SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.06574 · 2020

Dynamically Stable Infinite-Width Limits of Neural Classifiers

arXiv · PDF · Open in the Atlas

Code that ran

11 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; 0 honoured the contract we drafted, 0 violated it and 11 have no contract verdict; 0 failures of our instrument; 3 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/2006.06574.json · badge: /api/ran/2006.06574.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
deepmipt/research canonical (classifier label)
classifier basis: census
11 ran · 3 no recorded run
FunctionStatusWhere it lives
cached_default_loader Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/data_loaders.py
code served (permissive licence) · get_code("e34888c8043c86a7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Image. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import tempfile
import os
from PIL import Image
from torchvision import datasets

torch.manual_seed(0)

# Create a temporary image file to use as the path argument
temp_dir = tempfile.mkdtemp()
temp_image_path = os.path.join(temp_dir, "test_image.jpg")

# Create a minimal valid image file
img = Image.new('RGB', (4, 3), color='red')
img.save(temp_image_path)

ARGS = {
    'path': temp_image_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
defaultdict_to_dict Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/dict_utils.py
code served (permissive licence) · get_code("9f3d51b0bf30577a")
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.

from collections import defaultdict

# Create a nested defaultdict structure to test the function
def_a = defaultdict(dict)

# Add some simple values
def_a['key1'] = 'value1'
def_a['key2'] = 42

# Add a nested defaultdict
nested_def = defaultdict(dict)
nested_def['nested_key1'] = 'nested_value1'
nested_def['nested_key2'] = [1, 2, 3]

def_a['nested'] = nested_def

# Add another level of nesting
deeply_nested = defaultdict(dict)
deeply_nested['deep_key'] = {'a': 1, 'b': 2}
nested_def['deep_nested'] = deeply_nested

ARGS = {
    'def_a': def_a
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_activation Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/models.py
code served (permissive licence) · get_code("e46ab9c27af265c9")
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 type. 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)

# Test with different activation types
ARGS = {
    'activation': 'relu'
}
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 type — a type output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a type output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_loss_function Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/perform_epoch.py
code served (permissive licence) · get_code("8bdd1d87adf242ec")
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 CrossEntropyLoss. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# The function takes a single parameter: loss_name (a string)
# It returns different loss functions based on the loss_name value
# We need to provide a valid loss_name that doesn't raise ValueError

ARGS = {
    'loss_name': 'ce'
}
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 CrossEntropyLoss — a CrossEntropyLoss output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a CrossEntropyLoss output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_model_with_modified_width Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/train_and_eval.py
code served (permissive licence) · get_code("b2072ed40ef3798a")
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 SimpleModel. 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 copy import copy, deepcopy

torch.manual_seed(0)

# Define a minimal model class that accepts a width parameter
class SimpleModel(nn.Module):
    def __init__(self, width=64, input_dim=4):
        super().__init__()
        self.linear = nn.Linear(input_dim, width)
    
    def forward(self, x):
        return self.linear(x)

# Set up the arguments
ARGS = {
    'model_class': SimpleModel,
    'reference_model_kwargs': {
        'width': 64,
        'input_dim': 4
    },
    'width_arg_name': 'width',
    'width_factor': 2,
    'device': 'cpu'
}
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 SimpleModel — a SimpleModel output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleModel output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_normalization_layer Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/models.py
code served (permissive licence) · get_code("e9251fc769b7987f")
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 CustomBatchNorm1d. 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)

# Test case 1: batch normalization with dim=1
ARGS = {
    'width': 16,
    'normalization': 'batch',
    'dim': 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 CustomBatchNorm1d — a CustomBatchNorm1d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a CustomBatchNorm1d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_optimizer Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/train_and_eval.py
code served (permissive licence) · get_code("94244eb9cae14d21")
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 Adam. 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 torch.optim import Adam

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.input_layer = nn.Linear(4, 8)
        self.hidden_layers = nn.Sequential(
            nn.Linear(8, 16),
            nn.ReLU(),
            nn.Linear(16, 8)
        )
        self.output_layer = nn.Linear(8, 2)
    
    def forward(self, x):
        x = self.input_layer(x)
        x = self.hidden_layers(x)
        x = self.output_layer(x)
        return x

model = SimpleModel()

optimizer_class = Adam
optimizer_kwargs = {'lr': 0.001, 'weight_decay': 1e-5}

ARGS = {
    'optimizer_class': optimizer_class,
    'optimizer_kwargs': optimizer_kwargs,
    'model': model
}
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 Adam — a Adam output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Adam output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_optimizer_class_and_default_lr Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/fcnet_epoch_dependence.py
code served (permissive licence) · get_code("93fe7120dfafc350")
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 torch.optim as optim

torch.manual_seed(0)

# Define SGDMomentum class since it's referenced in the function
class SGDMomentum(optim.Optimizer):
    def __init__(self, params, lr=0.1, momentum=0.9):
        defaults = dict(lr=lr, momentum=momentum)
        super(SGDMomentum, self).__init__(params, defaults)
    
    def step(self, closure=None):
        pass

# Create a minimal model for testing
class TinyModel(nn.Module):
    def __init__(self):
        super(TinyModel, self).__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Test with different optimizer names
ARGS = {
    'optimizer_name': 'sgd'
}
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, 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
get_shape Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/data_loaders.py
code served (permissive licence) · get_code("ad72d8173724b09a")
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

torch.manual_seed(0)

# The function only takes a dataset_name string parameter
# and returns a tuple of (shape, num_classes) based on the dataset name
# No complex objects needed - just a string

ARGS = {
    'dataset_name': 'mnist'
}
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
make_conv_block Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/models.py
code served (permissive licence) · get_code("262911c2cac73ce0")
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

torch.manual_seed(0)

ARGS = {
    'input_shape': (3, 16, 16),
    'num_convs': 2,
    'num_kernels': [8, 16],
    'kernel_size': [3, 3],
    'padding': 1,
    'activation': 'relu',
    'final_pool': 'max',
    'bias': True,
    'use_bn': True,
    'bn_affine': True,
    'dropout_rate': 0.2,
}
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
perform_epoch Ran
ran on a synthesized input
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/perform_epoch.py
code served (permissive licence) · get_code("e12b94e5b4eb51af")
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

torch.manual_seed(0)

# Create a simple model that takes input and outputs logits
class SimpleModel(nn.Module):
    def __init__(self, input_dim, output_dim, loss_name='ce'):
        super().__init__()
        self.loss_name = loss_name
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        return self.linear(x)

# Create a data loader as a list of (X, y) tuples
def create_loader(batch_size=2, num_batches=3, input_dim=4, num_classes=3, loss_name='ce'):
    loader = []
    for _ in range(num_batches):
        X = torch.randn(batch_size, input_dim)
        if loss_name == 'bce':
            y = torch.randint(0, 2, (batch_size,))
        else:
            y = torch.randint(0, num_classes, (batch_size,))
        loader.append((X, y))
    return loader

# Set up the model for cross-entropy loss (default case)
model_ce = SimpleModel(input_dim=4, output_dim=3, loss_name='ce')
loader_ce = create_loader(batch_size=2, num_batches=3, input_dim=4, num_classes=3, loss_name='ce')
optimizer_ce = torch.optim.SGD(model_ce.parameters(), lr=0.01)

# Set up the model for binary cross-entropy loss
model_bce = SimpleModel(input_dim=4, output_dim=1, loss_name='bce')
loader_bce = create_loader(batch_size=2, num_batches=3, input_dim=4, num_classes=2, loss_name='bce')
optimizer_bce = torch.optim.SGD(model_bce.parameters(), lr=0.01)

ARGS = {
    'model': model_ce,
    'loader': loader_ce,
    'optimizer': optimizer_ce,
    'max_batch_count': None,
    'device': 'cpu',
    'loss_name': 'ce'
}
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, 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
dict_to_defaultdict Not yet run
unverified (harvested, no recorded run)
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/dict_utils.py
code served (permissive licence) · get_code("46d9a00d8f29502d")
How it ran

no invocation recorded.

get_loaders Not yet run
unverified (harvested, no recorded run)
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/data_loaders.py
code served (permissive licence) · get_code("a3acc2c4331bedae")
How it ran

no invocation recorded.

get_tangent_kernels Not yet run
unverified (harvested, no recorded run)
deepmipt/research/Infinite_Width_Limits_of_Neural_Classifiers/utils/perform_epoch.py
code served (permissive licence) · get_code("def95cf3f393cd93")
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, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 14 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.

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

Recent research has been focused on two different approaches to studying neural networks training in the limit of infinite width (1) a mean-field (MF) and (2) a constant neural tangent kernel (NTK) approximations. These two approaches have different scaling of hyperparameters with the width of a network layer and as a result, different infinite-width limit models. We propose a general framework to study how the limit behavior of neural models depends on the scaling of hyperparameters with network width. Our framework allows us to derive scaling for existing MF and NTK limits, as well as an uncountable number of other scalings that lead to a dynamically stable limit behavior of corresponding models. However, only a finite number of distinct limit models are induced by these scalings. Each distinct limit model corresponds to a unique combination of such properties as boundedness of logits and tangent kernels at initialization or stationarity of tangent kernels. Existing MF and NTK limit models, as well as one novel limit model, satisfy most of the properties demonstrated by finite-width models. We also propose a novel initialization-corrected mean-field limit that satisfies all properties noted above, and its corresponding model is a simple modification for a finite-width model.

For agents

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

get_harvested_code_for_paper("2006.06574")
get_code_for_paper("2006.06574")
have("2006.06574")

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

curl https://syntology.ai/api/ran/2006.06574.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.06574.svg)](https://syntology.ai/paper/2006.06574)

Connect an agent — have() is free.