SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.07591 · ICLR · 2024

Robustifying and Boosting Training-Free Neural Architecture Search

Zhongxiang Dai, Bryan Kian, Hsiang Low, Yao Shu, Zhenfeng He

arXiv · PDF · Open in the Atlas

Code that ran

9 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; 4 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 3 failures of our instrument; 6 no recorded run); 2 of them were also run on the host, not in the jail, by the peer-agreement lane that set their level, 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/2403.07591.json · badge: /api/ran/2403.07591.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
hzf1174/RoBoT canonical (classifier label)
classifier basis: census
6 ran · 3 no recorded run
hzf1174/robot canonical (classifier label)
classifier basis: classifier_2025
3 ran · 3 no recorded run
FunctionStatusWhere it lives
calculate_md5 Ran
ran; our contract draft was wrong, not the code
hzf1174/RoBoT/foresight/imagenet16.py
code served (permissive licence) · get_code("ad0636a128295d5a")
How it ran

Ran in Syntology's jail on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned str. The image it ran in was not recorded. The comparison behind this sample's level is a different run: the peer-agreement lane ran it again on the host, not in the jail (report dated 2026-09-05T01:26:15Z). 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 os
import tempfile
import hashlib

# Create a temporary file with some content for testing
temp_dir = tempfile.mkdtemp()
test_file = os.path.join(temp_dir, "test_file.bin")

# Write some test data to the file
with open(test_file, 'wb') as f:
    f.write(b"Hello, this is test data for MD5 calculation." * 100)

ARGS = {
    "fpath": test_file,
    "chunk_size": 1024 * 1024
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
check_md5 Ran
ran, violated the contract we drafted
hzf1174/RoBoT/foresight/imagenet16.py
code served (permissive licence) · get_code("380a00b777aad655")
How it ran

Ran in Syntology's jail on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. The comparison behind this sample's level is a different run: the peer-agreement lane ran it again on the host, not in the jail (report dated 2026-09-04T23:17:45Z). 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 os
import sys
import hashlib
import torch
import tempfile

# Create a temporary file with known content
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test_file.bin")

# Write some test data to the file
test_data = b"Hello, this is test data for MD5 calculation!"
with open(temp_file, 'wb') as f:
    f.write(test_data)

# Calculate the MD5 hash of the test data
md5_hash = hashlib.md5(test_data).hexdigest()

# Define arguments for check_md5 function
ARGS = {
    'fpath': temp_file,
    'md5': md5_hash
}
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 bool, 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
accuracy Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hzf1174/RoBoT/darts_space/utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from shuyao95/HNAS, is pointer only (licence: NONE) · get_code("eab4f74c38cc3c8f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from shuyao95/HNAS, 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.

adjust_lr Ran
ran, honoured the contract we drafted
hzf1174/RoBoT/darts_space/train_imagenet.py
code served (permissive licence) · get_code("e127535d5e962cac")
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 float, 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
from torch.optim import SGD
import sys
from io import StringIO

# Set random seed
torch.manual_seed(0)

# Mock the args from get_args() by capturing it
old_argv = sys.argv
sys.argv = ['test']

import argparse
parser = argparse.ArgumentParser('imagenet')
parser.add_argument('--data_dir', type=str, default='./data/imagenet', help='location of ImageNet')
parser.add_argument('--class_num', type=int, default=1000, help='the number of classes')
parser.add_argument('--batch_size', type=int, default=1024, help='batch size')
parser.add_argument('--learning_rate', type=float, default=0.5, help='init learning rate')
parser.add_argument('--learning_rate_min', type=float, default=0.0, help='minimal init learning rate')
parser.add_argument('--lr_scheduler', type=str, default='linear', help='lr scheduler, linear, exp or cosine')
parser.add_argument('--momentum', type=float, default=0.9, help='momentum')
parser.add_argument('--weight_decay', type=float, default=3e-05, help='weight decay')
parser.add_argument('--report_freq', type=float, default=100, help='report frequency')
parser.add_argument('--epochs', type=int, default=250, help='num of training epochs')
parser.add_argument('--start_epoch', type=int, default=0, help='current epoch for reload')
parser.add_argument('--init_channels', type=int, default=48, help='num of init channels')
parser.add_argument('--layers', type=int, default=14, help='total number of layers')
parser.add_argument('--auxiliary', action='store_true', default=False, help='use auxiliary tower')
parser.add_argument('--auxiliary_weight', type=float, default=0.4, help='weight for auxiliary loss')
parser.add_argument('--drop_path_prob', type=float, default=0.0, help='drop path probability')
parser.add_argument('--save', type=str, default='ImageNet', help='experiment name')
parser.add_argument('--reload', action='store_true', default=False, help='reload initial weights')
parser.add_argument('--seed', type=int, default=0, help='random seed')
parser.add_argument('--arch', type=str, default='NASI', help='which architecture to use')
parser.add_argument('--grad_clip', type=float, default=5, help='gradient clipping')
parser.add_argument('--label_smooth', type=float, default=0.1, help='label smoothing')
parser.add_argument('--gamma', type=float, default=0.97, help='learning rate decay')
parser.add_argument('--decay_period', type=int, default=1, help='epochs between two learning rate decays')
parser.add_argument('--tensorboard_dir', default='./tensorboard/', type=str, help='tensorboard log')

args = parser.parse_args()
sys.argv = old_argv

# Create a simple model for the optimizer
model = nn.Linear(4, 3)

# Create optimizer
optimizer = SGD(model.parameters(), lr=args.learning_rate, momentum=args.momentum, weight_decay=args.weight_decay)

# Define ARGS dict for adjust_lr function
ARGS = {
    'optimizer': optimizer,
    'epoch': 10
}
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 float, 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
check_integrity Ran
ran, violated the contract we drafted
hzf1174/RoBoT/foresight/imagenet16.py
code served (permissive licence) · get_code("e947a6effcaaf32b")
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 bool, 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 os
import tempfile
import hashlib

# Create a temporary file with known content for testing
temp_dir = tempfile.mkdtemp()
test_file = os.path.join(temp_dir, "test_file.txt")

# Write some test content to the file
with open(test_file, 'w') as f:
    f.write("test content for md5 checking")

# Calculate the MD5 hash of the test file
md5_hash = hashlib.md5()
with open(test_file, 'rb') as f:
    for chunk in iter(lambda: f.read(1024 * 1024), b''):
        md5_hash.update(chunk)
correct_md5 = md5_hash.hexdigest()

# Define arguments for check_integrity function
ARGS = {
    'fpath': test_file,
    'md5': correct_md5
}
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 bool, 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
copynet Ran
ran; our contract draft was wrong, not the code
hzf1174/robot/foresight/pruners/predictive.py
code served (permissive licence) · get_code("c1f272141fc24942")
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 SimpleModel. 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 types
import copy

torch.manual_seed(0)

def no_op(self, x):
    return x

def copynet(self, bn):
    net = copy.deepcopy(self)
    if bn == False:
        for l in net.modules():
            if isinstance(l, nn.BatchNorm2d) or isinstance(l, nn.BatchNorm1d):
                l.forward = types.MethodType(no_op, l)
    return net

# Build a small model with BatchNorm layers
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3, 4, kernel_size=3)
        self.bn1 = nn.BatchNorm2d(4)
        self.linear = nn.Linear(4, 2)
        self.bn2 = nn.BatchNorm1d(2)
    
    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = x.view(x.size(0), -1)
        x = self.linear(x)
        x = self.bn2(x)
        return x

model = SimpleModel()

ARGS = {
    'self': model,
    'bn': 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 SimpleModel — a SimpleModel output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleModel 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
count_parameters_in_MB Ran
ran, honoured the contract we drafted
hzf1174/RoBoT/darts_space/utils.py
code served (permissive licence) · get_code("e615a9d09865bd9f")
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 float64, ndim 0, 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
import numpy as np

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
        self.auxiliary_layer = nn.Linear(3, 2)
    
    def forward(self, x):
        return self.linear(x)

model = SimpleModel()

ARGS = {
    "model": model
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
no_op Ran
ran, honoured the contract we drafted
hzf1174/robot/foresight/pruners/predictive.py
code served (permissive licence) · get_code("63f0914560e34c40")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class NoOpModule(nn.Module):
    def __init__(self):
        super().__init__()
    
    def no_op(self, x):
        return x

module = NoOpModule()
x = torch.randn(2, 4)

ARGS = {
    'self': module,
    'x': x
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sum_arr Ran
ran, honoured the contract we drafted
hzf1174/robot/darts_space/search.py
code served (permissive licence) · get_code("7f332db65cde5273")
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 float, 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)

# The function expects arr to be an iterable of tensors
# It iterates through arr, calls torch.sum on each element, and accumulates
arr = [
    torch.randn(2, 4),
    torch.randn(3, 5),
    torch.randn(2, 3)
]

ARGS = {
    "arr": arr
}
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 float, 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
drop_path Not yet run
unverified (harvested, no recorded run)
hzf1174/RoBoT/darts_space/utils.py
code served (permissive licence) · get_code("2c6bc306e5f9c38b")
How it ran

no invocation recorded.

find_measures Not yet run
unverified (harvested, no recorded run)
hzf1174/robot/foresight/pruners/predictive.py
code served (permissive licence) · get_code("bfe0770011ad29c4")
How it ran

no invocation recorded.

find_measures_arrays Not yet run
unverified (harvested, no recorded run)
hzf1174/robot/foresight/pruners/predictive.py
code served (permissive licence) · get_code("20066c8b8cca9c9a")
How it ran

no invocation recorded.

get_cifar_dataloaders Not yet run
unverified (harvested, no recorded run)
hzf1174/RoBoT/foresight/dataset.py
code served (permissive licence) · get_code("c8dca8dc011090bb")
How it ran

no invocation recorded.

get_mnist_dataloaders Not yet run
unverified (harvested, no recorded run)
hzf1174/RoBoT/foresight/dataset.py
code served (permissive licence) · get_code("7491772a906c4046")
How it ran

no invocation recorded.

search Not yet run
unverified (harvested, no recorded run)
hzf1174/robot/search_nb201.py
code served (permissive licence) · get_code("a9a567af7564f3f0")
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, 6 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 0 the same month, 3 later) and 9 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
accuracy 19 other papers
12 dated earlier, 7 later
1805.08318 (2018-05)
Self-Attention Generative Adversarial Networks
2105.14148 (2021-05)
OpenMatch: Open-set Consistency Regularization for Semi-supervised Learning with Outliers
2110.03006 (2021-10)
Unsupervised Selective Labeling for More Effective Semi-Supervised Learning
and 16 more not named here
calculate_md5 13 other papers
11 dated earlier, 1 later, 1 same year, order not determinable
2001.00326 (2020-01)
NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search
2003.09553 (2020-03)
Adversarial Continual Learning
2106.05390 (2021-06)
Optimizing Reusable Knowledge for Continual Learning via Metalearning
and 10 more not named here
check_md5 13 other papers
11 dated earlier, 1 later, 1 same year, order not determinable
2001.00326 (2020-01)
NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search
2003.09553 (2020-03)
Adversarial Continual Learning
2106.05390 (2021-06)
Optimizing Reusable Knowledge for Continual Learning via Metalearning
and 10 more not named here
adjust_lr 8 other papers
8 dated earlier
1806.09055 (2018-06)
DARTS: Differentiable Architecture Search
1904.12760 (2019-04)
Progressive Differentiable Architecture Search: Bridging the Depth Gap between Search and Evaluation
1906.07528 (2019-06)
Prune and Replace NAS
and 5 more not named here
check_integrity 6 other papers
5 dated earlier, 1 same year, order not determinable
2001.00326 (2020-01)
NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search
2108.11939 (2021-08)
Understanding and Accelerating Neural Architecture Search with Training-Free and Theory-Grounded Metrics
2205.05662 (2022-05)
Deep Architecture Connectivity Matters for Its Convergence: A Fine-Grained Analysis
and 3 more not named here
sum_arr 1 other paper
1 dated earlier
2110.08616 (2021-10)
GradSign: Model Performance Inference with Theoretical Insights

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

Neural architecture search (NAS) has become a key component of AutoML and a standard tool to automate the design of deep neural networks. Recently, trainingfree NAS as an emerging paradigm has successfully reduced the search costs of standard training-based NAS by estimating the true architecture performance with only training-free metrics. Nevertheless, the estimation ability of these metrics typically varies across different tasks, making it challenging to achieve robust and consistently good search performance on diverse tasks with only a single trainingfree metric. Meanwhile, the estimation gap between training-free metrics and the true architecture performances limits training-free NAS to achieve superior performance. To address these challenges, we propose the robustifying and boosting training-free NAS (RoBoT) algorithm which (a) employs the optimized combination of existing training-free metrics explored from Bayesian optimization to develop a robust and consistently better-performing metric on diverse tasks, and (b) applies greedy search, i.e., the exploitation, on the newly developed metric to bridge the aforementioned gap and consequently to boost the search performance of standard training-free NAS further. Remarkably, the expected performance of our RoBoT can be theoretically guaranteed, which improves over the existing training-free NAS under mild conditions with additional interesting insights. Our extensive experiments on various NAS benchmark tasks yield substantial empirical evidence to support our theoretical results. Our code has been made publicly available at https://github.com/hzf1174/RoBoT.

For agents

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

get_harvested_code_for_paper("2403.07591")
get_code_for_paper("2403.07591")
have("2403.07591")

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

curl https://syntology.ai/api/ran/2403.07591.json

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

[![Syntology run record](https://syntology.ai/api/ran/2403.07591.svg)](https://syntology.ai/paper/2403.07591)

Connect an agent — have() is free.