SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.06221 · NeurIPS · 2023

CSOT: Curriculum and Structure-Aware Optimal Transport for Learning with Noisy Labels

Wanxing Chang, Ye Shi, Jingya Wang

arXiv · PDF · Open in the Atlas

Code that ran

6 of 12 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, 0 violated it and 0 have no contract verdict; 5 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/2312.06221.json · badge: /api/ran/2312.06221.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
changwxx/CSOT-for-LNL canonical (classifier label)
classifier basis: classifier_2025
1 ran · 1 no recorded run
changwxx/csot-for-lnl community
Papers with Code archive
4 ran · 4 no recorded run
lijichang/LNL-NCE community
Papers with Code archive
1 ran · 1 no recorded run
FunctionStatusWhere it lives
entropic_COT Ran
ran; our contract draft was wrong, not the code
changwxx/CSOT-for-LNL/plabel_allocator.py
code served (permissive licence) · get_code("8038d1de287350ab")
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 (2, 4) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors for the function
# a: distribution over dim_a (first dimension of M)
# b: distribution over dim_b (second dimension of M)
# M: cost matrix of shape (dim_a, dim_b)

dim_a = 2
dim_b = 4

a = torch.ones(dim_a, dtype=torch.float64)
b = torch.ones(dim_b, dtype=torch.float64)

# Cost matrix - distinct dimensions to probe shapes
M = torch.randn(dim_a, dim_b, dtype=torch.float64)

# Regularization parameter
reg = 0.1

# Other parameters
numItermax = 100
stopThr = 1e-9
verbose = False
log = False

ARGS = {
    'a': a,
    'b': b,
    'M': M,
    'reg': reg,
    'numItermax': numItermax,
    'stopThr': stopThr,
    'verbose': verbose,
    'log': log
}
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 (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
entropic_COT_fast Ran
ran; our contract draft was wrong, not the code
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("db10c25daf92d9cb")
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, 3) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors for the function
# a: distribution over dim_a (rows of M)
# b: distribution over dim_b (columns of M)
# M: cost matrix of shape (dim_a, dim_b)

dim_a = 4
dim_b = 3

# a and b should be probability distributions (sum to 1)
a = torch.ones(dim_a, dtype=torch.float64) / dim_a
b = torch.ones(dim_b, dtype=torch.float64) / dim_b

# M is a cost matrix
M = torch.randn(dim_a, dim_b, dtype=torch.float64)

# reg: regularization parameter (should be positive)
reg = 0.1

# numItermax: maximum number of iterations
numItermax = 100

# stopThr: stopping threshold
stopThr = 1e-09

# verbose: whether to print verbose output
verbose = False

# log: whether to return log dictionary
log = False

ARGS = {
    'a': a,
    'b': b,
    'M': M,
    'reg': reg,
    'numItermax': numItermax,
    'stopThr': stopThr,
    'verbose': verbose,
    'log': log
}
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, 3) — shape (4, 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
mixup Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lijichang/LNL-NCE/cifar/utils.py
pointer only (licence: NONE) · get_code("e612a78761dac3a7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, lijichang/LNL-NCE 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.

omega Ran
ran; our contract draft was wrong, not the code
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("600754cf02efbf4a")
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 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

torch.manual_seed(0)

# Analyze the function:
# - S, M, Q are all used with torch operations (mul, matmul, sum, t)
# - They appear to be tensors
# - M and Q are multiplied together, then matmul with transpose
# - S is multiplied with the result
# - All are summed at the end

# Create small tensors with distinct dimensions for clarity
S = torch.randn(4, 4)  # 4x4 matrix
M = torch.randn(4, 3)  # 4x3 matrix
Q = torch.randn(4, 3)  # 4x3 matrix

ARGS = {
    "S": S,
    "M": M,
    "Q": Q
}
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 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
omega_df Ran
ran; our contract draft was wrong, not the code
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("987deaa594328394")
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, 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 torch

torch.manual_seed(0)

# Construct valid arguments for omega_df(S, M, Q)
# The function performs: gradient = 2 * torch.mul(torch.matmul(S, temp), M)
# where temp = torch.mul(M, Q)
# 
# This requires:
# - S: matrix for matmul with temp (result of M*Q)
# - M: used in element-wise mul with Q, and element-wise mul with matmul result
# - Q: used in element-wise mul with M

# For matmul(S, temp) to work:
# If M and Q are (batch, dim) or (dim, dim), temp will have same shape
# S needs compatible shape for matmul

# Let's use:
# M: (4, 3) - represents a 4x3 matrix
# Q: (4, 3) - same shape for element-wise multiplication
# S: (4, 4) - square matrix that can matmul with (4, 3) -> (4, 3)

S = torch.randn(4, 4)
M = torch.randn(4, 3)
Q = torch.randn(4, 3)

ARGS = {
    "S": S,
    "M": M,
    "Q": Q,
}
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, 3) — shape (4, 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
scalar_search_armijo Ran
ran, honoured the contract we drafted
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("6fd4c5c02034cba6")
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

torch.manual_seed(0)

# The function phi should be callable and take a scalar alpha value
# and return a scalar value (phi(alpha)).
# We'll create a simple quadratic function: phi(alpha) = alpha^2 - 2*alpha + 1
def phi(alpha):
    """Simple quadratic function for line search testing"""
    return alpha ** 2 - 2 * alpha + 1

# phi0 is the value of phi at alpha=0
phi0 = 1.0

# derphi0 is the derivative of phi at alpha=0
# For phi(alpha) = alpha^2 - 2*alpha + 1, phi'(alpha) = 2*alpha - 2
# At alpha=0: phi'(0) = -2
derphi0 = -2.0

# Standard parameters
c1 = 0.0001
alpha0 = 1.0
amin = 0

ARGS = {
    "phi": phi,
    "phi0": phi0,
    "derphi0": derphi0,
    "c1": c1,
    "alpha0": alpha0,
    "amin": amin,
}
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
create_model Not yet run
unverified (harvested, no recorded run)
changwxx/CSOT-for-LNL/main_cifar.py
code served (permissive licence) · get_code("da7dd5e490f22638")
How it ran

no invocation recorded.

curriculum_structure_aware_PL Not yet run
unverified (harvested, no recorded run)
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("a345b7c52a07008a")
How it ran

no invocation recorded.

entropic_COT_extra_reg Not yet run
unverified (harvested, no recorded run)
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("bb7cd0d56bf0c512")
How it ran

no invocation recorded.

entropic_COT_gcg Not yet run
unverified (harvested, no recorded run)
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("9dce4bf67e1fc812")
How it ran

no invocation recorded.

line_search_armijo Not yet run
unverified (harvested, no recorded run)
changwxx/csot-for-lnl/plabel_allocator.py
code served (permissive licence) · get_code("219fa6fca102ccb2")
How it ran

no invocation recorded.

train Not yet run
unverified (harvested, no recorded run)
lijichang/LNL-NCE/cifar/utils.py
pointer only (licence: NONE) · get_code("9a1ea56d8b41b7b2")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, lijichang/LNL-NCE 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.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 2 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 2 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
entropic_COT
brick of this paper's assembly: helper
1 other paper
1 later
2607.10548 (2026-07)
Beyond Looking Up, Try Looking Around: Harmonizing Global Structure and Local Consistency in Optimal Transport for Short Text Clustering
entropic_COT_gcg
brick of this paper's assembly: helper
1 other paper
1 later
2502.06501 (2025-02)
Learning Clustering-based Prototypes for Compositional Zero-shot Learning

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

Learning with noisy labels (LNL) poses a significant challenge in training a wellgeneralized model while avoiding overfitting to corrupted labels. Recent advances have achieved impressive performance by identifying clean labels and correcting corrupted labels for training. However, the current approaches rely heavily on the model's predictions and evaluate each sample independently without considering either the global or local structure of the sample distribution. These limitations typically result in a suboptimal solution for the identification and correction processes, which eventually leads to models overfitting to incorrect labels. In this paper, we propose a novel optimal transport (OT) formulation, called Curriculum and Structure-aware Optimal Transport (CSOT). CSOT concurrently considers the interand intra-distribution structure of the samples to construct a robust denoising and relabeling allocator. During the training process, the allocator incrementally assigns reliable labels to a fraction of the samples with the highest confidence. These labels have both global discriminability and local coherence. Notably, CSOT is a new OT formulation with a nonconvex objective function and curriculum constraints, so it is not directly compatible with classical OT solvers. Here, we develop a lightspeed computational method that involves a scaling iteration within a generalized conditional gradient framework to solve CSOT efficiently. Extensive experiments demonstrate the superiority of our method over the current state-of-the-arts in LNL. Code is available at https://github.com/changwxx/CSOT-for-LNL.

For agents

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

get_harvested_code_for_paper("2312.06221")
get_code_for_paper("2312.06221")
have("2312.06221")

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

curl https://syntology.ai/api/ran/2312.06221.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.06221.svg)](https://syntology.ai/paper/2312.06221)

Connect an agent — have() is free.