SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.09349 · ICML · 2024

Adversarial Robustness Limits via Scaling-Law and Human-Alignment Studies

Bhavya Kailkhura, Brian Bartoldson, James Diffenderfer, Konstantinos Parasyris

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 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, 1 violated it and 0 have no contract verdict; 2 failures of our instrument; 0 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/2404.09349.json · badge: /api/ran/2404.09349.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
bbartoldson/adversarial-robustness-limits canonical (classifier label)
classifier basis: classifier_2025
4 ran · 0 no recorded run
FunctionStatusWhere it lives
file_contains Ran
ran, violated the contract we drafted
bbartoldson/adversarial-robustness-limits/submit.py
code served (permissive licence) · get_code("29bf95206c3c5e5f")
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

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

with open(test_file, 'w') as f:
    f.write("line one\n")
    f.write("target string\n")
    f.write("line three\n")

# Define arguments for file_contains function
ARGS = {
    "fn": test_file,
    "string": "target string\n"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — 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
get_aa Ran
ran, honoured the contract we drafted
bbartoldson/adversarial-robustness-limits/eval-aa.py
code served (permissive licence) · get_code("3a91811af0bdb328")
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 os
import tempfile

# Create a temporary file with the expected format
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
temp_file.write("some line\n")
temp_file.write("robust accuracy: 0.95\n")
temp_file.write("another line\n")
temp_file.write("robust accuracy: 0.95\n")
temp_file.close()

ARGS = {
    'fn': temp_file.name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — 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
get_state Ran
ran; our contract draft was wrong, not the code
bbartoldson/adversarial-robustness-limits/submit.py
code served (permissive licence) · get_code("4fc810a6b13b1408")
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 str. 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 os
import tempfile

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()
log_subdir = os.path.join(temp_dir, 'test_desc')
os.makedirs(log_subdir, exist_ok=True)

# Create mock log files
train_log_path = os.path.join(log_subdir, 'log-train.log')
aa_log_path = os.path.join(log_subdir, 'log-aa.log')

with open(train_log_path, 'w') as f:
    f.write('Some training output\n')
    f.write('Script Completed.\n')

with open(aa_log_path, 'w') as f:
    f.write('Some AA output\n')

# Create a simple args object
class Args:
    def __init__(self, log_dir, desc):
        self.log_dir = log_dir
        self.desc = desc

args_obj = Args(log_dir=temp_dir, desc='test_desc')

ARGS = {
    'args': args_obj
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — 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
setup_chain_args Ran
ran; our contract draft was wrong, not the code
bbartoldson/adversarial-robustness-limits/submit.py
code served (permissive licence) · get_code("369b4e750a6f24aa")
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 MockChain. 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 argparse

class MockChain:
    """Mock argument parser chain object"""
    def __init__(self):
        self.arguments = []
    
    def add_argument(self, *args, **kwargs):
        """Mock add_argument method"""
        self.arguments.append((args, kwargs))
        return self
    
    def __repr__(self):
        return f"MockChain(arguments={len(self.arguments)})"

torch.manual_seed(0)

chain = MockChain()

ARGS = {
    'chain': chain
}
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 MockChain — a MockChain output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MockChain 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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 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.

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 paper revisits the simple, long-studied, yet still unsolved problem of making image classifiers robust to imperceptible perturbations. Taking CI-FAR10 as an example, SOTA clean accuracy is about 100%, but SOTA robustness to ℓ ∞ -norm bounded perturbations barely exceeds 70%. To understand this gap, we analyze how model size, dataset size, and synthetic data quality affect robustness by developing the first scaling laws for adversarial training. Our scaling laws reveal inefficiencies in prior art and provide actionable feedback to advance the field. For instance, we discovered that SOTA methods diverge notably from compute-optimal setups, using excess compute for their level of robustness. Leveraging a compute-efficient setup, we surpass the prior SOTA with 20% (70%) fewer training (inference) FLOPs. We trained various compute-efficient models, with our best achieving 74% AutoAttack accuracy (+3% gain). However, our scaling laws also predict robustness slowly grows then plateaus at 90%: dwarfing our new SOTA by scaling is impractical, and perfect robustness is impossible. To better understand this predicted limit, we carry out a small-scale human evaluation on the AutoAttack data that fools our top-performing model. Concerningly, we estimate that human performance also plateaus near 90%, which we show to be attributable to ℓ ∞ -constrained attacks' generation of invalid images not consistent with their original labels. Having characterized limiting roadblocks, we outline promising paths for future research.

For agents

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

get_harvested_code_for_paper("2404.09349")
get_code_for_paper("2404.09349")
have("2404.09349")

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

curl https://syntology.ai/api/ran/2404.09349.json

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

[![Syntology run record](https://syntology.ai/api/ran/2404.09349.svg)](https://syntology.ai/paper/2404.09349)

Connect an agent — have() is free.