SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2309.14062 · NeurIPS · 2023

FeCAM: Exploiting the Heterogeneity of Class Distributions in Exemplar-Free Continual Learning

Bartłomiej Twardowski, Joost Van De Weijer, Dipam Goswami, Yuyang Liu

arXiv · PDF · Open in the Atlas

Code that ran

15 of 21 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 4 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 7 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/2309.14062.json · badge: /api/ran/2309.14062.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
dipamgoswami/fecam canonical (classifier label)
classifier basis: census
9 ran · 4 no recorded run
dipamgoswami/FeCAM canonical (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
gmum/hyperinterval community (classifier label)
classifier basis not recorded on the link
3 ran · 1 no recorded run
gmum/hint community
Papers with Code archive
1 ran · 1 no recorded run
FunctionStatusWhere it lives
IntervalModuleWithWeights Ran
ran on a synthesized input · constructed an object; no result was computed
gmum/hint/IntervalNets/interval_modules.py
code served (permissive licence) · get_code("4d1ed8dab640e691")
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 an instance of its own class, IntervalModuleWithWeights: the run constructed the object and did not call it, so no result was computed. 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)

# Since IntervalModuleWithWeights.__init__ takes only self,
# there are no additional parameters to configure
ARGS = {}
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 IntervalModuleWithWeights — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a IntervalModuleWithWeights 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_common_cov Ran
ran, honoured the contract we drafted
dipamgoswami/FeCAM/utils/maha_utils.py
code served (permissive licence) · get_code("c9427cae68d6444f")
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 (3, 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
import numpy as np
import torch.nn as nn
from torch.utils.data import DataLoader

torch.manual_seed(0)

class StandInModel(nn.Module):
    def __init__(self, dataset='cifar10', tukey=False):
        super().__init__()
        self.args = {
            'dataset': dataset,
            'tukey': tukey
        }
        self.linear = nn.Linear(4, 3)
    
    def _extract_vectors(self, train_loader):
        # Returns vectors of shape (batch_size, feature_dim) and labels
        vectors_list = []
        labels_list = []
        for x, y in train_loader:
            batch_vectors = self.linear(x)
            vectors_list.append(batch_vectors.detach().numpy())
            labels_list.append(y.numpy())
        vectors = np.vstack(vectors_list)
        labels = np.hstack(labels_list)
        return vectors, labels
    
    def _extract_vectors_common_cov(self, train_loader):
        # For imagenet100 dataset
        vectors_list = []
        for x, y in train_loader:
            batch_vectors = self.linear(x)
            vectors_list.append(batch_vectors.detach())
        vectors = torch.vstack(vectors_list)
        cov = torch.tensor(np.cov(vectors.T))
        return cov
    
    def _tukeys_transform(self, vectors):
        # Simple stand-in transformation
        return vectors * 1.1

# Create small dummy data loader
batch_size = 2
feature_dim = 4
num_batches = 2

data = []
for _ in range(num_batches):
    x = torch.randn(batch_size, feature_dim)
    y = torch.randint(0, 10, (batch_size,))
    data.append((x, y))

train_loader = data

# Create model with non-imagenet100 dataset
model = StandInModel(dataset='cifar10', tukey=False)

ARGS = {
    'train_loader': train_loader,
    'model': model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, 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 (3, 3) — shape (3, 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
compute_new_common_cov Ran
ran, honoured the contract we drafted
dipamgoswami/FeCAM/utils/maha_utils.py
code served (permissive licence) · get_code("1de4ff97b038199a")
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, 4) 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
import torch.nn as nn
from torch.utils.data import DataLoader

torch.manual_seed(0)

class StandInModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.args = {
            'dataset': 'cifar10',
            'tukey': False,
            'shrink': True
        }
        self._known_classes = 5
        self._total_classes = 10
        self._common_cov = torch.eye(4)
        
    def _extract_vectors(self, train_loader):
        # Return vectors of shape (batch_size * num_batches, feature_dim)
        # and labels
        vectors = torch.randn(4, 4)  # 4 samples, 4 features
        labels = torch.tensor([0, 1, 0, 1])
        return vectors, labels
    
    def _tukeys_transform(self, vectors):
        return vectors
    
    def shrink_cov(self, cov):
        # Simple shrinkage towards identity
        identity = torch.eye(cov.shape[0])
        return 0.9 * cov + 0.1 * identity

# Create a minimal train_loader
# DataLoader expects an iterable of (x, y) tuples
train_data = [
    (torch.randn(2, 4), torch.tensor([0, 1])),
    (torch.randn(2, 4), torch.tensor([1, 0]))
]
train_loader = train_data

# Create model instance
model = StandInModel()

ARGS = {
    'train_loader': train_loader,
    'model': model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, 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, 4) — shape (4, 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
conv1x1 Ran
ran; our contract draft was wrong, not the code
dipamgoswami/fecam/convs/resnet.py
code served (permissive licence) · get_code("d9def42110729a85")
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 Conv2d. 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 conv1x1 takes in_planes, out_planes, and stride as parameters
# It returns an nn.Conv2d layer configured for 1x1 convolution
# We need to provide valid integer arguments for these parameters

ARGS = {
    "in_planes": 3,
    "out_planes": 64,
    "stride": 1
}
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 Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d 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
conv1x1 Ran
ran; our contract draft was wrong, not the code
dipamgoswami/fecam/convs/modified_represnet.py
pointer only (licence: MIT); the stored copy get_code serves, from sajaddarabi/BiRealNet, is pointer only (licence: NONE) · get_code("d5cd7ffe2dc51b21")
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 sajaddarabi/BiRealNet, 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.

conv3x3 Ran
ran; our contract draft was wrong, not the code
dipamgoswami/fecam/convs/resnet.py
code served (permissive licence) · get_code("160bb14bd76201b4")
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 Conv2d. 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 = {
    "in_planes": 3,
    "out_planes": 64,
    "stride": 1,
    "groups": 1,
    "dilation": 1,
}
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 Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d 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
conv3x3 Ran
ran; our contract draft was wrong, not the code
dipamgoswami/fecam/convs/resnet_cbam.py
pointer only (licence: MIT); the stored copy get_code serves, from lixiaotong97/DSU, is pointer only (licence: NONE) · get_code("fac5364e2f53c6db")
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 lixiaotong97/DSU, 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.

conv3x3 Ran
ran; our contract draft was wrong, not the code
dipamgoswami/fecam/convs/modified_represnet.py
code served (permissive licence) · get_code("1907f2ae25449f39")
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 Conv2d. 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 = {
    "in_planes": 3,
    "out_planes": 64,
    "stride": 1,
}
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 Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d 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_current_targets Ran
ran on a synthesized input
gmum/hyperinterval/Utils/hnet_interval_regularizer.py
code served (permissive licence) · get_code("e488b43331809f71")
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 5, 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
import torch.nn as nn
from typing import Dict, List

torch.manual_seed(0)

class MockHypernetwork(nn.Module):
    """Mock hypernetwork that implements the interface used by get_current_targets."""
    
    def __init__(self, num_tasks: int = 3):
        super().__init__()
        self.num_tasks = num_tasks
        self._prev_hnet_weights = torch.randn(5, 4)
        self.linear = nn.Linear(4, 3)
        
    def forward(self, cond_id: List[int], ret_format: str = 'sequential', 
                perturbated_eps: float = 0.0, weights: Dict = None, 
                return_extended_output: bool = False):
        """
        Mock forward pass that returns weights in the expected format.
        
        Returns:
        - W_lower: list of lists of tensors (one per task in cond_id)
        - W_middle: list of lists of tensors (one per task in cond_id)
        - W_upper: list of lists of tensors (one per task in cond_id)
        - dummy: placeholder for fourth return value
        """
        num_conds = len(cond_id)
        
        # Create weight tensors for each task
        W_lower = [[torch.randn(3, 2) for _ in range(2)] for _ in range(num_conds)]
        W_middle = [[torch.randn(3, 2) for _ in range(2)] for _ in range(num_conds)]
        W_upper = [[torch.randn(3, 2) for _ in range(2)] for _ in range(num_conds)]
        
        return W_lower, W_middle, W_upper, None


# Construct arguments
task_id = 2
hnet = MockHypernetwork(num_tasks=3)
eps = 0.01

ARGS = {
    'task_id': task_id,
    'hnet': hnet,
    'eps': eps,
}
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 5, 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
get_current_targets Ran
ran on a synthesized input
gmum/hyperinterval/Utils/hnet_middle_regularizer.py
code served (permissive licence) · get_code("d8e7aca624e3e8de")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list. 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)

class SimpleHypernetwork(nn.Module):
    """Minimal hypernetwork that returns sequential output format."""
    
    def __init__(self, num_tasks=3):
        super().__init__()
        self.num_tasks = num_tasks
        self.linear = nn.Linear(10, 12)
    
    def forward(self, cond_id, ret_format='sequential', perturbated_eps=0.0, return_extended_output=False):
        """
        Simulates hypernetwork forward pass.
        
        Args:
            cond_id: list of task ids
            ret_format: format of return (should be 'sequential')
            perturbated_eps: perturbation value
            return_extended_output: whether to return extended output
        
        Returns:
            List of lists of tensors, one sublist per task_id
        """
        # Create dummy output: for each task_id, return a list of parameter tensors
        result = []
        for task_idx in cond_id:
            # Each task gets a list of weight tensors (simulating network parameters)
            task_weights = [
                torch.randn(4, 3),  # First layer weights
                torch.randn(5),     # First layer bias
                torch.randn(6, 5),  # Second layer weights
            ]
            result.append(task_weights)
        
        return result

# Construct arguments
hnet = SimpleHypernetwork(num_tasks=3)
hnet.train()  # Set to training mode initially

ARGS = {
    'task_id': 2,
    'hnet': hnet,
    'eps': 0.01
}
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 list — a list 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
normalize_cov Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
dipamgoswami/fecam/FeCAM_vit_core50.py
code served (permissive licence) · get_code("dbac5df5f6dc9433")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 3, 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 numpy as np

# Create a list of small covariance matrices
# Covariance matrices should be square and symmetric
# Using distinct sizes to be clear about dimensions

# Create first covariance matrix (3x3)
cov1 = np.array([
    [2.0, 0.5, 0.3],
    [0.5, 1.5, 0.2],
    [0.3, 0.2, 1.0]
], dtype=np.float32)

# Create second covariance matrix (3x3)
cov2 = np.array([
    [3.0, 0.8, 0.1],
    [0.8, 2.0, 0.4],
    [0.1, 0.4, 1.5]
], dtype=np.float32)

# Create third covariance matrix (3x3)
cov3 = np.array([
    [1.5, 0.2, 0.0],
    [0.2, 2.5, 0.6],
    [0.0, 0.6, 0.8]
], dtype=np.float32)

cov_mat_list = [cov1, cov2, cov3]

ARGS = {
    "cov_mat": cov_mat_list
}
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 list, ndim 3, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_logits Ran
ran on a synthesized input
gmum/hyperinterval/IntervalNets/interval_modules.py
code served (permissive licence) · get_code("a14c0ad5907d1cd5")
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 map. 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
from torch import Tensor

torch.manual_seed(0)

# Create a tensor of shape (batch_size=2, 3, tensor_dimension=4)
# The tensor needs to have a named dimension 'bounds' for the second dimension
x = torch.randn(2, 3, 4)
x = x.rename(None, 'bounds', None)  # Name the second dimension 'bounds'

ARGS = {
    'x': x
}
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 map — a map output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a map 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_rep Ran
ran; our contract draft was wrong, not the code
dipamgoswami/fecam/convs/modified_represnet.py
pointer only (licence: MIT); the stored copy get_code serves, from jinglin-liang/OLAS, is pointer only (licence: NONE) · get_code("ecaa7ca05d715804")
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 jinglin-liang/OLAS, 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.

shrink_cov Ran
ran, honoured the contract we drafted
dipamgoswami/fecam/FeCAM_vit_core50.py
code served (permissive licence) · get_code("17ed71f058d7674a")
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 (4, 4) 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

# Create a small sample covariance matrix
# A covariance matrix should be square and symmetric
np.random.seed(0)
cov_matrix = np.random.randn(4, 4)
cov_matrix = (cov_matrix + cov_matrix.T) / 2  # Make it symmetric

ARGS = {
    "cov": cov_matrix
}
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 (4, 4) — shape (4, 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
shrink_cov Ran
ran, honoured the contract we drafted
dipamgoswami/fecam/FeCAM_vit_imagenet-r.py
code served (permissive licence) · get_code("0365639e81fbf382")
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 (4, 4) 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 a small covariance matrix for testing
# A 4x4 symmetric positive semi-definite matrix
cov = np.array([
    [2.0, 0.5, 0.3, 0.1],
    [0.5, 1.5, 0.2, 0.4],
    [0.3, 0.2, 1.8, 0.2],
    [0.1, 0.4, 0.2, 1.2]
], dtype=np.float64)

ARGS = {
    "cov": cov
}
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 (4, 4) — shape (4, 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
IntervalLinear Not yet run
unverified (harvested, no recorded run)
gmum/hint/IntervalNets/interval_modules.py
code served (permissive licence) · get_code("5ac470f50d929fcf")
How it ran

no invocation recorded.

get_dataset Not yet run
unverified (harvested, no recorded run)
dipamgoswami/fecam/FeCAM_vit_imagenet-r.py
code served (permissive licence) · get_code("f5c8f70797dc3539")
How it ran

no invocation recorded.

prepare_subset_imagenet_tasks Not yet run
unverified (harvested, no recorded run)
gmum/hyperinterval/DatasetHandlers/subset_image_net.py
code served (permissive licence) · get_code("a740ad44865dbc9b")
How it ran

no invocation recorded.

resnet18 Not yet run
unverified (harvested, no recorded run)
dipamgoswami/fecam/convs/resnet.py
code served (permissive licence) · get_code("966d18702829b1c0")
How it ran

no invocation recorded.

resnet18_cbam Not yet run
unverified (harvested, no recorded run)
dipamgoswami/fecam/convs/resnet_cbam.py
pointer only (licence: MIT); the stored copy get_code serves, from jinglin-liang/OLAS, is pointer only (licence: NONE) · get_code("9d7277889d69cf2d")
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 jinglin-liang/OLAS, 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. No invocation recorded.

resnet34_cbam Not yet run
unverified (harvested, no recorded run)
dipamgoswami/fecam/convs/resnet_cbam.py
pointer only (licence: MIT); the stored copy get_code serves, from jinglin-liang/OLAS, is pointer only (licence: NONE) · get_code("9eb3d43ee2e8ae7d")
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 jinglin-liang/OLAS, 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. No invocation recorded.

Also present in other papers' code

Of 21 samples lifted from this paper's linked repositories, 21 carry a function identity and 0 do not. Of the 21 distinct functions with an identity, 9 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 4 the same month, 9 later) and 12 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
conv3x3 686 other papers
511 dated earlier, 5 the same month, 163 later, 7 same year, order not determinable · 31 only as an assembly brick
1406.2661 (2014-06)
Generative Adversarial Networks
1409.0575 (2014-09)
ImageNet Large Scale Visual Recognition Challenge
1412.6572 (2014-12)
Explaining and Harnessing Adversarial Examples
and 683 more not named here
conv1x1 431 other papers
284 dated earlier, 3 the same month, 135 later, 8 same year, order not determinable, 1 undated · 38 only as an assembly brick
1409.1556 (2014-09)
Very Deep Convolutional Networks for Large-Scale Image Recognition
1506.01186 (2015-06)
Cyclical Learning Rates for Training Neural Networks
1507.05717 (2015-07)
An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition
and 428 more not named here
conv3x3 312 other papers
185 dated earlier, 4 the same month, 115 later, 7 same year, order not determinable, 1 undated · 31 only as an assembly brick
1409.1556 (2014-09)
Very Deep Convolutional Networks for Large-Scale Image Recognition
1507.05717 (2015-07)
An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition
1512.03385 (2015-12)
Deep Residual Learning for Image Recognition
and 309 more not named here
resnet18 95 other papers
53 dated earlier, 40 later, 1 same year, order not determinable, 1 undated
1805.00907 (2018-05)
Glow: Graph Lowering Compiler Techniques for Neural Networks
1902.08153 (2019-02)
Learned Step Size Quantization
1904.01355 (2019-04)
FCOS: Fully Convolutional One-Stage Object Detection
and 92 more not named here
conv3x3 74 other papers
47 dated earlier, 1 the same month, 26 later · 5 only as an assembly brick
1605.07146 (2016-05)
Wide Residual Networks
1707.01629 (2017-07)
Dual Path Networks
1802.05668 (2018-02)
Model compression via distillation and quantization
and 71 more not named here
resnet34_cbam 11 other papers
11 later
2310.15171 (2023-10)
RoboDepth: Robust Out-of-Distribution Depth Estimation under Corruptions
aaai_29079 (2024 (year only))
Dynamic Sub-graph Distillation for Robust Semi-supervised Continual Learning
2403.14101 (2024-03)
Text-Enhanced Data-free Approach for Federated Class-Incremental Learning
and 8 more not named here
resnet18_cbam 10 other papers
1 dated earlier, 9 later
2308.09297 (2023-08)
NAPA-VQ: Neighborhood Aware Prototype Augmentation with Vector Quantization for Continual Learning
2310.15171 (2023-10)
RoboDepth: Robust Out-of-Distribution Depth Estimation under Corruptions
aaai_29079 (2024 (year only))
Dynamic Sub-graph Distillation for Robust Semi-supervised Continual Learning
and 7 more not named here
conv1x1 9 other papers
3 dated earlier, 6 later
1808.00278 (2018-08)
Bi-Real Net: Enhancing the Performance of 1-bit CNNs With Improved Representational Capability and Advanced Training Algorithm
1812.11800 (2018-12)
Regularized Binary Network Training
2108.05722 (2021-08)
MT-ORL: Multi-Task Occlusion Relationship Learning
and 6 more not named here
resnet18_rep 6 other papers
6 later
aaai_29079 (2024 (year only))
Dynamic Sub-graph Distillation for Robust Semi-supervised Continual Learning
2403.14101 (2024-03)
Text-Enhanced Data-free Approach for Federated Class-Incremental Learning
2404.00986 (2024-04)
Make Continual Learning Stronger via C-Flat
and 3 more not named here

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

Exemplar-free class-incremental learning (CIL) poses several challenges since it prohibits the rehearsal of data from previous tasks and thus suffers from catastrophic forgetting. Recent approaches to incrementally learning the classifier by freezing the feature extractor after the first task have gained much attention. In this paper, we explore prototypical networks for CIL, which generate new class prototypes using the frozen feature extractor and classify the features based on the Euclidean distance to the prototypes. In an analysis of the feature distributions of classes, we show that classification based on Euclidean metrics is successful for jointly trained features. However, when learning from non-stationary data, we observe that the Euclidean metric is suboptimal and that feature distributions are heterogeneous. To address this challenge, we revisit the anisotropic Mahalanobis distance for CIL. In addition, we empirically show that modeling the feature covariance relations is better than previous attempts at sampling features from normal distributions and training a linear classifier. Unlike existing methods, our approach generalizes to both many-and few-shot CIL settings, as well as to domain-incremental settings. Interestingly, without updating the backbone network, our method obtains state-ofthe-art results on several standard continual learning benchmarks. Code is available at https://github.com/dipamgoswami/FeCAM. This paper investigates methods to enhance the representation of class prototypes in CIL, aiming to improve plasticity within the stability-favoring classifier-incremental setting. A standard practice in few-shot CIL [31,42,76,70,72] is to obtain the feature embeddings of new class samples and average them to generate class-wise prototypes. The test image features are then classified by computing the Euclidean distance to the mean prototypes. The Euclidean distance is used in the NCM classifier, following [17], which claims that the highly non-linear nature of learned representations eliminates the need to learn the Mahalanobis [65,11] metric previously used [38]. Our analysis shows that this holds true for classes that are considered during training, however, for new classes, the Euclidean distance is suboptimal. To address this problem, we propose to use the anisotropic Mahalanobis distance. In Fig. 1, we explain how the feature representations vary in CIL settings. Here, the highstability case in CIL is explored, where the model does not achieve spherical representations for new classes in the feature space, unlike joint training. Thus, it is intuitive to take into account the feature covariances while computing the distance. The covariance relations between the feature dimensions better captures the more complex class structure in the high-dimensional feature space. Additionally, in Fig. 3, we analyze singular values for old and new class features to observe the changes in variances in their feature distributions, suggesting a shift towards more anisotropic representations. While previous methods [38] proposed learning Mahalanobis metrics, we propose using an optimal Bayes classifier by modeling the covariance relations of the features and employing class prototypes. We term this approach Feature Covariance-Aware Metric (FeCAM). We compute the covariance matrix for each class from the feature embeddings corresponding to training samples and perform correlation normalization to ensure similar variances across all class representations, which is crucial for distance comparisons. We investigate various ways of using covariance relations in continual settings. We posit that utilizing a Bayes classifier enables better learning of optimal decision boundaries compared to previous attempts [67] involving feature sampling from Gaussian distributions and training linear classifiers. The proposed approach is simple to implement and requires no training since we employ a Bayes classifier. The Bayes classifier FeCAM can be used for both many-shot CIL and few-shot CIL, unlike existing methods. Additionally, we achieve superior performance with pretrained models on both class-incremental and domain-incremental benchmarks. Many-shot class-incremental learning (MSCIL) is the conventional setting, where sufficient training data is available for all classes. A critical aspect in many-shot CIL methods is the semantic drift in

For agents

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

get_harvested_code_for_paper("2309.14062")
get_code_for_paper("2309.14062")
have("2309.14062")

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

curl https://syntology.ai/api/ran/2309.14062.json

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

[![Syntology run record](https://syntology.ai/api/ran/2309.14062.svg)](https://syntology.ai/paper/2309.14062)

Connect an agent — have() is free.