SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.11973 · 2023

Continual Learning: Forget-free Winning Subnetworks for Video Representations

arXiv · PDF · Open in the Atlas

Code that ran

11 of 15 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 4 have no contract verdict; 6 failures of our instrument; 4 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2312.11973.json · badge: /api/ran/2312.11973.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
ihaeyong/pfnr canonical (classifier label)
classifier basis: classifier_2025
2 ran · 1 no recorded run
ihaeyong/pnr community (classifier label)
classifier basis: plan_row
9 ran · 3 no recorded run
FunctionStatusWhere it lives
ActivationLayer Ran
ran; our contract draft was wrong, not the code
ihaeyong/pnr/model_nerv.py
pointer only (licence: MIT); the stored copy get_code serves, from haochen-rye/nerv, is pointer only (licence: NONE) · get_code("878fd412e7a78234")
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 haochen-rye/nerv, 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.

MLP Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ihaeyong/pnr/model_nerv.py
pointer only (licence: MIT); the stored copy get_code serves, from haochen-rye/nerv, is pointer only (licence: NONE) · get_code("4b4bad794e3e1cab")
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 haochen-rye/nerv, 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.

NormLayer Ran
ran; our contract draft was wrong, not the code
ihaeyong/pnr/model_nerv.py
pointer only (licence: MIT); the stored copy get_code serves, from haochen-rye/nerv, is pointer only (licence: NONE) · get_code("853ab1b19577355a")
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 haochen-rye/nerv, 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.

compute_conv_output_size Ran
ran, honoured the contract we drafted
ihaeyong/pnr/model/subnet.py
code served (permissive licence) · get_code("e71c2a2bc2fc8909")
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 int, ndim 0, dtype family int, 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 numpy as np

# This function computes the output size of a convolutional layer
# given input size and convolution parameters.
# All parameters are simple numeric values.

ARGS = {
    "Lin": 28,           # Input length (e.g., 28x28 image)
    "kernel_size": 3,    # Kernel/filter size
    "stride": 1,         # Stride of convolution
    "padding": 1,        # Padding added to input
    "dilation": 1        # Dilation factor
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_consolidated_masks Ran
ran; our contract draft was wrong, not the code
ihaeyong/pfnr/train_nerv.py
code served (permissive licence) · get_code("c24fca3836221111")
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 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 numpy as np
from copy import deepcopy

torch.manual_seed(0)

# Construct per_task_masks - a list of dicts containing masks for different tasks
# The function accesses per_task_masks[task_id] and iterates over keys
# Keys can contain 'real' or 'imag' (which have list values) or other keys (which have tensor values)

per_task_masks = [
    # Task 0 masks
    {
        'weight': torch.tensor([[0.5, 0.3], [0.7, 0.2]]),
        'bias': torch.tensor([0.1, 0.9]),
        'real_part': [torch.tensor([0.4, 0.6]), torch.tensor([0.2, 0.8])],
        'imag_part': [torch.tensor([0.3, 0.7]), torch.tensor([0.5, 0.5])]
    },
    # Task 1 masks
    {
        'weight': torch.tensor([[0.6, 0.4], [0.3, 0.8]]),
        'bias': torch.tensor([0.2, 0.7]),
        'real_part': [torch.tensor([0.5, 0.5]), torch.tensor([0.3, 0.7])],
        'imag_part': [torch.tensor([0.4, 0.6]), torch.tensor([0.6, 0.4])]
    },
    # Task 2 masks
    {
        'weight': torch.tensor([[0.7, 0.2], [0.4, 0.6]]),
        'bias': torch.tensor([0.3, 0.5]),
        'real_part': [torch.tensor([0.6, 0.4]), torch.tensor([0.4, 0.6])],
        'imag_part': [torch.tensor([0.5, 0.5]), torch.tensor([0.7, 0.3])]
    }
]

# Test with task_id = 0 (consolidated_masks should be None initially)
ARGS = {
    'per_task_masks': per_task_masks,
    'task_id': 0,
    'consolidated_masks': None
}
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 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_coused_sparsity Ran
ran on a synthesized input
ihaeyong/pnr/train_nerv_eval.py
code served (permissive licence) · get_code("f0e8e55d92917b22")
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

torch.manual_seed(0)

# Build per_task_masks and consolidated_masks
# The function iterates through per_task_masks[tid] which contains keys with 'real'/'imag' or regular keys
# It accesses per_task_masks[task_id][key][idx] for 'real'/'imag' keys (lists of tensors)
# or per_task_masks[task_id][key] for regular keys (single tensors)

# Create masks for multiple tasks
per_task_masks = {}

# Task 0
per_task_masks[0] = {
    'weight_real': [torch.randn(2, 4), torch.randn(2, 4)],  # List of tensors
    'weight_imag': [torch.randn(2, 4), torch.randn(2, 4)],  # List of tensors
    'bias': torch.randn(3),  # Single tensor
    'last_layer': torch.randn(2, 2),  # Should be skipped (contains 'last')
}

# Task 1
per_task_masks[1] = {
    'weight_real': [torch.randn(2, 4), torch.randn(2, 4)],  # List of tensors
    'weight_imag': [torch.randn(2, 4), torch.randn(2, 4)],  # List of tensors
    'bias': torch.randn(3),  # Single tensor
    'last_layer': torch.randn(2, 2),  # Should be skipped
}

# Task 2
per_task_masks[2] = {
    'weight_real': [torch.randn(2, 4), torch.randn(2, 4)],  # List of tensors
    'weight_imag': [torch.randn(2, 4), torch.randn(2, 4)],  # List of tensors
    'bias': torch.randn(3),  # Single tensor
    'last_layer': torch.randn(2, 2),  # Should be skipped
}

# consolidated_masks is not used in the function body
consolidated_masks = {}

ARGS = {
    'task_id': 2,
    'per_task_masks': per_task_masks,
    'consolidated_masks': consolidated_masks,
}
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 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_reused_sparsity Ran
ran on a synthesized input
ihaeyong/pnr/train_nerv_eval.py
code served (permissive licence) · get_code("49fb8932e40b133e")
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

torch.manual_seed(0)

# Construct per_task_masks: dict of dicts containing lists of tensors and single tensors
per_task_masks = {
    0: {
        'weight_real': [torch.randint(0, 2, (4, 3)).float(), torch.randint(0, 2, (2, 5)).float()],
        'bias_imag': [torch.randint(0, 2, (3,)).float(), torch.randint(0, 2, (4,)).float()],
        'other_param': torch.randint(0, 2, (6, 2)).float(),
        'last_layer': torch.randint(0, 2, (2, 2)).float(),
    },
    1: {
        'weight_real': [torch.randint(0, 2, (4, 3)).float(), torch.randint(0, 2, (2, 5)).float()],
        'bias_imag': [torch.randint(0, 2, (3,)).float(), torch.randint(0, 2, (4,)).float()],
        'other_param': torch.randint(0, 2, (6, 2)).float(),
        'last_layer': torch.randint(0, 2, (2, 2)).float(),
    },
    2: {
        'weight_real': [torch.randint(0, 2, (4, 3)).float(), torch.randint(0, 2, (2, 5)).float()],
        'bias_imag': [torch.randint(0, 2, (3,)).float(), torch.randint(0, 2, (4,)).float()],
        'other_param': torch.randint(0, 2, (6, 2)).float(),
        'last_layer': torch.randint(0, 2, (2, 2)).float(),
    },
}

# Construct consolidated_masks (used as a parameter but not in function body)
consolidated_masks = {
    'weight_real': [torch.randint(0, 2, (4, 3)).float(), torch.randint(0, 2, (2, 5)).float()],
    'bias_imag': [torch.randint(0, 2, (3,)).float(), torch.randint(0, 2, (4,)).float()],
    'other_param': torch.randint(0, 2, (6, 2)).float(),
}

ARGS = {
    'task_id': 1,
    'per_task_masks': per_task_masks,
    'consolidated_masks': consolidated_masks,
}
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 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_task_sparsity Ran
ran; our contract draft was wrong, not the code
ihaeyong/pfnr/train_nerv.py
code served (permissive licence) · get_code("144b98b55047b2da")
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 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

torch.manual_seed(0)

# Create per_task_masks structure
# The function iterates over per_task_masks[task_id] when g_sparsity=False
# or per_task_masks directly when g_sparsity=True
# Each value in the masks dict should be either None or a tensor

# For g_sparsity=False case: per_task_masks is a dict of task_id -> masks_dict
per_task_masks_false = {
    0: {
        'layer1_weight': torch.randint(0, 2, (4, 3)),  # binary mask tensor
        'layer2_weight': torch.randint(0, 2, (5, 4)),
        'layer3_bias': None,
        'last_layer': torch.randint(0, 2, (2, 2)),  # should be skipped due to 'last'
    },
    1: {
        'layer1_weight': torch.randint(0, 2, (4, 3)),
        'layer2_weight': torch.randint(0, 2, (5, 4)),
    }
}

# For g_sparsity=True case: per_task_masks is directly a masks_dict
per_task_masks_true = {
    'layer1_weight': torch.randint(0, 2, (4, 3)),
    'layer2_weight': torch.randint(0, 2, (5, 4)),
    'layer3_bias': None,
    'last_layer': torch.randint(0, 2, (2, 2)),
}

# Test case 1: g_sparsity=False
ARGS = {
    'task_id': 0,
    'per_task_masks': per_task_masks_false,
    'g_sparsity': False,
}
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 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_task_sparsity Ran
ran on a synthesized input
ihaeyong/pnr/train_nerv_eval.py
code served (permissive licence) · get_code("000804ce4ed442ee")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Build per_task_masks structure
# The function expects either:
# - if g_sparsity=False: per_task_masks[task_id] = dict of masks
# - if g_sparsity=True: per_task_masks = dict of masks directly

# Create mask tensors - some regular, some with 'real'/'imag' in key
mask1 = torch.randint(0, 2, (4, 3))  # Regular mask
mask2 = torch.randint(0, 2, (5, 2))  # Another regular mask
mask_real = [torch.randint(0, 2, (3, 2)), torch.randint(0, 2, (2, 4))]  # List of real masks
mask_imag = [torch.randint(0, 2, (2, 3))]  # List of imag masks
mask_last = torch.randint(0, 2, (2, 2))  # This should be skipped due to 'last' in key

# Build per_task_masks as dict of dicts for multiple tasks
per_task_masks = {
    0: {
        'layer1_weight': mask1,
        'layer2_weight': mask2,
        'complex_real': mask_real,
        'complex_imag': mask_imag,
        'last_layer': mask_last,
        'none_mask': None,
    },
    1: {
        'layer1_weight': torch.randint(0, 2, (4, 3)),
        'layer2_weight': torch.randint(0, 2, (5, 2)),
        'complex_real': [torch.randint(0, 2, (3, 2))],
        'last_layer': torch.randint(0, 2, (2, 2)),
    }
}

ARGS = {
    'task_id': 0,
    'per_task_masks': per_task_masks,
    'g_sparsity': 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 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
percentile Ran
ran; our contract draft was wrong, not the code
ihaeyong/pnr/model/subnet.py
pointer only (licence: MIT); the stored copy get_code serves, from ihaeyong/SoftNet-3DLS, is pointer only (licence: NONE) · get_code("913e2d426938c7d4")
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 ihaeyong/SoftNet-3DLS, 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.

quantize_per_tensor Ran
ran on a synthesized input
ihaeyong/pnr/utils.py
code served (permissive licence) · get_code("9ba78d582d59f8ed")
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 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 torch

torch.manual_seed(0)

# Test case 1: axis=-1 (global quantization)
t1 = torch.randn(2, 4) * 10
t1[0, 0] = 0  # Add some zeros

# Test case 2: axis=0 (per-row quantization, 4D tensor)
t2 = torch.randn(2, 3, 4, 5) * 10
t2[0, 0, 0, 0] = 0

# Test case 3: axis=1 (per-column quantization, 2D tensor)
t3 = torch.randn(3, 4) * 10
t3[0, 0] = 0

ARGS = {
    "t": t1,
    "bit": 8,
    "axis": -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 tuple, ndim 3, 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
all_gather Not yet run
unverified (harvested, no recorded run)
ihaeyong/pnr/utils.py
code served (permissive licence) · get_code("a1ba6b0c4d96716f")
How it ran

no invocation recorded.

all_reduce Not yet run
unverified (harvested, no recorded run)
ihaeyong/pnr/utils.py
code served (permissive licence) · get_code("439067335e7b67b5")
How it ran

no invocation recorded.

evaluate Not yet run
unverified (harvested, no recorded run)
ihaeyong/pfnr/train_nerv.py
code served (permissive licence) · get_code("d8684b8a9ca3ef7a")
How it ran

no invocation recorded.

get_consolidated_masks Not yet run
unverified (harvested, no recorded run)
ihaeyong/pnr/plot_nerv_eval.py
code served (permissive licence) · get_code("cc46e27629d64950")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 7 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 0 the same month, 3 later) and 8 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
compute_conv_output_size 7 other papers
6 dated earlier, 1 later
1905.11614 (2019-05)
Uncertainty-based Continual Learning with Adaptive Regularization
2003.09553 (2020-03)
Adversarial Continual Learning
2008.11364 (2020-08)
Improving Semi-supervised Federated Learning by Reducing the Gradient Diversity of Models
and 4 more not named here
NormLayer 4 other papers
4 dated earlier · 2 only as an assembly brick
2110.13903 (2021-10)
NeRV: Neural Representations for Videos
2207.08132 (2022-07)
E-NeRV: Expedite Neural Video Representation with Disentangled Spatial-Temporal Context
2210.06823 (2022-10)
Scalable Neural Video Representations with Learnable Positional Features
and 1 more not named here
ActivationLayer 2 other papers
2 dated earlier
2110.13903 (2021-10)
NeRV: Neural Representations for Videos
2210.06823 (2022-10)
Scalable Neural Video Representations with Learnable Positional Features
MLP 2 other papers
2 dated earlier
2110.13903 (2021-10)
NeRV: Neural Representations for Videos
2210.06823 (2022-10)
Scalable Neural Video Representations with Learnable Positional Features
all_gather 2 other papers
1 dated earlier, 1 later
2306.02018 (2023-06)
VideoComposer: Compositional Video Synthesis with Motion Controllability
Pei_D2ST-Adapter_Disentangled-and-Deformable_Spatio-Temporal_Adapter_for_Few-shot_Action_Recognition_ICCV_2025_paper (2025 (year only))
D 2 ST-Adapter: Disentangled-and-Deformable Spatio-Temporal Adapter for Few-shot Action Recognition
all_reduce 2 other papers
1 dated earlier, 1 later
2306.02018 (2023-06)
VideoComposer: Compositional Video Synthesis with Motion Controllability
Pei_D2ST-Adapter_Disentangled-and-Deformable_Spatio-Temporal_Adapter_for_Few-shot_Action_Recognition_ICCV_2025_paper (2025 (year only))
D 2 ST-Adapter: Disentangled-and-Deformable Spatio-Temporal Adapter for Few-shot Action Recognition
percentile 1 other paper
1 dated earlier
2209.07529 (2022-09)
On the Soft-Subnetwork for Few-shot Class Incremental Learning

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

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

Abstract

Inspired by the Lottery Ticket Hypothesis (LTH), which highlights the existence of efficient subnetworks within larger, dense networks, a high-performing Winning Subnetwork (WSN) in terms of task performance under appropriate sparsity conditions is considered for various continual learning tasks. It leverages pre-existing weights from dense networks to achieve efficient learning in Task Incremental Learning (TIL) and Task-agnostic Incremental Learning (TaIL) scenarios. In Few-Shot Class Incremental Learning (FSCIL), a variation of WSN referred to as the Soft subnetwork (SoftNet) is designed to prevent overfitting when the data samples are scarce. Furthermore, the sparse reuse of WSN weights is considered for Video Incremental Learning (VIL). The use of Fourier Subneural Operator (FSO) within WSN is considered. It enables compact encoding of videos and identifies reusable subnetworks across varying bandwidths. We have integrated FSO into different architectural frameworks for continual learning, including VIL, TIL, and FSCIL. Our comprehensive experiments demonstrate FSO's effectiveness, significantly improving task performance at various convolutional representational levels. Specifically, FSO enhances higher-layer performance in TIL and FSCIL and lower-layer performance in VIL.

For agents

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

get_harvested_code_for_paper("2312.11973")
get_code_for_paper("2312.11973")
have("2312.11973")

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

curl https://syntology.ai/api/ran/2312.11973.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.11973.svg)](https://syntology.ai/paper/2312.11973)

Connect an agent — have() is free.