SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1806.00451 · 2018

Do CIFAR-10 Classifiers Generalize to CIFAR-10?

arXiv · PDF · Open in the Atlas

Code that ran

7 of 11 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 5 have no contract verdict; 2 failures of our instrument; 4 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/1806.00451.json · badge: /api/ran/1806.00451.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
modestyachts/CIFAR-10.1 canonical (classifier label)
classifier basis: classifier_2025
1 ran · 2 no recorded run
aisosalo/CIFAR-10 community (classifier label)
classifier basis not recorded on the link
6 ran · 2 no recorded run
FunctionStatusWhere it lives
aslist Ran
ran on a synthesized input
aisosalo/CIFAR-10/imageclassification/training/arguments.py
code served (permissive licence) · get_code("ab5cf2c3305cebc8")
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 int; 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 ast

def aslist(lst):
    return ast.literal_eval(lst)

# The function aslist takes a single parameter 'lst' which should be a string
# representation of a Python literal (list, dict, tuple, etc.)
# and converts it to an actual Python object using ast.literal_eval

ARGS = {
    "lst": "[1, 2, 3, 4, 5]"
}
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 list, ndim 1, dtype family int — 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
ev Ran
ran on a synthesized input
aisosalo/CIFAR-10/run_evaluation.py
code served (permissive licence) · get_code("5e66e1cf8a48f819")
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
import numpy as np

torch.manual_seed(0)

# Create a minimal model that takes inputs and outputs logits
class MinimalModel(nn.Module):
    def __init__(self, num_classes=3):
        super().__init__()
        self.linear = nn.Linear(4, num_classes)
    
    def forward(self, x):
        return self.linear(x)

# Create a minimal data loader - list of dicts with 'img' and 'label' keys
def create_loader(num_batches=2, batch_size=2, num_classes=3):
    loader = []
    for _ in range(num_batches):
        batch = {
            'img': torch.randn(batch_size, 4),
            'label': torch.randint(0, num_classes, (batch_size,))
        }
        loader.append(batch)
    return loader

model = MinimalModel(num_classes=3)
device = 'cpu'
loader = create_loader(num_batches=2, batch_size=2, num_classes=3)

ARGS = {
    '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, torch, tqdm — 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
init_dataset Ran
ran on a synthesized input
aisosalo/CIFAR-10/imageclassification/training/dataset.py
code served (permissive licence) · get_code("d0941755b1d1eba6")
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 os
import pickle
import tempfile
import numpy as np

# Create temporary directory and mock CIFAR10 data files
temp_dir = tempfile.mkdtemp()
cifar10_dir = os.path.join(temp_dir, 'CIFAR10')
os.makedirs(cifar10_dir, exist_ok=True)

# Create mock CIFAR10 train batch files
for i in range(1, 6):
    batch_data = {
        b'data': np.random.randint(0, 256, (10, 3072), dtype=np.uint8),
        b'labels': np.random.randint(0, 10, (10,))
    }
    batch_file = os.path.join(cifar10_dir, f'data_batch_{i}')
    with open(batch_file, 'wb') as f:
        pickle.dump(batch_data, f)

# Create mock CIFAR10 test batch file
test_data = {
    b'data': np.random.randint(0, 256, (10, 3072), dtype=np.uint8),
    b'labels': np.random.randint(0, 10, (10,))
}
test_file = os.path.join(cifar10_dir, 'test_batch')
with open(test_file, 'wb') as f:
    pickle.dump(test_data, f)

# Create mock CIFAR100 files
cifar100_dir = os.path.join(temp_dir, 'CIFAR100')
os.makedirs(cifar100_dir, exist_ok=True)

train_data = {
    b'data': np.random.randint(0, 256, (10, 3072), dtype=np.uint8),
    b'fine_labels': np.random.randint(0, 100, (10,))
}
train_file = os.path.join(cifar100_dir, 'train')
with open(train_file, 'wb') as f:
    pickle.dump(train_data, f)

test_data_100 = {
    b'data': np.random.randint(0, 256, (10, 3072), dtype=np.uint8),
    b'fine_labels': np.random.randint(0, 100, (10,))
}
test_file_100 = os.path.join(cifar100_dir, 'test')
with open(test_file_100, 'wb') as f:
    pickle.dump(test_data_100, f)

ARGS = {
    'path': temp_dir,
    'dataset': 'CIFAR10',
    'batch': 'train'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, termcolor — 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
init_mean_std Ran
ran on a synthesized input
aisosalo/CIFAR-10/imageclassification/training/transformations.py
code served (permissive licence) · get_code("be2b61117e25938d")
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 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 os
import tempfile
import torch
import numpy as np
from torch.utils.data import DataLoader

torch.manual_seed(0)

# Create a temporary directory for saving mean_std files
temp_dir = tempfile.mkdtemp()

# Create a minimal dataset class that returns batches with 'img' key
class MinimalDataset:
    def __init__(self, num_samples=4, num_channels=3, height=8, width=8):
        self.num_samples = num_samples
        self.num_channels = num_channels
        self.height = height
        self.width = width
    
    def __len__(self):
        return self.num_samples
    
    def __getitem__(self, idx):
        # Return a dict with 'img' key containing a tensor of shape (C, H, W)
        img = torch.randn(self.num_channels, self.height, self.width)
        return {'img': img}

# Instantiate the minimal dataset
dataset = MinimalDataset(num_samples=4, num_channels=3, height=8, width=8)

# Define ARGS dictionary with all parameters
ARGS = {
    'dataset': dataset,
    'batch_size': 2,
    'n_threads': 0,
    'save_mean_std': temp_dir,
    'color_space': 'rgb'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, termcolor, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, 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
np_to_png Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
modestyachts/CIFAR-10.1/code/utils.py
code served (permissive licence) · get_code("ae13c8ccf56adae8")
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 bytes. 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
import torch

torch.manual_seed(0)

# Create a small numpy array representing an image
# The function expects uint8 values (0-255) for an image
# Using a small 8x8 RGB image
a = np.random.randint(0, 256, size=(8, 8, 3), dtype=np.uint8)

# fmt is a string for the image format
fmt = 'png'

# scale is an integer multiplier for resizing
scale = 2

ARGS = {
    'a': a,
    'fmt': fmt,
    'scale': scale
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
PIL, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bytes — a bytes output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a bytes 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
unpickle Ran
ran; our contract draft was wrong, not the code
aisosalo/CIFAR-10/build_dataset.py
code served (permissive licence) · get_code("b8200f6c2f73edda")
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
import tempfile
import os

# Create a temporary pickle file with sample data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, 'test.pkl')

# Create sample data that matches CIFAR format
sample_data = {
    b'batch_label': b'training batch 1 of 5',
    b'labels': [0, 1, 2, 3, 4],
    b'data': b'sample_pixel_data',
    b'filenames': [b'file1.png', b'file2.png']
}

# Write the pickle file
with open(temp_file, 'wb') as f:
    pickle.dump(sample_data, f)

# Define arguments for the unpickle function
ARGS = {
    'file': temp_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
unpickle Ran
ran on a synthesized input
aisosalo/CIFAR-10/imageclassification/training/dataset.py
code served (permissive licence) · get_code("54d88bd375890f88")
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
import tempfile
import os

# Create a temporary pickle file with sample CIFAR-like data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test_batch")

# Create a sample dictionary similar to CIFAR format
sample_data = {
    b'batch_label': b'training batch 1',
    b'labels': [0, 1, 2, 3, 4],
    b'data': b'sample_image_data',
    b'filenames': [b'lena_s.png', b'airplane_s.png']
}

# Pickle the sample data
with open(temp_file, 'wb') as f:
    pickle.dump(sample_data, f)

# Define ARGS
ARGS = {
    'file': temp_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
build_cifar100_metadata Not yet run
unverified (harvested, no recorded run)
aisosalo/CIFAR-10/build_dataset.py
code served (permissive licence) · get_code("e28e30f82310aaae")
How it ran

no invocation recorded.

build_cifar10_metadata Not yet run
unverified (harvested, no recorded run)
aisosalo/CIFAR-10/build_dataset.py
code served (permissive licence) · get_code("fb3c67bcbad48f16")
How it ran

no invocation recorded.

load_distances_to_cifar10 Not yet run
unverified (harvested, no recorded run)
modestyachts/CIFAR-10.1/code/utils.py
code served (permissive licence) · get_code("a9822666515c5938")
How it ran

no invocation recorded.

load_new_test_data Not yet run
unverified (harvested, no recorded run)
modestyachts/CIFAR-10.1/code/utils.py
code served (permissive licence) · get_code("c21e963cc74afdd7")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 later) and 10 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
unpickle 2 other papers
1 dated earlier, 1 later
1502.03044 (2015-02)
Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
1904.12043 (2019-04)
Dynamic Mini-batch SGD for Elastic Distributed Training: Learning in the Limbo of Resources

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

Machine learning is currently dominated by largely experimental work focused on improvements in a few key tasks. However, the impressive accuracy numbers of the best performing models are questionable because the same test sets have been used to select these models for multiple years now. To understand the danger of overfitting, we measure the accuracy of CIFAR-10 classifiers by creating a new test set of truly unseen images. Although we ensure that the new test set is as close to the original data distribution as possible, we find a large drop in accuracy (4% to 10%) for a broad range of deep learning models. Yet more recent models with higher original accuracy show a smaller drop and better overall performance, indicating that this drop is likely not due to overfitting based on adaptivity. Instead, we view our results as evidence that current accuracy numbers are brittle and susceptible to even minute natural variations in the data distribution.

For agents

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

get_harvested_code_for_paper("1806.00451")
get_code_for_paper("1806.00451")
have("1806.00451")

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

curl https://syntology.ai/api/ran/1806.00451.json

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

[![Syntology run record](https://syntology.ai/api/ran/1806.00451.svg)](https://syntology.ai/paper/1806.00451)

Connect an agent — have() is free.