SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2002.07933 · ICML · 2020

Improving Generalization by Controlling Label-Noise Information in Neural Network Weights

Hrayr Harutyunyan, Kyle Reing, Greg Steeg, Aram Galstyan

arXiv · PDF · Open in the Atlas

Code that ran

6 of 7 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; 1 failure of our instrument; 1 no recorded run), measured 2026-11-21; 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/2002.07933.json · badge: /api/ran/2002.07933.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
hrayrhar/limit-label-memorization canonical (classifier label)
classifier basis: census
6 ran · 1 no recorded run
FunctionStatusWhere it lives
dmi Ran
ran on a synthesized input
hrayrhar/limit-label-memorization/modules/losses.py
code served (permissive licence) · get_code("01eef8537d9a77ab")
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 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)

# The function dmi expects:
# - target: a tensor that will be transposed and used in torch.mm
# - pred: a tensor that will be used in torch.mm with target.T
# 
# torch.mm(target.T, pred) requires:
# - target.T has shape (D, N)
# - pred has shape (N, D)
# where target has shape (N, D)
#
# For torch.det to work, the resulting matrix must be square,
# so target.T @ pred must be (D, D)
# This means: (D, N) @ (N, D) = (D, D) ✓

# Create target with shape (2, 4) - batch_size=2, dim=4
target = torch.randn(2, 4)

# Create pred with shape (2, 4) - same as target for square result after transpose
# (2, 4).T @ (2, 4) = (4, 2) @ (2, 4) = (4, 4) ✓
pred = torch.randn(2, 4)

ARGS = {
    "target": target,
    "pred": pred,
}
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 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
fw Ran
ran on a synthesized input
hrayrhar/limit-label-memorization/modules/losses.py
code served (permissive licence) · get_code("a78e5788e438f4b2")
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 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Analyze the function:
# - target: used with torch.sum and torch.log operations, needs shape (batch, classes)
# - pred: clamped, used in torch.mm(pred, T_est), so shape (batch, intermediate)
# - T_est: used in torch.mm(pred, T_est), so shape (intermediate, classes)
# The result of mm should be (batch, classes) to match target for element-wise mult

batch_size = 2
intermediate_dim = 3
num_classes = 4

# target: (batch, classes) - probability distribution over classes
target = torch.tensor([
    [0.1, 0.2, 0.3, 0.4],
    [0.25, 0.25, 0.25, 0.25]
], dtype=torch.float32)

# pred: (batch, intermediate) - predictions that will be matrix multiplied with T_est
pred = torch.tensor([
    [0.5, 0.3, 0.2],
    [0.4, 0.4, 0.2]
], dtype=torch.float32)

# T_est: (intermediate, classes) - transformation matrix
T_est = torch.tensor([
    [0.1, 0.2, 0.3, 0.4],
    [0.4, 0.3, 0.2, 0.1],
    [0.2, 0.2, 0.3, 0.3]
], dtype=torch.float32)

ARGS = {
    "target": target,
    "pred": pred,
    "T_est": T_est,
}
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 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
gce Ran
ran on a synthesized input
hrayrhar/limit-label-memorization/modules/losses.py
code served (permissive licence) · get_code("f2cfffff6c7c3e88")
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 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)

# Create small tensors for the gce function
# target: one-hot or probability distribution over classes, shape (batch_size, num_classes)
# pred: predicted probabilities, shape (batch_size, num_classes)
batch_size = 2
num_classes = 4

target = torch.tensor([
    [1.0, 0.0, 0.0, 0.0],
    [0.0, 1.0, 0.0, 0.0]
], dtype=torch.float32)

pred = torch.tensor([
    [0.7, 0.2, 0.05, 0.05],
    [0.1, 0.8, 0.05, 0.05]
], dtype=torch.float32)

q = 1.5

ARGS = {
    "target": target,
    "pred": pred,
    "q": q
}
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 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
get_grad_noise_class Ran
ran on a synthesized input
hrayrhar/limit-label-memorization/modules/nn_utils.py
code served (permissive licence) · get_code("ba2589ac23b2a996")
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 FunctionMeta. 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 numpy as np

torch.manual_seed(0)

ARGS = {
    'standard_dev': 0.1,
    'q_dist': 'Gaussian'
}
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 FunctionMeta — a FunctionMeta output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a FunctionMeta 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_grad_replacement_class Ran
ran on a synthesized input
hrayrhar/limit-label-memorization/modules/nn_utils.py
code served (permissive licence) · get_code("c935d78981b4c242")
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 FunctionMeta. 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 numpy as np

torch.manual_seed(0)

ARGS = {
    'sample': False,
    'standard_dev': None,
    'q_dist': 'Gaussian'
}
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 FunctionMeta — a FunctionMeta output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a FunctionMeta 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
make_resnet18k Ran
ran; our contract draft was wrong, not the code
hrayrhar/limit-label-memorization/modules/resnet18_double_descent.py
code served (permissive licence) · get_code("f2c0aa86345bdfa8")
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 PreActResNet. 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 = {
    "k": 64,
    "num_classes": 10,
}
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 PreActResNet — a PreActResNet output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PreActResNet 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
compute_accuracy_with_bootstrapping Not yet run
unverified (harvested, no recorded run)
hrayrhar/limit-label-memorization/modules/evaluation.py
code served (permissive licence) · get_code("4ae47e7c73b92c57")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 6 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
make_resnet18k 1 other paper
1 later
2502.01585 (2025-02)
The φ Curve: The Shape of Generalization through the Lens of Norm-based Capacity Control

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 2026-11-21.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

In the presence of noisy or incorrect labels, neural networks have the undesirable tendency to memorize information about the noise. Standard regularization techniques such as dropout, weight decay or data augmentation sometimes help, but do not prevent this behavior. If one considers neural network weights as random variables that depend on the data and stochasticity of training, the amount of memorized information can be quantified with the Shannon mutual information between weights and the vector of all training labels given inputs, I(w; y | x). We show that for any training algorithm, low values of this term correspond to reduction in memorization of label-noise and better generalization bounds. To obtain these low values, we propose training algorithms that employ an auxiliary network that predicts gradients in the final layers of a classifier without accessing labels. We illustrate the effectiveness of our approach on versions of MNIST, CIFAR-10, and CIFAR-100 corrupted with various noise models, and on a large-scale dataset Clothing1M that has noisy labels.

For agents

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

get_harvested_code_for_paper("2002.07933")
get_code_for_paper("2002.07933")
have("2002.07933")

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

curl https://syntology.ai/api/ran/2002.07933.json

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

[![Syntology run record](https://syntology.ai/api/ran/2002.07933.svg)](https://syntology.ai/paper/2002.07933)

Connect an agent — have() is free.