SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · aaai_33972 · AAAI · 2025

SAP: Corrective Machine Unlearning with Scaled Activation Projection for Label Noise Robustness

Kaushik Roy, Gobinda Saha, Sangamesh Kodge, Deepak Ravikumar

Open in the Atlas

Code that ran

12 of 17 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; 11 failures of our instrument; 5 no recorded run), measured 2026-11-22; 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/aaai_33972.json · badge: /api/ran/aaai_33972.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
sangamesh-kodge/SAP canonical (classifier label)
classifier basis: census
12 ran · 5 no recorded run
FunctionStatusWhere it lives
compute_conv_output_size Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/models/model_utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from sangamesh-kodge/class_forgetting, is pointer only (licence: NONE) · get_code("cec2ecd572648904")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from sangamesh-kodge/class_forgetting, 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.

forward_cache_activations Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/models/model_utils.py
code served (permissive licence) · get_code("ded7f8ff8bcea57c")
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
from collections import OrderedDict
from copy import deepcopy

torch.manual_seed(0)

# Create a small Conv2d layer
conv_layer = nn.Conv2d(in_channels=3, out_channels=4, kernel_size=3, stride=1, padding=1, dilation=1)

# Create input tensor for Conv2d: (batch=2, channels=3, height=5, width=5)
x_conv = torch.randn(2, 3, 5, 5)

# Create a small Linear layer
linear_layer = nn.Linear(in_features=8, out_features=6)

# Create input tensor for Linear: (batch=2, features=8)
x_linear = torch.randn(2, 8)

# Create projection matrices for the case with prev_recur_proj_mat
# For Conv2d pre: (kernel_size * in_channels, proj_dim) = (3*3*3, 10)
prev_recur_proj_mat_conv = {
    'pre': {
        'conv_key': torch.randn(3*3*3, 10)
    },
    'post': {
        'conv_key': torch.randn(4, 4)
    }
}

# For Linear pre: (in_features, proj_dim) = (8, 7)
prev_recur_proj_mat_linear = {
    'pre': {
        'linear_key': torch.randn(8, 7)
    },
    'post': {
        'linear_key': torch.randn(6, 5)
    }
}

# Initialize activation dict
act_dict = {'pre': OrderedDict(), 'post': OrderedDict()}

# Test case 1: Conv2d without projection matrix
ARGS = {
    'x': x_conv,
    'layer': conv_layer,
    'key': 'conv_key',
    'prev_recur_proj_mat': None,
    'act': deepcopy(act_dict)
}
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
get_corrupt_targets Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/demo_spiral.py
code served (permissive licence) · get_code("88caca6c6146a0d7")
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 numpy as np
import random
import copy

torch.manual_seed(0)
np.random.seed(0)
random.seed(0)

# Create sample data: (features, targets)
# features: batch of 2 samples, each with 4 dimensions
# targets: batch of 2 binary labels (0 or 1)
features = torch.randn(2, 4)
targets = torch.tensor([0, 1])
data = (features, targets)

# Fraction of data points to corrupt
fraction_corrupt = 0.5

ARGS = {
    "data": data,
    "fraction_corrupt": fraction_corrupt
}
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 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
get_dataset Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/demo_spiral.py
code served (permissive licence) · get_code("e969309a21a6d5ca")
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 numpy as np
from numpy import pi

torch.manual_seed(0)
np.random.seed(0)

ARGS = {
    'train_samples_per_class': 2,
    'val_samples_per_class': 2,
    'test_samples_per_class': 2,
    'intrinsic_randomness': False,
}
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 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
get_mislabeled_dataset Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/utils.py
code served (permissive licence) · get_code("92ebf6ffc609374c")
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 numpy as np
import random
import torch

torch.manual_seed(0)
np.random.seed(0)
random.seed(0)

# Create a minimal dataset class that satisfies the function's requirements
class MinimalDataset:
    def __init__(self, num_samples=10, num_classes=4):
        # Create dummy data: (num_samples, 3, 4, 4) for small images
        self.data = torch.randn(num_samples, 3, 4, 4)
        # Create targets as a list of integers
        self.targets = [i % num_classes for i in range(num_samples)]
    
    def __len__(self):
        return len(self.data)
    
    def __getitem__(self, idx):
        # Return (image, label) tuple
        return (self.data[idx], self.targets[idx])

# Create dataset instance
dataset = MinimalDataset(num_samples=10, num_classes=4)

# Create a temporary directory-like path (no actual file I/O)
import tempfile
temp_dir = tempfile.gettempdir()
path = os.path.join(temp_dir, "test_dataset")

ARGS = {
    "dataset": dataset,
    "percentage_mislabeled": 0.3,
    "num_classes": 4,
    "return_clean_partition": False,
    "path": path
}
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 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
get_test_set_clothing Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/utils.py
code served (permissive licence) · get_code("cebea7e1cf9b44ad")
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 ImageFolder. 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 tempfile
import torch
from torchvision import datasets
from torchvision.transforms import v2
from PIL import Image
import numpy as np

torch.manual_seed(0)

# Create a temporary directory structure with test images
temp_dir = tempfile.mkdtemp()
test_dir = os.path.join(temp_dir, 'test')
class1_dir = os.path.join(test_dir, 'class1')
class2_dir = os.path.join(test_dir, 'class2')

os.makedirs(class1_dir, exist_ok=True)
os.makedirs(class2_dir, exist_ok=True)

# Create dummy test images
for i in range(2):
    img_array = np.random.randint(0, 256, (224, 224, 3), dtype=np.uint8)
    img = Image.fromarray(img_array)
    img.save(os.path.join(class1_dir, f'img_{i}.jpg'))
    
    img_array = np.random.randint(0, 256, (224, 224, 3), dtype=np.uint8)
    img = Image.fromarray(img_array)
    img.save(os.path.join(class2_dir, f'img_{i}.jpg'))

# Create args object as a simple class
class Args:
    def __init__(self, arch, data_path):
        self.arch = arch
        self.data_path = data_path
        self.resize_image = None

ARGS = {
    'args': Args(arch='resnet50', data_path=temp_dir)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ImageFolder — a ImageFolder 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
interpolate_embeddings Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/models/vit.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from sangamesh-kodge/class_forgetting, is pointer only (licence: NONE) · get_code("e9cd32628e2a4c34")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from sangamesh-kodge/class_forgetting, 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.

make_layers Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/models/vgg.py
code served (permissive licence) · get_code("ac62432dc5134b0d")
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 Sequential. 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 two parameters: cfg and batch_norm
# cfg should be a list/sequence of either integers (channel counts) or 'M' (maxpool marker)
# batch_norm is a boolean flag

# Using one of the predefined configurations from the module-level cfg dict
cfg_list = [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M']

ARGS = {
    'cfg': cfg_list,
    'batch_norm': False
}
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 Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 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
reshape_conv_input_activation Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
sangamesh-kodge/SAP/models/model_utils.py
code served (permissive licence) · get_code("1f16cd6ab762523f")
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 2, dtype family float, shape (32, 27) 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)

# Create a small Conv2d layer for testing
conv_layer = nn.Conv2d(in_channels=3, out_channels=16, kernel_size=3, stride=1, padding=1, dilation=1)

# Create a small input tensor: (batch=2, channels=3, height=4, width=4)
x = torch.randn(2, 3, 4, 4)

ARGS = {
    "x": x,
    "conv_layer": conv_layer,
    "kernel_size": 3,
    "stride": 1,
    "padding": 0,
    "dilation": 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 Tensor, ndim 2, dtype family float, shape (32, 27) — shape (32, 27) 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
train Ran
ran, honoured the contract we drafted
sangamesh-kodge/SAP/demo_spiral.py
code served (permissive licence) · get_code("0bc5fd5d3b82848e")
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 float, 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
import torch.optim as optim

torch.manual_seed(0)

# Define a minimal model that:
# - has a train() method (inherited from nn.Module)
# - takes data as input and returns log probabilities
# - works with F.nll_loss
class SimpleModel(nn.Module):
    def __init__(self, input_dim=4, num_classes=3):
        super().__init__()
        self.linear = nn.Linear(input_dim, num_classes)
        self.log_softmax = nn.LogSoftmax(dim=1)
    
    def forward(self, x):
        return self.log_softmax(self.linear(x))

# Create model
model = SimpleModel(input_dim=4, num_classes=3)

# Create device
device = 'cpu'

# Create optimizer
optimizer = optim.SGD(model.parameters(), lr=0.01)

# Create scheduler (with step method that takes loss)
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min')

# Create train_loader: list of (data, target) tuples
# batch_size=2, input_dim=4, num_classes=3
train_loader = [
    (torch.randn(2, 4), torch.randint(0, 3, (2,))),
    (torch.randn(2, 4), torch.randint(0, 3, (2,)))
]
# Add dataset attribute for len(train_loader.dataset)
class SimpleDataset:
    def __init__(self, size):
        self.size = size
    def __len__(self):
        return self.size

train_loader_obj = train_loader
# Convert to object with dataset attribute
class DataLoaderWrapper:
    def __init__(self, data_list, dataset_size):
        self.data_list = data_list
        self.dataset = SimpleDataset(dataset_size)
    
    def __iter__(self):
        return iter(self.data_list)

train_loader = DataLoaderWrapper(train_loader, dataset_size=4)

ARGS = {
    'model': model,
    'train_loader': train_loader,
    'device': device,
    'optimizer': optimizer,
    'schedular': scheduler
}
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 float, 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
vgg11 Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/models/vgg.py
code served (permissive licence) · get_code("7806cd2294939410")
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 VGG. 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 math

torch.manual_seed(0)

# Define stub functions that are referenced in VGG but not provided
def auto_get_activations(x, module, name, prev_recur_proj_mat, act):
    """Stub function for get_activations"""
    return act, x

def auto_project_weights(module, name, projection_mat_dict, proj_classifier=False):
    """Stub function for project_weights"""
    pass

ARGS = {
    'num_classes': 10,
    'dataset': 'imagenet',
    'do_log_softmax': False,
}
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 VGG — a VGG output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a VGG 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
vgg11_bn Ran
ran; our contract draft was wrong, not the code
sangamesh-kodge/SAP/models/vgg.py
code served (permissive licence) · get_code("5a9e5f8cb1d2d696")
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 VGG. 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 math
from collections import OrderedDict

torch.manual_seed(0)

ARGS = {
    'num_classes': 10,
    'dataset': 'imagenet',
    'do_log_softmax': False
}
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 VGG — a VGG output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a VGG 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
ResNet18 Not yet run
unverified (harvested, no recorded run)
sangamesh-kodge/SAP/models/resnet.py
code served (permissive licence) · get_code("7aad35d4ddbec001")
How it ran

no invocation recorded.

ResNet34 Not yet run
unverified (harvested, no recorded run)
sangamesh-kodge/SAP/models/resnet.py
code served (permissive licence) · get_code("a5f542287fef57fe")
How it ran

no invocation recorded.

ResNet50 Not yet run
unverified (harvested, no recorded run)
sangamesh-kodge/SAP/models/resnet.py
code served (permissive licence) · get_code("7cf54bdcae924f20")
How it ran

no invocation recorded.

get_dataset Not yet run
unverified (harvested, no recorded run)
sangamesh-kodge/SAP/utils.py
code served (permissive licence) · get_code("5f6e7cee2efbd51c")
How it ran

no invocation recorded.

inceptionresnetv2 Not yet run
unverified (harvested, no recorded run)
sangamesh-kodge/SAP/models/inceptionresnetv2.py
code served (permissive licence) · get_code("4e50b2fee4494463")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 4 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 1 later) and 13 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_layers 59 other papers
55 dated earlier, 1 later, 3 same year, order not determinable
1411.4038 (2014-11)
Fully Convolutional Networks for Semantic Segmentation
1511.00561 (2015-11)
SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation
1511.05950 (2015-11)
Staleness-aware Async-SGD for Distributed Deep Learning
and 56 more not named here
inceptionresnetv2 4 other papers
4 dated earlier
1905.00953 (2019-05)
Omni-Scale Feature Learning for Person Re-Identification
1910.06180 (2019-10)
KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment
2010.05435 (2020-10)
Top-DB-Net: Top DropBlock for Activation Enhancement in Person Re-Identification
and 1 more not named here
compute_conv_output_size 1 other paper
1 dated earlier
2312.00761 (2023-12)
Deep Unlearning: Fast and Efficient Gradient-free Approach to Class Forgetting
interpolate_embeddings 1 other paper
1 dated earlier
2312.00761 (2023-12)
Deep Unlearning: Fast and Efficient Gradient-free Approach to Class Forgetting

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

Label corruption, where training samples are mislabeled due to non-expert annotation or adversarial attacks, significantly degrades model performance. Acquiring large, perfectly labeled datasets is costly, and retraining models from scratch is computationally expensive. To address this, we introduce Scaled Activation Projection (SAP), a novel SVD (Singular Value Decomposition)-based corrective machine unlearning algorithm. SAP mitigates label noise by identifying a small subset of trusted samples using cross-entropy loss and projecting model weights onto a clean activation space estimated using SVD on these trusted samples. This process suppresses the noise introduced in activations due to the mislabeled samples. In our experiments, we demonstrate SAP's effectiveness on synthetic noise with different settings and real-world label noise. SAP applied to the CIFAR dataset with 25% synthetic corruption show upto 6% generalization improvements. Additionally, SAP can improve the generalization over noise robust training approaches on CIFAR dataset by ∼ 3.2% on average. Further, we observe generalization improvements of 2.31% for a Vision Transformer model trained on naturally corrupted Clothing1M.

For agents

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

get_harvested_code_for_paper("aaai_33972")
get_code_for_paper("aaai_33972")
have("aaai_33972")

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

curl https://syntology.ai/api/ran/aaai_33972.json

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

[![Syntology run record](https://syntology.ai/api/ran/aaai_33972.svg)](https://syntology.ai/paper/aaai_33972)

Connect an agent — have() is free.