SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.13009 · NeurIPS · 2020

Iterative Deep Graph Learning for Graph Neural Networks: Better and Robust Node Embeddings

Yu Chen, Lingfei Wu, Mohammed Zaki

arXiv · PDF · Open in the Atlas

Code that ran

8 of 14 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 1 violated it and 4 have no contract verdict; 2 failures of our instrument; 6 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/2006.13009.json · badge: /api/ran/2006.13009.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
hugochan/IDGL canonical (classifier label)
classifier basis: census
8 ran · 6 no recorded run
FunctionStatusWhere it lives
SquaredFrobeniusNorm Ran
ran; our contract draft was wrong, not the code
hugochan/IDGL/src/core/model_handler.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from WtaoZhao/GraphGLOW, is pointer only (licence: NONE) · get_code("4e393b15daa9461c")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from WtaoZhao/GraphGLOW, 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.

accuracy Ran
ran, honoured the contract we drafted
hugochan/IDGL/src/core/model.py
code served (permissive licence) · get_code("4cf9dcdd60711db4")
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 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

torch.manual_seed(0)

# The function expects:
# - labels: a tensor of class indices
# - output: a tensor of logits/scores where max(1)[1] gets predicted class

# Create small tensors
batch_size = 2
num_classes = 4

# labels: tensor of true class indices (shape: [batch_size])
labels = torch.tensor([0, 2], dtype=torch.long)

# output: tensor of logits (shape: [batch_size, num_classes])
# max(1)[1] will extract the class with highest score for each sample
output = torch.randn(batch_size, num_classes)

ARGS = {
    "labels": labels,
    "output": output,
}
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 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
compute_normalized_laplacian Ran
ran, violated the contract we drafted
hugochan/IDGL/src/core/layers/graphlearn.py
code served (permissive licence) · get_code("698d3807a0c41ec9")
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 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
# The function expects a 2D tensor (matrix)
# We'll create a 4x4 adjacency matrix for a small graph
adj = torch.tensor([
    [0.0, 1.0, 1.0, 0.0],
    [1.0, 0.0, 1.0, 1.0],
    [1.0, 1.0, 0.0, 1.0],
    [0.0, 1.0, 1.0, 0.0]
], dtype=torch.float32)

ARGS = {
    "adj": adj
}
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 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
dropout Ran
ran on a synthesized input
hugochan/IDGL/src/core/layers/common.py
code served (permissive licence) · get_code("90744bd46d234211")
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
import torch.nn as nn

torch.manual_seed(0)

# Create a small input tensor
# Shape: (batch_size=2, embedding_dim=4)
x = torch.randn(2, 4)

# Drop probability
drop_prob = 0.5

# Shared axes - empty list means no axes are shared
shared_axes = []

# Training mode flag
training = True

ARGS = {
    "x": x,
    "drop_prob": drop_prob,
    "shared_axes": shared_axes,
    "training": training
}
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
normalize_text Ran
ran; our contract draft was wrong, not the code
hugochan/IDGL/src/core/utils/eval_utils.py
code served (permissive licence) · get_code("38ffdcfda0ea6889")
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 str. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:55:33Z. 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 re
import string

def normalize_text(s):
    """Lower text and remove punctuation, articles and extra whitespace."""

    def remove_articles(text):
        return re.sub('\\b(a|an|the)\\b', ' ', text)

    def white_space_fix(text):
        return ' '.join(text.split())

    def remove_punc(text):
        exclude = set(string.punctuation)
        return ''.join((ch for ch in text if ch not in exclude))

    def lower(text):
        return text.lower()
    return white_space_fix(remove_articles(remove_punc(lower(s))))

ARGS = {
    "s": "The Quick Brown Fox, jumps over a lazy dog!"
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); 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_anchors Ran
ran on a synthesized input
hugochan/IDGL/src/core/layers/anchor.py
code served (permissive licence) · get_code("784e90120e8129cf")
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. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:55:33Z. 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

torch.manual_seed(0)

# Create a small node_vec tensor (batch size 2, embedding dim 4)
node_vec = torch.randn(2, 4)

# Set s to be smaller than the number of nodes
s = 1

ARGS = {
    "node_vec": node_vec,
    "s": s,
}
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 tuple — a tuple output has no array shape of its own; none is stated
Environment
image syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); 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_batch Ran
ran on a synthesized input
hugochan/IDGL/src/core/model.py
code served (permissive licence) · get_code("11a56e75a1cddbe8")
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 torch.nn as nn
from unittest.mock import Mock

torch.manual_seed(0)

# Mock network output
class NetworkOutput:
    def __init__(self, device='cpu'):
        self.loss = torch.tensor(0.5, device=device, requires_grad=True)
        self.loss_value = torch.tensor(0.5, device=device)
        self.decoded_tokens = torch.randint(0, 100, (5, 2), device=device)  # seq_len x batch_size

# Mock OOV dict
class OOVDict:
    def __init__(self):
        self.ext_vocab_size = 150

# Minimal network module
class MockNetwork(nn.Module):
    def __init__(self):
        super().__init__()
        self.device = 'cpu'
        self.linear = nn.Linear(4, 4)
    
    def train(self, mode=True):
        super().train(mode)
    
    def forward(self, batch, targets=None, criterion=None, forcing_ratio=0.5,
                partial_forcing=False, sample=False, ext_vocab_size=None,
                include_cover_loss=False, saved_out=None, criterion_reduction=True,
                criterion_nll_only=False, visualize=True):
        return NetworkOutput(self.device)

# Mock criterion
class MockCriterion:
    def __call__(self, *args, **kwargs):
        return torch.tensor(0.5, requires_grad=True)

# Mock vocab
vocab = {'<unk>': 0, '<pad>': 1, 'hello': 2, 'world': 3}

# Mock batch
batch = {
    'context': torch.randn(2, 4),  # batch_size=2, dim=4
    'targets': torch.randint(0, 100, (5, 2)),  # seq_len=5, batch_size=2
    'oov_dict': OOVDict(),
    'target_src': [['hello', 'world'], ['foo', 'bar']],
}

# Mock functions needed by the function body
def eval_batch_output(target_src, vocab, oov_dict, decoded_tokens, baseline=None):
    """Mock eval function returning metrics"""
    metrics = {
        'rouge': 0.5,
        'bleu': 0.6,
    }
    if baseline is None:
        return [metrics]
    else:
        return [metrics, metrics]

def batch_decoded_index2word(decoded_indices, vocab, oov_dict):
    """Mock function to convert indices to words"""
    return [['hello', 'world'] for _ in decoded_indices]

def to_cuda(tensor, device):
    """Mock function"""
    return tensor

# Inject mocks into the module namespace
import sys
current_module = sys.modules[__name__]
current_module.eval_batch_output = eval_batch_output
current_module.batch_decoded_index2word = batch_decoded_index2word
current_module.to_cuda = to_cuda

# Config dict
config = {
    'partial_forcing': False,
    'sample': False,
    'show_cover_loss': False,
    'rl_reward_metric': 'rouge',
    'rl_wmd_ratio': 0.0,
    'max_wmd_reward': 1.0,
}

# Mock WMD (word mover distance)
wmd_mock = Mock()
wmd_mock.distance = Mock(return_value=0.5)

ARGS = {
    'batch': batch,
    'network': MockNetwork(),
    'vocab': vocab,
    'criterion': MockCriterion(),
    'forcing_ratio': 0.5,
    'rl_ratio': 0.0,
    'config': config,
    'wmd': wmd_mock,
}
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 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
vectorize_input Ran
ran on a synthesized input
hugochan/IDGL/src/core/utils/data_utils.py
code served (permissive licence) · get_code("81d592687b5f5dd5")
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 torch

torch.manual_seed(0)

class Batch:
    def __init__(self, sent1_word, sent1_length, labels, has_sent2=False, sent2_word=None, sent2_length=None):
        self.sent1_word = sent1_word
        self.sent1_length = sent1_length
        self.labels = labels
        self.has_sent2 = has_sent2
        self.sent2_word = sent2_word
        self.sent2_length = sent2_length

batch = Batch(
    sent1_word=[[1, 2, 3, 4], [5, 6, 7, 8]],
    sent1_length=[4, 4],
    labels=[0.5, 1.5],
    has_sent2=True,
    sent2_word=[[9, 10, 11, 12], [13, 14, 15, 16]],
    sent2_length=[4, 4]
)

config = {
    'task_type': 'regression'
}

ARGS = {
    'batch': batch,
    'config': config,
    'training': True,
    'device': None
}
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 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
batch_diff Not yet run
unverified (harvested, no recorded run)
hugochan/IDGL/src/core/model_handler.py
code served (permissive licence) · get_code("e264fe290089eb78")
How it ran

no invocation recorded.

batch_sample_anchors Not yet run
unverified (harvested, no recorded run)
hugochan/IDGL/src/core/layers/anchor.py
code served (permissive licence) · get_code("844daf63f8277a07")
How it ran

no invocation recorded.

batch_select_from_tensor Not yet run
unverified (harvested, no recorded run)
hugochan/IDGL/src/core/layers/anchor.py
code served (permissive licence) · get_code("837b6503909c3bcc")
How it ran

no invocation recorded.

diff Not yet run
unverified (harvested, no recorded run)
hugochan/IDGL/src/core/model_handler.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from WtaoZhao/GraphGLOW, is pointer only (licence: NONE) · get_code("29922ecc7c935e89")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from WtaoZhao/GraphGLOW, 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. No invocation recorded.

get_binarized_kneighbors_graph Not yet run
unverified (harvested, no recorded run)
hugochan/IDGL/src/core/layers/graphlearn.py
code served (permissive licence) · get_code("80476f95d3822c09")
How it ran

no invocation recorded.

prepare_datasets Not yet run
unverified (harvested, no recorded run)
hugochan/IDGL/src/core/utils/data_utils.py
code served (permissive licence) · get_code("7533b3dc9f5bbfd6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 6 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 5 later) and 8 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
compute_normalized_laplacian 3 other papers
3 later
2104.09036 (2021-04)
Mining Latent Structures for Multimedia Recommendation
2311.00423 (2023-11)
LLMRec: Large Language Models with Graph Augmentation for Recommendation
2402.17188 (2024-02)
PromptMM: Multi-Modal Knowledge Distillation for Recommendation with Prompt-Tuning
normalize_text 3 other papers
1 dated earlier, 2 later
1909.03329 (2019-09)
LAMOL: LAnguage MOdeling for Lifelong Language Learning
2203.10652 (2022-03)
Continual Sequence Generation with Adaptive Compositional Modules
2025.emnlp-industry.89 (2025 (year only))
L4: Mutual Learning Helps Lifelong Language Learning
accuracy 2 other papers
2 later
2208.08302 (2022-08)
Position-aware Structure Learning for Graph Topology-imbalance by Relieving Under-reaching and Over-squashing
2502.19670 (2025-02)
Training Robust Graph Neural Networks by Modeling Noise Dependencies
dropout 2 other papers
2 dated earlier
1908.04942 (2019-08)
Reinforcement Learning Based Graph-to-Sequence Model for Natural Question Generation
2004.06015 (2020-04)
Toward Subgraph-Guided Knowledge Graph Question Generation with Graph Neural Networks
SquaredFrobeniusNorm 1 other paper
1 later
2502.19670 (2025-02)
Training Robust Graph Neural Networks by Modeling Noise Dependencies
diff 1 other paper
1 later
2502.19670 (2025-02)
Training Robust Graph Neural Networks by Modeling Noise Dependencies

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

In this paper, we propose an end-to-end graph learning framework, namely Iterative Deep Graph Learning (IDGL), for jointly and iteratively learning graph structure and graph embedding. The key rationale of IDGL is to learn a better graph structure based on better node embeddings, and vice versa (i.e., better node embeddings based on a better graph structure). Our iterative method dynamically stops when the learned graph structure approaches close enough to the graph optimized for the downstream prediction task. In addition, we cast the graph learning problem as a similarity metric learning problem and leverage adaptive graph regularization for controlling the quality of the learned graph. Finally, combining the anchorbased approximation technique, we further propose a scalable version of IDGL, namely IDGL-ANCH, which significantly reduces the time and space complexity of IDGL without compromising the performance. Our extensive experiments on nine benchmarks show that our proposed IDGL models can consistently outperform or match the state-of-the-art baselines. Furthermore, IDGL can be more robust to adversarial graphs and cope with both transductive and inductive learning. Recent years have seen a significantly growing amount of interest in graph neural networks (GNNs), especially on efforts devoted to developing more effective GNNs for node classification [29,36,17,52], graph classification [60,43] and graph generation [47,37,61]. Despite GNNs' powerful ability in learning expressive node embeddings, unfortunately, they can only be used when graph-structured data is available. Many real-world applications naturally admit network-structured data (e.g., social networks). However, these intrinsic graph-structures are not always optimal for the downstream tasks. This is partially because the raw graphs were constructed from the original feature space, which may not reflect the "true" graph topology after feature extraction and transformation. Another potential reason is that real-world graphs are often noisy or even incomplete due to the inevitably error-prone data measurement or collection. Furthermore, many applications such as those in natural language processing [7,57,58] may only have sequential data or even just the original feature matrix, requiring additional graph construction from the original data matrix. To address these limitations, we propose an end-to-end graph learning framework, namely Iterative Deep Graph Learning (IDGL), for jointly and iteratively learning the graph structure and the GNN parameters that are optimized toward the downstream prediction task. The key rationale of our IDGL ˚Corresponding author. 34th Conference on Neural Information Processing Systems (NeurIPS 2020),

For agents

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

get_harvested_code_for_paper("2006.13009")
get_code_for_paper("2006.13009")
have("2006.13009")

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

curl https://syntology.ai/api/ran/2006.13009.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.13009.svg)](https://syntology.ai/paper/2006.13009)

Connect an agent — have() is free.