SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1905.12019 · 2019

Unified Probabilistic Deep Continual Learning through Generative Replay and Open Set Recognition

arXiv · PDF · Open in the Atlas

Code that ran

12 of 13 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 11 have no contract verdict; 0 failures of our instrument; 1 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/1905.12019.json · badge: /api/ran/1905.12019.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
MrtnMndt/OCDVAEContinualLearning canonical (classifier label)
classifier basis: census
11 ran · 1 no recorded run
TimmHess/OCDVAEContinualLearning community (classifier label)
classifier basis not recorded on the link
1 ran · 0 no recorded run
FunctionStatusWhere it lives
KLD Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/loss_functions.py
code served (permissive licence) · get_code("35a868da9fed37c9")
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)

# Construct tiny tensors for mu and std
# mu and std are used in arithmetic operations and torch.sum, torch.log, torch.numel
# Keep them small: batch_size=2, dim=4
mu = torch.randn(2, 4)
std = torch.randn(2, 4).abs() + 0.1  # ensure positive values for log stability

ARGS = {
    "mu": mu,
    "std": std,
}
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
augment_data Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/augmentation.py
code served (permissive licence) · get_code("acecd7a6217e354d")
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 4, dtype family float, shape (2, 3, 4, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import random

torch.manual_seed(0)

class Args:
    def __init__(self):
        self.hflip_p = 0.5
        self.vflip_p = 0.3
        self.random_translation = True

# Create a small batch of images
# Shape: (batch_size, channels, height, width) = (2, 3, 4, 5)
batch = torch.randn(2, 3, 4, 5)

args = Args()

ARGS = {
    "batch": batch,
    "args": 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 Tensor, ndim 4, dtype family float, shape (2, 3, 4, 5) — shape (2, 3, 4, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
augment_flip Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/augmentation.py
code served (permissive licence) · get_code("4333c3e5ee86bb81")
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 4, dtype family float, shape (2, 3, 4, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import random

torch.manual_seed(0)

# Create a small image tensor
# Shape: (batch=2, channels=3, height=4, width=5)
# The function uses torch.flip with dims [2] and [1], so we need at least 3 dimensions
img = torch.randn(2, 3, 4, 5)

# Probability parameters for horizontal and vertical flip
hflip_p = 0.5
vflip_p = 0.3

ARGS = {
    "img": img,
    "hflip_p": hflip_p,
    "vflip_p": vflip_p,
}
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 4, dtype family float, shape (2, 3, 4, 5) — shape (2, 3, 4, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
augment_random_translate Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/augmentation.py
code served (permissive licence) · get_code("77b460b279b0da2d")
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 3, dtype family float, shape (2, 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
import random

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

# Create a small image tensor: (batch=2, height=4, width=3, channels=1)
# The function expects img.size(1) to be height, img.size(2) to be width
img = torch.randn(2, 4, 3)

ARGS = {
    "img": img,
    "random_factor": 0.1
}
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 3, dtype family float, shape (2, 4, 3) — shape (2, 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
calc_distances_to_means Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/OpenSet/meta_recognition.py
code served (permissive licence) · get_code("4bf4bed0805b9389")
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 2, 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)

# Create small tensors for testing
# means: list of tensors representing class means
means = [
    torch.randn(4),  # mean for class 0
    torch.randn(4),  # mean for class 1
]

# tensors: list of tensors representing individual samples
tensors = [
    torch.randn(4),  # sample z for class 0
    torch.randn(4),  # sample z for class 1
]

# distance_function: string specifying which distance metric to use
distance_function = 'cosine'

ARGS = {
    'means': means,
    'tensors': tensors,
    'distance_function': distance_function,
}
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, ndim 2, 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
encoder_loss_with_fake Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/loss_functions.py
code served (permissive licence) · get_code("b3576c10e8d51fce")
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
import torch.nn as nn

torch.manual_seed(0)

# Define a minimal args object with the margin attribute
class Args:
    def __init__(self):
        self.margin = 10.0

args = Args()

# Define device
device = torch.device('cpu')

# Create tiny tensors for the function inputs
batch_size = 2
num_classes = 4
feature_dim = 4
input_dim = 3

# Classification output and target
output_samples_classification = torch.randn(batch_size, num_classes)
target = torch.randint(0, num_classes, (batch_size,))

# Reconstruction output and input
output_samples_recon = torch.sigmoid(torch.randn(batch_size, input_dim))
inp = torch.randint(0, 2, (batch_size, input_dim)).float()

# Mean and std for real, reconstructed, and fake samples
real_mu = torch.randn(batch_size, feature_dim)
real_std = torch.abs(torch.randn(batch_size, feature_dim)) + 0.1

rec_mu = torch.randn(batch_size, feature_dim)
rec_std = torch.abs(torch.randn(batch_size, feature_dim)) + 0.1

fake_mu = torch.randn(batch_size, feature_dim)
fake_std = torch.abs(torch.randn(batch_size, feature_dim)) + 0.1

# Construct the ARGS dictionary
ARGS = {
    'output_samples_classification': output_samples_classification,
    'target': target,
    'output_samples_recon': output_samples_recon,
    'inp': inp,
    'real_mu': real_mu,
    'real_std': real_std,
    'rec_mu': rec_mu,
    'rec_std': rec_std,
    'fake_mu': fake_mu,
    'fake_std': fake_std,
    'device': device,
    'args': 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 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
eval_dataset Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/evaluate.py
code served (permissive licence) · get_code("5bb73bb78d01fb53")
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 dict. 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)

# Build a minimal model that satisfies the function's usage
class MinimalVAE(nn.Module):
    def __init__(self, input_dim=4, latent_dim=3, num_classes=2):
        super().__init__()
        self.input_dim = input_dim
        self.latent_dim = latent_dim
        self.num_classes = num_classes
        
        # Encoder
        self.encoder_fc = nn.Linear(input_dim, latent_dim * 2)
        
        # Classifier
        self.classifier_fc = nn.Linear(latent_dim, num_classes)
        
        # Decoder
        self.decoder_fc = nn.Linear(latent_dim, input_dim)
        
        # PixelCNN for autoregressive decoding
        self.pixelcnn = nn.Linear(input_dim, input_dim)
    
    def encode(self, x):
        # x shape: (batch, input_dim)
        x_flat = x.view(x.size(0), -1)
        h = self.encoder_fc(x_flat)
        mu = h[:, :self.latent_dim]
        sigma = torch.exp(h[:, self.latent_dim:])
        return mu, sigma
    
    def reparameterize(self, mu, sigma):
        eps = torch.randn_like(mu)
        z = mu + sigma * eps
        return z
    
    def classifier(self, z):
        return self.classifier_fc(z)
    
    def decode(self, z):
        return self.decoder_fc(z).view(z.size(0), 1, 2, 2)
    
    def pixelcnn(self, inputs, priors):
        return self.pixelcnn(inputs.view(inputs.size(0), -1)).view(inputs.size(0), 1, 2, 2)

class WrappedModel(nn.Module):
    def __init__(self, vae):
        super().__init__()
        self.module = vae
    
    def forward(self, x):
        return self.module(x)

# Create model
vae = MinimalVAE(input_dim=4, latent_dim=3, num_classes=2)
model = WrappedModel(vae)

# Create minimal data loader
# Each batch should have shape (batch_size, input_dim) with some structure
data = [
    (torch.randn(2, 1, 2, 2), torch.tensor([0, 1])),
    (torch.randn(2, 1, 2, 2), torch.tensor([1, 0])),
]

ARGS = {
    'model': model,
    'data_loader': data,
    'num_classes': 2,
    'device': 'cpu',
    'samples': 1,
    'calc_reconstruction': False,
    'autoregression': 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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
eval_openset_dataset Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/evaluate.py
code served (permissive licence) · get_code("ff4c1bd619f2a04b")
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 dict. 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 numpy as np
import math

torch.manual_seed(0)

# Define a minimal model that satisfies all the method calls in eval_openset_dataset
class MinimalVAEModel(nn.Module):
    def __init__(self, num_classes=3, latent_dim=4, input_channels=1, input_size=8):
        super().__init__()
        self.num_classes = num_classes
        self.latent_dim = latent_dim
        self.input_channels = input_channels
        self.input_size = input_size
        
        # Encoder
        self.encoder_fc = nn.Linear(input_channels * input_size * input_size, 16)
        self.encoder_mu = nn.Linear(16, latent_dim)
        self.encoder_sigma = nn.Linear(16, latent_dim)
        
        # Classifier
        self.classifier_fc = nn.Linear(latent_dim, num_classes)
        
        # Decoder
        self.decoder_fc1 = nn.Linear(latent_dim, 16)
        self.decoder_fc2 = nn.Linear(16, input_channels * input_size * input_size)
        
        # PixelCNN stub for autoregressive mode
        self.pixelcnn = nn.Linear(1, 1)
    
    def encode(self, x):
        batch_size = x.size(0)
        flat_x = x.view(batch_size, -1)
        h = torch.relu(self.encoder_fc(flat_x))
        mu = self.encoder_mu(h)
        sigma = torch.exp(self.encoder_sigma(h))
        return mu, sigma
    
    def reparameterize(self, mu, sigma):
        eps = torch.randn_like(mu)
        z = mu + sigma * eps
        return z
    
    def classifier(self, z):
        logits = self.classifier_fc(z)
        return logits
    
    def decode(self, z):
        h = torch.relu(self.decoder_fc1(z))
        recon = self.decoder_fc2(h)
        recon = recon.view(z.size(0), self.input_channels, self.input_size, self.input_size)
        return recon

class ModelWrapper(nn.Module):
    def __init__(self, model):
        super().__init__()
        self.module = model

# Create data loader as list of (inputs, classes) tuples
def create_data_loader(batch_size=2, num_batches=1, input_channels=1, input_size=8, num_classes=3):
    data_loader = []
    for _ in range(num_batches):
        inputs = torch.randn(batch_size, input_channels, input_size, input_size)
        classes = torch.randint(0, num_classes, (batch_size,))
        data_loader.append((inputs, classes))
    return data_loader

# Setup arguments
base_model = MinimalVAEModel(num_classes=3, latent_dim=4, input_channels=1, input_size=8)
model = ModelWrapper(base_model)
data_loader = create_data_loader(batch_size=2, num_batches=1, input_channels=1, input_size=8, num_classes=3)

ARGS = {
    'model': model,
    'data_loader': data_loader,
    'num_classes': 3,
    'device': 'cpu',
    'samples': 1,
    'calc_reconstruction': False,
    'autoregression': 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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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_feat_size Ran
ran, honoured the contract we drafted
MrtnMndt/OCDVAEContinualLearning/lib/Models/architectures.py
code served (permissive licence) · get_code("8b8425aa050b8c28")
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 int; 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

torch.manual_seed(0)

# Create a simple block that processes input and outputs feature maps
class SimpleBlock(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, padding=1, stride=2)
    
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        return x

block = SimpleBlock()

ARGS = {
    "block": block,
    "spatial_size": 4,
    "ncolors": 3
}
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 int — 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_latent_embedding Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/Training/evaluate.py
code served (permissive licence) · get_code("f28bb4760db0af78")
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)

# Build a minimal model with encode and reparameterize methods
class SimpleVAE(nn.Module):
    def __init__(self, input_dim=4, latent_dim=3):
        super().__init__()
        self.encoder = nn.Linear(input_dim, latent_dim * 2)
        self.input_dim = input_dim
        self.latent_dim = latent_dim
    
    def encode(self, x):
        h = self.encoder(x)
        mu = h[:, :self.latent_dim]
        logstd = h[:, self.latent_dim:]
        std = torch.exp(logstd)
        return mu, std
    
    def reparameterize(self, mu, std):
        eps = torch.randn_like(std)
        z = mu + eps * std
        return z

# Wrap model to match the function's use of model.module
class ModelWrapper(nn.Module):
    def __init__(self, vae):
        super().__init__()
        self.module = vae

# Create the VAE model
vae = SimpleVAE(input_dim=4, latent_dim=3)
model = ModelWrapper(vae)

# Create minimal data loader: list of (inputs, classes) tuples
# Batch size 2, input dim 4, 2 samples per class, 3 classes
data_loader = [
    (torch.randn(2, 4), torch.tensor([0, 1])),
    (torch.randn(2, 4), torch.tensor([2, 0])),
]

num_classes = 3
device = 'cpu'

ARGS = {
    'model': model,
    'data_loader': data_loader,
    'num_classes': num_classes,
    'device': device,
}
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
get_means Ran
ran on a synthesized input
MrtnMndt/OCDVAEContinualLearning/lib/OpenSet/meta_recognition.py
code served (permissive licence) · get_code("9a2c9f801548b8b5")
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 2, 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)

# Create a list of tensors representing per-class z values
# Each tensor has shape (batch_size, feature_dim) with distinct dimensions
tensor_1 = torch.randn(2, 4)  # 2 samples, 4 features
tensor_2 = torch.randn(3, 4)  # 3 samples, 4 features
tensor_3 = torch.randn(2, 4)  # 2 samples, 4 features

tensors_list = [tensor_1, tensor_2, tensor_3]

ARGS = {
    "tensors_list": tensors_list
}
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, ndim 2, 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
surrogate_loss Ran
ran on a synthesized input
TimmHess/OCDVAEContinualLearning/lib/Models/si.py
code served (permissive licence) · get_code("41cc1bfa6d07666f")
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)

# Create a minimal model with named parameters
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.layer1 = nn.Linear(4, 3)
        self.layer2 = nn.Linear(3, 2)
    
    def forward(self, x):
        return self.layer2(self.layer1(x))

model = SimpleModel()

# Create a storage_unit class that mimics the expected interface
class StorageUnit:
    def __init__(self):
        self._device_val = 'cpu'
        
        # For each parameter in the model, create SI_prev_task and SI_omega attributes
        # Replace '.' with '__' in parameter names as the function does
        self.layer1__weight_SI_prev_task = torch.randn(3, 4)
        self.layer1__weight_SI_omega = torch.abs(torch.randn(3, 4))
        self.layer1__bias_SI_prev_task = torch.randn(3)
        self.layer1__bias_SI_omega = torch.abs(torch.randn(3))
        self.layer2__weight_SI_prev_task = torch.randn(2, 3)
        self.layer2__weight_SI_omega = torch.abs(torch.randn(2, 3))
        self.layer2__bias_SI_prev_task = torch.randn(2)
        self.layer2__bias_SI_omega = torch.abs(torch.randn(2))
    
    def _device(self):
        return self._device_val

storage_unit = StorageUnit()

ARGS = {
    'model': model,
    'storage_unit': storage_unit
}
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
joint_loss_function Not yet run
unverified (harvested, no recorded run)
MrtnMndt/OCDVAEContinualLearning/lib/Training/loss_functions.py
code served (permissive licence) · get_code("13389065de3e98b6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 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 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
get_feat_size 1 other paper
1 later
2105.08997 (2021-05)
When Deep Classifiers Agree: Analyzing Correlations between Learning Order and Image Statistics

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

Modern deep neural networks are well known to be brittle in the face of unknown data instances and recognition of the latter remains a challenge. Although it is inevitable for continual-learning systems to encounter such unseen concepts, the corresponding literature appears to nonetheless focus primarily on alleviating catastrophic interference with learned representations. In this work, we introduce a probabilistic approach that connects these perspectives based on variational inference in a single deep autoencoder model. Specifically, we propose to bound the approximate posterior by fitting regions of high density on the basis of correctly classified data points. These bounds are shown to serve a dual purpose: unseen unknown out-of-distribution data can be distinguished from already trained known tasks towards robust application. Simultaneously, to retain already acquired knowledge, a generative replay process can be narrowed to strictly in-distribution samples, in order to significantly alleviate catastrophic interference.

For agents

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

get_harvested_code_for_paper("1905.12019")
get_code_for_paper("1905.12019")
have("1905.12019")

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

curl https://syntology.ai/api/ran/1905.12019.json

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

[![Syntology run record](https://syntology.ai/api/ran/1905.12019.svg)](https://syntology.ai/paper/1905.12019)

Connect an agent — have() is free.