SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2302.14040 · NeurIPS · 2023

Permutation Equivariant Neural Functionals

J Zico Kolter, Chelsea Finn, Samuel Sokota, Allan Zhou, Yiding Jiang, Adriano Cardace, Kaien Yang, Kaylee Burns

arXiv · PDF · Open in the Atlas

Code that ran

6 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (4 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 1 failure 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/2302.14040.json · badge: /api/ran/2302.14040.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
allanyangzhou/nfn canonical (classifier label)
classifier basis: classifier_2025
1 ran · 1 no recorded run
jkalogero/scalegmn community
Papers with Code archive
5 ran · 3 no recorded run
FunctionStatusWhere it lives
InvariantLayer Ran
ran on a synthesized input · constructed an object; no result was computed
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("25b277b54e5787ff")
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, InvariantLayer: 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)

# Define mlp_args configuration for the InvariantLayer
mlp_args = {
    'd_k': [8, 16],
    'activation': 'elu',
    'batch_norm': False,
    'layer_norm': False,
    'bias': True,
    'skip': False,
    'dropout': 0.0,
    'final_activation': 'identity',
    'weight_init': None,
    'weight_init_args': None,
    'w0': 30.0
}

# Create sample tensors
x = torch.randn(2, 4)  # batch_size=2, in_features=4
extra_features = torch.randn(2, 3)  # batch_size=2, d_extra=3

ARGS = {
    'in_features': 4,
    'out_features': 5,
    'mlp_args': mlp_args,
    'sym': 'sign',
    'sign_symmetrization': True,
    'd_extra': 3,
    'final_mlp': True,
}
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 InvariantLayer — 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 InvariantLayer 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
MLPNet Ran
ran on a synthesized input
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("59052528a73a219f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, 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
import torch.nn as nn

# INIT_ARGS for MLPNet.__init__
INIT_ARGS = {
    'd_in': 4,
    'd_out': 3,
    'mlp_args': {
        'd_k': [8, 6],  # Two hidden layers with sizes 8 and 6
        'activation': 'relu',
        'batch_norm': False,
        'layer_norm': False,
        'bias': True,
        'skip': False,
        'dropout': 0.1,
        'final_activation': 'identity',
        'weight_init': None,
        'weight_init_args': None,
        'w0': 30.0
    },
    'pos_embed_dim': 0
}

# ARGS for MLPNet.forward
ARGS = {
    'x': torch.randn(2, 4),  # Batch size 2, input features 4
    'extra_features': None
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
NetworkSpec Ran
ran on a synthesized input · constructed an object; no result was computed
allanyangzhou/nfn/nfn/layers/layers.py
code served (permissive licence) · get_code("355080531ee83dab")
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, NetworkSpec: 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
from dataclasses import dataclass
from typing import List, Tuple

@dataclass(frozen=True)
class ArraySpec:
    shape: Tuple[int, ...]

torch.manual_seed(0)

# Create weight and bias specs for a simple 2-layer network
# Layer 1: input_dim=5, hidden_dim=3
# Layer 2: hidden_dim=3, output_dim=2

weight_spec = [
    ArraySpec(shape=(3, 5)),  # First layer: 3 neurons, 5 inputs
    ArraySpec(shape=(2, 3)),  # Second layer: 2 neurons, 3 inputs
]

bias_spec = [
    ArraySpec(shape=(3,)),  # First layer bias
    ArraySpec(shape=(2,)),  # Second layer bias
]

ARGS = {
    "weight_spec": weight_spec,
    "bias_spec": bias_spec,
}
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 NetworkSpec — 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 NetworkSpec 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
Sine Ran
ran on a synthesized input · constructed an object; no result was computed
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("a0e7d080ac1bf58c")
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, Sine: 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)

ARGS = {
    "w0": 2.0
}
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 Sine — 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 Sine 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
choose_activation Ran
ran; our contract draft was wrong, not the code
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("313b13b7bd56f7cf")
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 Sine. 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 Sine(nn.Module):
    def __init__(self, w0=1.0):
        super().__init__()
        self.w0 = w0

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return torch.sin(self.w0 * x)

def choose_activation(activation, **kwargs):
    activations = {'elu': nn.ELU(), 'relu': nn.ReLU(), 'silu': nn.SiLU(), 'tanh': nn.Tanh(), 'sigmoid': nn.Sigmoid(), 'softmax': nn.Softmax(), 'sine': Sine(kwargs['w0']), 'identity': lambda x: x}
    if activation in activations:
        return activations[activation]
    else:
        raise NotImplementedError

ARGS = {
    'activation': 'sine',
    'w0': 2.0
}
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 Sine — a Sine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sine 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
mlp Ran
ran on a synthesized input · constructed an object; no result was computed
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("67d3763fe4254040")
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, mlp: 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)

ARGS = {
    'in_features': 4,
    'out_features': 3,
    'd_k': [8, 6],
    'activation': 'elu',
    'batch_norm': False,
    'layer_norm': False,
    'bias': True,
    'skip': False,
    'dropout': 0.0,
    'final_activation': 'identity',
    'weight_init': None,
    'weight_init_args': None,
    'w0': 30.0,
}
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 mlp — 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 mlp 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
EquivariantLayer Not yet run
unverified (harvested, no recorded run)
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("da9f6cc5270f3205")
How it ran

no invocation recorded.

EquivariantNet Not yet run
unverified (harvested, no recorded run)
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("48fdd1872aded0c3")
How it ran

no invocation recorded.

InvariantNet Not yet run
unverified (harvested, no recorded run)
jkalogero/scalegmn/src/scalegmn/layers.py
code served (permissive licence) · get_code("0d5da287e13d8e84")
How it ran

no invocation recorded.

NPLinear Not yet run
unverified (harvested, no recorded run)
allanyangzhou/nfn/nfn/layers/layers.py
code served (permissive licence) · get_code("af80d775104f9282")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 9 are also present in other papers whose code we hold (8 with at least one paper dated earlier, 0 the same month, 1 later) and 1 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
EquivariantLayer
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces
EquivariantNet
brick of this paper's assembly: entry
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces
InvariantLayer
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces
InvariantNet
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces
MLPNet
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces
NPLinear
brick of this paper's assembly: entry
1 other paper
1 later
2409.11697 (2024-09)
Monomial Matrix Group Equivariant Neural Functional Networks
Sine
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces
choose_activation
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces
mlp
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2301.12780 (2023-01)
Equivariant Architectures for Learning in Deep Weight Spaces

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

This work studies the design of neural networks that can process the weights or gradients of other neural networks, which we refer to as neural functional networks (NFNs). Despite a wide range of potential applications, including learned optimization, processing implicit neural representations, network editing, and policy evaluation, there are few unifying principles for designing effective architectures that process the weights of other networks. We approach the design of neural functionals through the lens of symmetry, in particular by focusing on the permutation symmetries that arise in the weights of deep feedforward networks because hidden layer neurons have no inherent order. We introduce a framework for building permutation equivariant neural functionals, whose architectures encode these symmetries as an inductive bias. The key building blocks of this framework are NF-Layers (neural functional layers) that we constrain to be permutation equivariant through an appropriate parameter sharing scheme. In our experiments, we find that permutation equivariant neural functionals are effective on a diverse set of tasks that require processing the weights of MLPs and CNNs, such as predicting classifier generalization, producing "winning ticket" sparsity masks for initializations, and classifying or editing implicit neural representations (INRs). In addition, we provide code for our models and experiments 1 .

For agents

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

get_harvested_code_for_paper("2302.14040")
get_code_for_paper("2302.14040")
have("2302.14040")

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

curl https://syntology.ai/api/ran/2302.14040.json

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

[![Syntology run record](https://syntology.ai/api/ran/2302.14040.svg)](https://syntology.ai/paper/2302.14040)

Connect an agent — have() is free.