code
icml2024357/hombasis-gnn/brec/HomGIN/models_misc.py
code served (permissive licence) · get_code("8cece8d22b52bc54")
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 ReLU. 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 = {
    'activation': 'relu'
}
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 ReLU — a ReLU output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ReLU 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
eval Ran
ran, honoured the contract we drafted
icml2024357/hombasis-gnn/hombasis-bench/zinc_utils.py
code served (permissive licence) · get_code("1abf815848e61eff")
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 numpy as np

torch.manual_seed(0)

# Define a minimal batch class with required attributes
class Batch:
    def __init__(self, x, edge_index, counts, batch, y, dimension=1):
        self.x = x
        self.edge_index = edge_index
        self.counts = counts
        self.batch = batch
        self.y = y
        self.dimension = dimension
        self.cochains = {}
        for dim in range(dimension + 1):
            self.cochains[dim] = type('obj', (object,), {'x': torch.randn(4, 3)})()
    
    def to(self, device):
        self.x = self.x.to(device)
        self.edge_index = self.edge_index.to(device)
        self.counts = self.counts.to(device)
        self.batch = self.batch.to(device)
        self.y = self.y.to(device)
        for dim in range(self.dimension + 1):
            self.cochains[dim].x = self.cochains[dim].x.to(device)
        return self

# Define a minimal model
class MinimalGraphReg(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(3, 1)
    
    def forward(self, x, edge_index, counts, use_counts, batch):
        return self.linear(x).squeeze(-1)
    
    def eval(self):
        super().eval()
        return self

# Create test data
num_nodes = 4
num_edges = 5
batch_size = 2

x = torch.randn(num_nodes, 3)
edge_index = torch.randint(0, num_nodes, (2, num_edges))
counts = torch.ones(num_nodes, 2)
batch_indices = torch.tensor([0, 0, 1, 1])
y = torch.randn(num_nodes, 1)

batch1 = Batch(x=x.clone(), edge_index=edge_index.clone(), counts=counts.clone(), 
               batch=batch_indices.clone(), y=y.clone())
batch2 = Batch(x=x.clone(), edge_index=edge_index.clone(), counts=counts.clone(), 
               batch=batch_indices.clone(), y=y.clone())

loader = [batch1, batch2]

model = MinimalGraphReg()
device = torch.device('cpu')

ARGS = {
    'model_name': 'GINGraphReg',
    'use_counts': True,
    'model': model,
    'device': device,
    'loader': loader
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, torch — 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
graph6_to_pyg Ran
ran; our contract draft was wrong, not the code
icml2024357/hombasis-gnn/brec/ProvablyPowerfulGraphNetworks_torch/BRECDataset_v3.py
code served (permissive licence) · get_code("513e5ba0da199b50")
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 Data. 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 networkx as nx
from torch_geometric.utils.convert import from_networkx, to_networkx

torch.manual_seed(0)

# Create a simple graph in graph6 format
# graph6 format is a compact representation of graphs
# We'll create a small graph and encode it to bytes
simple_graph = nx.Graph()
simple_graph.add_edges_from([(0, 1), (1, 2), (2, 0)])

# Convert to graph6 bytes format
graph6_bytes = nx.to_graph6_bytes(simple_graph, header=False)

ARGS = {
    "x": graph6_bytes
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx, torch, torch_geometric — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Data — a Data 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
test Ran
ran; our contract draft was wrong, not the code
ejin700/hombasis-gnn/hombasis-bench/run-collab.py
code served (permissive licence) · get_code("a6e419d6bf287a8f")
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 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 torch
import torch.nn as nn

torch.manual_seed(0)

# Define a minimal model that satisfies the function's usage
class MinimalModel(nn.Module):
    def __init__(self, input_dim, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, hidden_dim)
    
    def forward(self, x, adj_t, use_counts, counts):
        # Returns node embeddings of shape [num_nodes, hidden_dim]
        return self.linear(x)
    
    def eval(self):
        super().eval()
        return self

# Define a minimal predictor that takes two node embeddings and returns a score
class MinimalPredictor(nn.Module):
    def __init__(self, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(hidden_dim * 2, 1)
    
    def forward(self, h1, h2):
        # Concatenate embeddings and predict link score
        combined = torch.cat([h1, h2], dim=-1)
        return self.linear(combined)
    
    def eval(self):
        super().eval()
        return self

# Define a minimal evaluator
class MinimalEvaluator:
    def __init__(self):
        self.K = 10
    
    def eval(self, input_dict):
        # Return hits@K metric
        y_pred_pos = input_dict['y_pred_pos']
        y_pred_neg = input_dict['y_pred_neg']
        # Simple mock evaluation: compute hits based on ranking
        hits = (y_pred_pos.mean() > y_pred_neg.mean()).float().item()
        return {f'hits@{self.K}': hits}

# Define a minimal data object
class MinimalData:
    def __init__(self):
        num_nodes = 8
        input_dim = 4
        self.x = torch.randn(num_nodes, input_dim)
        self.adj_t = torch.eye(num_nodes)
        self.full_adj_t = torch.eye(num_nodes)
        self.use_counts = torch.ones(num_nodes)
        self.counts = torch.ones(num_nodes)

# Create instances
num_nodes = 8
input_dim = 4
hidden_dim = 3
batch_size = 2

model = MinimalModel(input_dim, hidden_dim)
predictor = MinimalPredictor(hidden_dim)
data = MinimalData()
evaluator = MinimalEvaluator()

# Create split_edge dictionary with edge tensors
num_train_edges = 4
num_valid_edges = 2
num_test_edges = 2

split_edge = {
    'train': {
        'edge': torch.randint(0, num_nodes, (num_train_edges, 2))
    },
    'valid': {
        'edge': torch.randint(0, num_nodes, (num_valid_edges, 2)),
        'edge_neg': torch.randint(0, num_nodes, (num_valid_edges, 2))
    },
    'test': {
        'edge': torch.randint(0, num_nodes, (num_test_edges, 2)),
        'edge_neg': torch.randint(0, num_nodes, (num_test_edges, 2))
    }
}

ARGS = {
    'model': model,
    'predictor': predictor,
    'data': data,
    'split_edge': split_edge,
    'evaluator': evaluator,
    'batch_size': batch_size
}
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 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
test Ran
ran, honoured the contract we drafted
ejin700/hombasis-gnn/qm9/src/experiments/run_gc.py
code served (permissive licence) · get_code("5bfd4028374d07c7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned float, ndim 0, dtype family float, shape () 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

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(5, 3)
    
    def forward(self, data):
        # data is expected to have some feature attribute
        # Return shape (batch_size, num_classes)
        x = data.x if hasattr(data, 'x') else torch.randn(data.y.size(0), 5)
        return self.fc(x)

class DataObject:
    def __init__(self, x, y):
        self.x = x
        self.y = y
    
    def to(self, device):
        self.x = self.x.to(device)
        self.y = self.y.to(device)
        return self

# Create model
model = SimpleModel()

# Create loader - list of data objects
data1 = DataObject(torch.randn(2, 5), torch.tensor([0, 1]))
data2 = DataObject(torch.randn(2, 5), torch.tensor([2, 0]))
data3 = DataObject(torch.randn(2, 5), torch.tensor([1, 2]))

loader = [data1, data2, data3]

# Add dataset attribute to loader (needed for len(loader.dataset))
class LoaderWithDataset:
    def __init__(self, data_list):
        self.data_list = data_list
        self.dataset = list(range(sum(d.y.size(0) for d in data_list)))
    
    def __iter__(self):
        return iter(self.data_list)

loader = LoaderWithDataset([data1, data2, data3])

ARGS = {
    'model': model,
    'loader': loader
}
Setup block drafted by
a language model, Claude Sonnet 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 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
train Ran
ran, honoured the contract we drafted
ejin700/hombasis-gnn/hombasis-bench/run-collab.py
code served (permissive licence) · get_code("b10c2f48c45238a5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) 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.utils.data import DataLoader

torch.manual_seed(0)

# Model that takes x, adj_t, use_counts, counts and returns embeddings
class StandInModel(nn.Module):
    def __init__(self, in_dim, out_dim):
        super().__init__()
        self.lin = nn.Linear(in_dim, out_dim)
    
    def forward(self, x, adj_t, use_counts, counts):
        return self.lin(x)

# Predictor that takes two node embeddings and returns a score
class StandInPredictor(nn.Module):
    def __init__(self, emb_dim):
        super().__init__()
        self.lin = nn.Linear(emb_dim * 2, 1)
    
    def forward(self, h1, h2):
        combined = torch.cat([h1, h2], dim=-1)
        return torch.sigmoid(self.lin(combined)).squeeze(-1)

# Create model and predictor
model = StandInModel(in_dim=5, out_dim=4)
predictor = StandInPredictor(emb_dim=4)

# Create data object with required attributes
class StandInData:
    def __init__(self):
        self.num_nodes = 6
        self.x = torch.randn(6, 5)  # 6 nodes, 5 features
        self.adj_t = torch.randn(6, 6)  # adjacency
        self.use_counts = torch.ones(6)
        self.counts = torch.ones(6)

data = StandInData()

# Create split_edge with train edges
split_edge = {
    'train': {
        'edge': torch.tensor([[0, 1], [1, 2], [2, 3], [3, 4]], dtype=torch.long)  # 4 edges
    }
}

# Create optimizer
optimizer = torch.optim.Adam(list(model.parameters()) + list(predictor.parameters()), lr=0.01)

batch_size = 2

ARGS = {
    'model': model,
    'predictor': predictor,
    'data': data,
    'split_edge': split_edge,
    'optimizer': optimizer,
    'batch_size': batch_size
}
Setup block drafted by
a language model, Claude Sonnet 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 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
train Ran
ran, honoured the contract we drafted
icml2024357/hombasis-gnn/hombasis-bench/zinc_utils.py
code served (permissive licence) · get_code("b367358dc74e5496")
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 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
from torch_geometric.data import Data, DataLoader

torch.manual_seed(0)

# Create a minimal GNN model that matches the function's usage
class MinimalGINGraphReg(nn.Module):
    def __init__(self):
        super().__init__()
        self.lin = nn.Linear(4, 1)
    
    def forward(self, x, edge_index, counts, use_counts, batch):
        # The function expects pred to have same shape as targets
        # targets has shape batch.y reshaped to pred.shape
        # So we need to return a tensor with batch size = number of graphs
        out = self.lin(x)  # [num_nodes, 1]
        # Simple global pooling: sum over nodes for each graph
        num_graphs = batch.max().item() + 1
        graph_out = torch.zeros(num_graphs, 1)
        for i in range(num_graphs):
            graph_out[i] = out[batch == i].sum()
        return graph_out
    
    def train(self):
        super().train()

# Create a minimal data batch
def create_batch():
    # 2 graphs
    x = torch.randn(4, 4)  # 4 nodes, 4 features
    edge_index = torch.tensor([[0, 1, 2, 3], [1, 2, 3, 0]], dtype=torch.long)
    counts = torch.tensor([1.0, 1.0, 1.0, 1.0])
    batch = torch.tensor([0, 0, 1, 1])  # 2 nodes in each of 2 graphs
    y = torch.randn(2, 1)  # 2 graphs, 1 target each
    
    data = Data(x=x, edge_index=edge_index, counts=counts, batch=batch, y=y)
    return data

# Create loader with batches
loader = [create_batch(), create_batch()]

# Create model and optimizer
model = MinimalGINGraphReg()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

ARGS = {
    "model_name": "GINGraphReg",
    "use_counts": True,
    "model": model,
    "device": torch.device("cpu"),
    "loader": loader,
    "optimizer": optimizer,
}
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 list, ndim 1, 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
val Ran
ran, honoured the contract we drafted
ejin700/hombasis-gnn/qm9/src/experiments/run_gc.py
code served (permissive licence) · get_code("49732bab8713a245")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned float, ndim 0, dtype family float, shape () 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

torch.manual_seed(0)

# Create a simple model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(5, 3)
    
    def forward(self, data):
        # Model expects data object with x attribute
        return self.linear(data.x)

model = SimpleModel()

# Create a stand-in data class that mimics PyTorch Geometric Data
class Data:
    def __init__(self, x, y, num_graphs):
        self.x = x
        self.y = y
        self.num_graphs = num_graphs
    
    def to(self, device):
        # Return self since we're staying on CPU
        return self

# Create a simple dataset/loader
# loader should be iterable yielding Data objects
data1 = Data(x=torch.randn(6, 5), y=torch.randn(6, 3), num_graphs=2)
data2 = Data(x=torch.randn(4, 5), y=torch.randn(4, 3), num_graphs=2)

class SimpleLoader:
    def __init__(self, data_list):
        self.data_list = data_list
        self.dataset = self  # loader.dataset should exist
    
    def __iter__(self):
        return iter(self.data_list)
    
    def __len__(self):
        # This should return the total number of graphs in the dataset
        return sum(d.num_graphs for d in self.data_list)

loader = SimpleLoader([data1, data2])

# Create a loss function
loss_fun = nn.MSELoss()

ARGS = {
    'model': model,
    'loader': loader,
    'loss_fun': loss_fun
}
Setup block drafted by
a language model, Claude Sonnet 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 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
train Not yet run
unverified (harvested, no recorded run)
ejin700/hombasis-gnn/qm9/src/experiments/run_gc.py
code served (permissive licence) · get_code("d1a3f22e2fe768ac")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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, 2 later) and 7 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
choose_activation 3 other papers
2 dated earlier, 1 later
2107.01952 (2021-07)
Partition and Code: learning how to compress graphs
2304.07702 (2023-04)
An Empirical Study of Realized GNN Expressiveness
2511.08028 (2025-11)
Generalizable Insights for Graph Transformers in Theory and Practice
graph6_to_pyg 2 other papers
1 dated earlier, 1 later
2304.07702 (2023-04)
An Empirical Study of Realized GNN Expressiveness
2511.08028 (2025-11)
Generalizable Insights for Graph Transformers in Theory and Practice

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 large body of work has investigated the properties of graph neural networks and identified several limitations, particularly pertaining to their expressive power. Their inability to count certain patterns (e.g., cycles) in a graph lies at the heart of such limitations, since many functions to be learned rely on the ability of counting such patterns. Two prominent paradigms aim to address this limitation by enriching the graph features with subgraph or homomorphism pattern counts. In this work, we show that both of these approaches are sub-optimal in a certain sense and argue for a more fine-grained approach, which incorporates the homomorphism counts of all structures in the "basis" of the target pattern. This yields strictly more expressive architectures without incurring any additional overhead in terms of computational complexity compared to existing approaches. We prove a series of theoretical results on node-level and graph-level motif parameters and empirically validate them on standard benchmark datasets.

For agents

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

get_harvested_code_for_paper("2402.08595")
get_code_for_paper("2402.08595")
have("2402.08595")

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

curl https://syntology.ai/api/ran/2402.08595.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.08595.svg)](https://syntology.ai/paper/2402.08595)

Connect an agent — have() is free.