SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.08984 · 2021

Contrastive Learning Improves Model Robustness Under Label Noise

arXiv · PDF · Open in the Atlas

Code that ran

11 of 13 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; 4 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 5 failures of our instrument; 2 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/2104.08984.json · badge: /api/ran/2104.08984.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
arghosh/noisy_label_pretrain canonical (classifier label)
classifier basis: census
11 ran · 2 no recorded run
FunctionStatusWhere it lives
accuracy Ran
ran; our contract draft was wrong, not the code
arghosh/noisy_label_pretrain/models/losses.py
code served (permissive licence) · get_code("11bc56dd2ecb7644")
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 1, dtype family float, shape (2,) 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)

# The function expects:
# - logits: a tensor with shape (batch_size, num_classes) that supports .topk() and torch.sort()
# - labels: a tensor with shape (batch_size, k) that supports torch.sort()
# - k: an integer for the top-k parameter

batch_size = 2
num_classes = 4
k = 3

# Create logits tensor: (batch_size, num_classes)
logits = torch.randn(batch_size, num_classes)

# Create labels tensor: (batch_size, k) with class indices
# Labels should contain k distinct class indices for each sample
labels = torch.tensor([[0, 1, 2], [1, 2, 3]], dtype=torch.long)

ARGS = {
    "logits": logits,
    "labels": labels,
    "k": k
}
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 1, dtype family float, shape (2,) — shape (2,) 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
asym_noise Ran
ran, honoured the contract we drafted
arghosh/noisy_label_pretrain/utils/utils.py
code served (permissive licence) · get_code("0aeed6613e3f5540")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (10, 10) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Setup for asym_noise function
# The function takes three parameters:
# 1. mixing_ratio: a float between 0 and 1
# 2. num_classes: an integer (either 10 or 100)
# 3. class_to_idx: optional dict mapping class names to indices (required when num_classes != 10)

# For num_classes=10 case, class_to_idx is not used
mixing_ratio_10 = 0.3
num_classes_10 = 10
class_to_idx_10 = None

# For num_classes=100 case, we need to build a class_to_idx mapping
# based on the super_class structure in the function
super_class = {}
super_class['aquatic mammals'] = ['beaver', 'dolphin', 'otter', 'seal', 'whale']
super_class['fish'] = ['aquarium_fish', 'flatfish', 'ray', 'shark', 'trout']
super_class['flowers'] = ['orchid', 'poppy', 'rose', 'sunflower', 'tulip']
super_class['food containers'] = ['bottle', 'bowl', 'can', 'cup', 'plate']
super_class['fruit and vegetables'] = ['apple', 'mushroom', 'orange', 'pear', 'sweet_pepper']
super_class['household electrical devices'] = ['clock', 'keyboard', 'lamp', 'telephone', 'television']
super_class['household furniture'] = ['bed', 'chair', 'couch', 'table', 'wardrobe']
super_class['insects'] = ['bee', 'beetle', 'butterfly', 'caterpillar', 'cockroach']
super_class['large carnivores'] = ['bear', 'leopard', 'lion', 'tiger', 'wolf']
super_class['large man-made outdoor things'] = ['bridge', 'castle', 'house', 'road', 'skyscraper']
super_class['large natural outdoor scenes'] = ['cloud', 'forest', 'mountain', 'plain', 'sea']
super_class['large omnivores and herbivores'] = ['camel', 'cattle', 'chimpanzee', 'elephant', 'kangaroo']
super_class['medium mammals'] = ['fox', 'porcupine', 'possum', 'raccoon', 'skunk']
super_class['non-insect invertebrates'] = ['crab', 'lobster', 'snail', 'spider', 'worm']
super_class['people'] = ['baby', 'boy', 'girl', 'man', 'woman']
super_class['reptiles'] = ['crocodile', 'dinosaur', 'lizard', 'snake', 'turtle']
super_class['small mammals'] = ['hamster', 'mouse', 'rabbit', 'shrew', 'squirrel']
super_class['trees'] = ['maple_tree', 'oak_tree', 'palm_tree', 'pine_tree', 'willow_tree']
super_class['vehicles 1'] = ['bicycle', 'bus', 'motorcycle', 'pickup_truck', 'train']
super_class['vehicles 2'] = ['lawn_mower', 'rocket', 'streetcar', 'tank', 'tractor']

# Build class_to_idx by flattening the super_class structure
class_to_idx_100 = {}
idx = 0
for k, v in super_class.items():
    for class_name in v:
        class_to_idx_100[class_name] = idx
        idx += 1

mixing_ratio_100 = 0.2
num_classes_100 = 100

# Test with num_classes=10 (simpler case, no class_to_idx needed)
ARGS = {
    'mixing_ratio': mixing_ratio_10,
    'num_classes': num_classes_10,
    'class_to_idx': class_to_idx_10
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (10, 10) — shape (10, 10) 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
get_config Ran
ran; our contract draft was wrong, not the code
arghosh/noisy_label_pretrain/scheduler.py
code served (permissive licence) · get_code("e31c345c9535138f")
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. 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)

ARGS = {
    'combo': {
        'dataset': 'clothing'
    }
}
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 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 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_cpu Ran
ran, honoured the contract we drafted
arghosh/noisy_label_pretrain/scheduler.py
code served (permissive licence) · get_code("53f0e581017b7778")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function get_cpu takes a 'combo' parameter which is a dict
# It checks combo['dataset'] to determine the return value
# So we need to provide a dict with a 'dataset' key

combo = {
    'dataset': 'clothing'
}

ARGS = {
    'combo': combo
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_memory Ran
ran, honoured the contract we drafted
arghosh/noisy_label_pretrain/scheduler.py
code served (permissive licence) · get_code("9ef98e4eebd96977")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function only accesses combo['dataset']
# It returns different values based on whether dataset == 'clothing' or not
# We'll test with a combo dict that has a 'dataset' key

ARGS = {
    'combo': {'dataset': 'clothing'}
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_train Ran
ran; our contract draft was wrong, not the code
arghosh/noisy_label_pretrain/imagenet32.py
code served (permissive licence) · get_code("97860ae8a6484f60")
How it ran

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

import numpy as np
import pickle
import os
import tempfile
import shutil

# Create a temporary directory for test data
temp_dir = tempfile.mkdtemp()

# Create mock pickle files that the function expects
def create_mock_pickle_file(filepath, batch_idx):
    """Create a mock pickle file with the expected structure"""
    img_size = 32
    img_size2 = img_size * img_size
    
    # Create mock data: 2 samples per batch
    # Each image has 3 channels flattened: 3 * 1024 = 3072 values
    data = np.random.randint(0, 256, (2, 3 * img_size2), dtype=np.uint8)
    labels = [batch_idx + 1, batch_idx + 2]  # Labels > 0 (will be decremented to 0, 1)
    
    mock_dict = {
        'data': data,
        'labels': labels
    }
    
    with open(filepath, 'wb') as f:
        pickle.dump(mock_dict, f)

# Create 10 batch files (train_data_batch_1 through train_data_batch_10)
for idx in range(1, 11):
    filepath = os.path.join(temp_dir, f'train_data_batch_{idx}')
    create_mock_pickle_file(filepath, idx)

ARGS = {
    'data_folder': temp_dir,
    'img_size': 32
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
mean_cumulative_gain Ran
ran; our contract draft was wrong, not the code
arghosh/noisy_label_pretrain/models/losses.py
code served (permissive licence) · get_code("439f4fcb999995b5")
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 1, dtype family float, shape (2,) 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 small tensors for the function
batch_size = 2
k = 3
num_classes = 4

# logits: shape (batch_size, num_classes)
# The function calls logits.topk(k, dim=1)[1] to get top-k indices
logits = torch.randn(batch_size, num_classes)

# labels: shape (batch_size, k)
# The function expects labels to have k elements per batch for comparison
# After topk, we get k indices, then we sort and compare
labels = torch.randint(0, num_classes, (batch_size, k))

ARGS = {
    "logits": logits,
    "labels": labels,
    "k": k,
}
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 1, dtype family float, shape (2,) — shape (2,) 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
noise_matrix Ran
ran, honoured the contract we drafted
arghosh/noisy_label_pretrain/utils/utils.py
code served (permissive licence) · get_code("2d3453d13dc8c3f4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (10, 10) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Define class_to_idx mapping for the 100-class case (CIFAR-100)
class_to_idx = {
    'beaver': 0, 'dolphin': 1, 'otter': 2, 'seal': 3, 'whale': 4,
    'aquarium_fish': 5, 'flatfish': 6, 'ray': 7, 'shark': 8, 'trout': 9,
    'orchid': 10, 'poppy': 11, 'rose': 12, 'sunflower': 13, 'tulip': 14,
    'bottle': 15, 'bowl': 16, 'can': 17, 'cup': 18, 'plate': 19,
    'apple': 20, 'mushroom': 21, 'orange': 22, 'pear': 23, 'sweet_pepper': 24,
    'clock': 25, 'keyboard': 26, 'lamp': 27, 'telephone': 28, 'television': 29,
    'bed': 30, 'chair': 31, 'couch': 32, 'table': 33, 'wardrobe': 34,
    'bee': 35, 'beetle': 36, 'butterfly': 37, 'caterpillar': 38, 'cockroach': 39,
    'bear': 40, 'leopard': 41, 'lion': 42, 'tiger': 43, 'wolf': 44,
    'bridge': 45, 'castle': 46, 'house': 47, 'road': 48, 'skyscraper': 49,
    'cloud': 50, 'forest': 51, 'mountain': 52, 'plain': 53, 'sea': 54,
    'camel': 55, 'cattle': 56, 'chimpanzee': 57, 'elephant': 58, 'kangaroo': 59,
    'fox': 60, 'porcupine': 61, 'possum': 62, 'raccoon': 63, 'skunk': 64,
    'crab': 65, 'lobster': 66, 'snail': 67, 'spider': 68, 'worm': 69,
    'baby': 70, 'boy': 71, 'girl': 72, 'man': 73, 'woman': 74,
    'crocodile': 75, 'dinosaur': 76, 'lizard': 77, 'snake': 78, 'turtle': 79,
    'hamster': 80, 'mouse': 81, 'rabbit': 82, 'shrew': 83, 'squirrel': 84,
    'maple_tree': 85, 'oak_tree': 86, 'palm_tree': 87, 'pine_tree': 88, 'willow_tree': 89,
    'bicycle': 90, 'bus': 91, 'motorcycle': 92, 'pickup_truck': 93, 'train': 94,
    'lawn_mower': 95, 'rocket': 96, 'streetcar': 97, 'tank': 98, 'tractor': 99,
}

ARGS = {
    'mixing_ratio': 0.1,
    'num_classes': 10,
    'noise_type': 'asym',
    'class_to_idx': class_to_idx,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (10, 10) — shape (10, 10) 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
pad Ran
ran on a synthesized input
arghosh/noisy_label_pretrain/utils/datautils.py
code served (permissive licence) · get_code("eab99ab911f1645a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family int, shape (8, 7, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import PIL

# Create a small test image as numpy array (height=4, width=3, channels=3)
img_array = np.random.randint(0, 256, size=(4, 3, 3), dtype=np.uint8)

# Convert to PIL Image
pil_img = PIL.Image.fromarray(img_array)

ARGS = {
    "img": pil_img,
    "size": 2,
    "mode": "constant"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (8, 7, 3) — shape (8, 7, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
uniform_mix_C Ran
ran on a synthesized input
arghosh/noisy_label_pretrain/utils/utils.py
code served (permissive licence) · get_code("ba6749b25e502f54")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (5, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct valid arguments for uniform_mix_C
mixing_ratio = 0.3
num_classes = 5

ARGS = {
    "mixing_ratio": mixing_ratio,
    "num_classes": num_classes
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (5, 5) — shape (5, 5) 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
unpickle Ran
ran; our contract draft was wrong, not the code
arghosh/noisy_label_pretrain/imagenet32.py
pointer only (licence: MIT); the stored copy get_code serves, from haiyunyao/diversemix, is pointer only (licence: NONE) · get_code("4ebc0e09cb353eac")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from haiyunyao/diversemix, 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.

data_loader Not yet run
unverified (harvested, no recorded run)
arghosh/noisy_label_pretrain/imagenet32.py
code served (permissive licence) · get_code("961e370efdf679f8")
How it ran

no invocation recorded.

gather Not yet run
unverified (harvested, no recorded run)
arghosh/noisy_label_pretrain/models/losses.py
code served (permissive licence) · get_code("3d5e11c2280cdb46")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 5 are also present in other papers whose code we hold (5 with at least one paper dated earlier, 0 the same month, 3 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
unpickle 7 other papers
1 dated earlier, 6 later
1911.07205 (2019-11)
REFIT: A Unified Watermark Removal Framework For Deep Learning Systems With Limited Data
2202.03299 (2022-02)
Training OOD Detectors in their Natural Habitats
2307.02129 (2023-07)
How Deep Neural Networks Learn Compositional Data: The Random Hierarchy Model
and 4 more not named here
uniform_mix_C 6 other papers
5 dated earlier, 1 later
1802.05300 (2018-02)
Using Trusted Data to Train Deep Networks on Labels Corrupted by Severe Noise
1901.09960 (2019-01)
Using Pre-Training Can Improve Model Robustness and Uncertainty
1902.07379 (2019-02)
Meta-Weight-Net: Learning an Explicit Mapping For Sample Weighting
and 3 more not named here
accuracy 3 other papers
1 dated earlier, 2 later
2002.05709 (2020-02)
A Simple Framework for Contrastive Learning of Visual Representations
2112.11450 (2021-12)
Max-Margin Contrastive Learning
2309.05300 (2023-09)
Decoupling Common and Unique Representations for Multimodal Self-supervised Learning
mean_cumulative_gain 1 other paper
1 dated earlier
2002.05709 (2020-02)
A Simple Framework for Contrastive Learning of Visual Representations
pad 1 other paper
1 dated earlier
2002.05709 (2020-02)
A Simple Framework for Contrastive Learning of Visual Representations

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

Deep neural network-based classifiers trained with the categorical cross-entropy (CCE) loss are sensitive to label noise in the training data. One common type of method that can mitigate the impact of label noise can be viewed as supervised robust methods; one can simply replace the CCE loss with a loss that is robust to label noise, or re-weight training samples and down-weight those with higher loss values. Recently, another type of method using semi-supervised learning (SSL) has been proposed, which augments these supervised robust methods to exploit (possibly) noisy samples more effectively. Although supervised robust methods perform well across different data types, they have been shown to be inferior to the SSL methods on image classification tasks under label noise. Therefore, it remains to be seen that whether these supervised robust methods can also perform well if they can utilize the unlabeled samples more effectively. In this paper, we show that by initializing supervised robust methods using representations learned through contrastive learning leads to significantly improved performance under label noise. Surprisingly, even the simplest method (training a classifier with the CCE loss) can outperform the state-of-the-art SSL method by more than 50\% under high label noise when initialized with contrastive learning. Our implementation will be publicly available at {\url{https://github.com/arghosh/noisy_label_pretrain}}.

For agents

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

get_harvested_code_for_paper("2104.08984")
get_code_for_paper("2104.08984")
have("2104.08984")

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

curl https://syntology.ai/api/ran/2104.08984.json

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

[![Syntology run record](https://syntology.ai/api/ran/2104.08984.svg)](https://syntology.ai/paper/2104.08984)

Connect an agent — have() is free.