SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2503.17138 · ICLR · 2025

Workshop on Neural Network Weights as a New Data Modality 2025 STRUCTURE IS NOT ENOUGH: LEVERAGING BEHAVIOR FOR NEURAL NETWORK WEIGHT RECONSTRUCTION

Damian Borth, Konstantin Sch Ürholt, Léo Meynent, Ivan Melev, Göran Kauermann

arXiv · PDF · Open in the Atlas

Code that ran

6 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; 0 honoured the contract we drafted, 2 violated it and 2 have no contract verdict; 2 failures of our instrument; 7 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/2503.17138.json · badge: /api/ran/2503.17138.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
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough canonical (classifier label)
classifier basis not recorded on the link
6 ran · 7 no recorded run
FunctionStatusWhere it lives
load_checkpoint Ran
ran on a synthesized input
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_base_trojai.py
code served (permissive licence) · get_code("e8143a66cdd639c4")
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 copy
import logging
from pathlib import Path
import tempfile
import os

torch.manual_seed(0)

# Create a temporary directory and checkpoint file
temp_dir = tempfile.mkdtemp()
checkpoint_path = Path(temp_dir)

# Create a minimal valid checkpoint file
checkpoint_data = {
    'model_state': torch.randn(2, 4),
    'optimizer_state': torch.randn(3, 5),
    'epoch': 10
}
torch.save(checkpoint_data, checkpoint_path / 'checkpoint.pt')

# Define helper functions that the main function calls
def test_checkpoint_for_nan(checkpoint):
    """Check if checkpoint contains NaN values"""
    for key, value in checkpoint.items():
        if isinstance(value, torch.Tensor):
            if torch.isnan(value).any():
                return True
    return False

def test_checkpoint_with_threshold(checkpoint, threshold):
    """Check if checkpoint values exceed threshold"""
    for key, value in checkpoint.items():
        if isinstance(value, torch.Tensor):
            if (value.abs() > threshold).any():
                return True
    return False

# Define a filter function that returns False (don't filter out)
def filter_func(path):
    """Filter function that accepts the checkpoint"""
    return False

# Construct arguments
ARGS = {
    'path': checkpoint_path,
    'weight_threshold': 10.0,
    'filter_function': filter_func
}
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
permute_model_vector Ran
ran; our contract draft was wrong, not the code
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/augmentations.py
pointer only (licence: MIT); the stored copy get_code serves, from HSG-AIML/SANE, is pointer only (licence: NONE) · get_code("889fe8497b0ba7c8")
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 HSG-AIML/SANE, 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.

read_properties_from_path Ran
ran; our contract draft was wrong, not the code
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_base.py
pointer only (licence: MIT); the stored copy get_code serves, from HSG-AIML/SANE, is pointer only (licence: NONE) · get_code("5ac745803cfc2796")
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 HSG-AIML/SANE, 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.

read_properties_from_path Ran
ran on a synthesized input
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_base_trojai.py
code served (permissive licence) · get_code("ca6ff6ea8664f6df")
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 tempfile
import json
from pathlib import Path

# Create temporary directory structure with necessary files
temp_dir = tempfile.mkdtemp()
temp_path = Path(temp_dir)

# Create config.json with required fields
config_data = {
    "POISONED": False,
    "model_name": "mobile_net_v2",
    "learning_rate": 0.001,
    "batch_size": 32,
    "epochs": 10,
    "some_float": 0.95,
    "some_int": 5,
    "some_bool": True
}

with open(str(temp_path / 'config.json'), 'w') as f:
    json.dump(config_data, f)

# Create stats.json with required fields
stats_data = {
    "final_accuracy": 0.92,
    "final_loss": 0.15,
    "training_time": 3600,
    "final_triggered_val_acc": 0.88,
    "final_triggered_val_loss": 0.20,
    "final_triggered_data_test_acc": 0.85,
    "new_metric": 42
}

with open(str(temp_path / 'stats.json'), 'w') as f:
    json.dump(stats_data, f)

ARGS = {
    "path": temp_path
}
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 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
test_checkpoint_for_nan Ran
ran, violated the contract we drafted
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_auxiliaries.py
pointer only (licence: MIT); the stored copy get_code serves, from HSG-AIML/SANE, is pointer only (licence: NONE) · get_code("ec1ff9919f412754")
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 HSG-AIML/SANE, 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.

test_checkpoint_with_threshold Ran
ran, violated the contract we drafted
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_auxiliaries.py
pointer only (licence: MIT); the stored copy get_code serves, from HSG-AIML/SANE, is pointer only (licence: NONE) · get_code("d2a9bd8273d8f5d4")
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 HSG-AIML/SANE, 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.

cast_properties_to_numbers Not yet run
unverified (harvested, no recorded run)
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_ffcv_trojai.py
code served (permissive licence) · get_code("6e1e8bc2c8ad772d")
How it ran

no invocation recorded.

fib Not yet run
unverified (harvested, no recorded run)
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/skeleton.py
code served (permissive licence) · get_code("227a3d4f5ac6c02d")
How it ran

no invocation recorded.

get_net_epoch_lst_from_label Not yet run
unverified (harvested, no recorded run)
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_auxiliaries.py
pointer only (licence: MIT); the stored copy get_code serves, from HSG-AIML/SANE, is pointer only (licence: NONE) · get_code("31ce1adb9a55256c")
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 HSG-AIML/SANE, 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.

load_checkpoint Not yet run
unverified (harvested, no recorded run)
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_base.py
code served (permissive licence) · get_code("336c00ba04cc94cf")
How it ran

no invocation recorded.

load_checkpoints_remote Not yet run
unverified (harvested, no recorded run)
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_base.py
code served (permissive licence) · get_code("5fb96d161e21926b")
How it ran

no invocation recorded.

load_checkpoints_remote Not yet run
unverified (harvested, no recorded run)
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_base_trojai.py
code served (permissive licence) · get_code("2b29b730acf17450")
How it ran

no invocation recorded.

read_properties_from_path Not yet run
unverified (harvested, no recorded run)
HSG-AIML/ICLR_WSL_2025-Structure_is_not_enough/skaling_hyper_reps/src/shrp/datasets/dataset_properties.py
pointer only (licence: MIT); the stored copy get_code serves, from HSG-AIML/SANE, is pointer only (licence: NONE) · get_code("790ee3cda94b46c4")
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 HSG-AIML/SANE, 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 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, 9 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 0 the same month, 8 later) and 4 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_net_epoch_lst_from_label 2 other papers
1 dated earlier, 1 later
2406.09997 (2024-06)
Towards Scalable and Versatile Weight Space Learning
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space Learning
permute_model_vector 2 other papers
1 dated earlier, 1 later
2406.09997 (2024-06)
Towards Scalable and Versatile Weight Space Learning
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space Learning
read_properties_from_path 2 other papers
1 dated earlier, 1 later
2406.09997 (2024-06)
Towards Scalable and Versatile Weight Space Learning
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space Learning
read_properties_from_path 2 other papers
1 dated earlier, 1 later
2406.09997 (2024-06)
Towards Scalable and Versatile Weight Space Learning
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space Learning
test_checkpoint_for_nan 2 other papers
1 dated earlier, 1 later
2406.09997 (2024-06)
Towards Scalable and Versatile Weight Space Learning
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space Learning
test_checkpoint_with_threshold 2 other papers
1 dated earlier, 1 later
2406.09997 (2024-06)
Towards Scalable and Versatile Weight Space Learning
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space Learning
fib 1 other paper
1 dated earlier
2110.06197 (2021-10)
Crystal Diffusion Variational Autoencoder for Periodic Material Generation
load_checkpoint 1 other paper
1 later
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space Learning
load_checkpoints_remote 1 other paper
1 later
2504.10141 (2025-04)
The Impact of Model Zoo Size and Composition on Weight Space 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

The weights of neural networks (NNs) have recently gained prominence as a new data modality in machine learning, with applications ranging from accuracy and hyperparameter prediction to representation learning or weight generation. One approach to leverage NN weights involves training autoencoders (AEs), using contrastive and reconstruction losses. This allows such models to be applied to a wide variety of downstream tasks, and they demonstrate strong predictive performance and low reconstruction error. However, despite the low reconstruction error, these AEs reconstruct NN models with deteriorated performance compared to the original ones, limiting their usability with regard to model weight generation. In this paper, we identify a limitation of weight-space AEs, specifically highlighting that a structural loss, that uses the Euclidean distance between original and reconstructed weights, fails to capture some features critical for reconstructing high-performing models. We analyze the addition of a behavioral loss for training AEs in weight space, where we compare the output of the reconstructed model with that of the original one, given some common input. We show a strong synergy between structural and behavioral signals, leading to increased performance in all downstream tasks evaluated, in particular NN weights reconstruction and generation.

For agents

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

get_harvested_code_for_paper("2503.17138")
get_code_for_paper("2503.17138")
have("2503.17138")

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

curl https://syntology.ai/api/ran/2503.17138.json

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

[![Syntology run record](https://syntology.ai/api/ran/2503.17138.svg)](https://syntology.ai/paper/2503.17138)

Connect an agent — have() is free.