SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1608.04644 · IEEE Symposium on Security and Privacy · 2016

Towards Evaluating the Robustness of Neural Networks

Nicholas Carlini, David Wagner

arXiv · PDF · Open in the Atlas

Code that ran

12 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; 1 honoured the contract we drafted, 0 violated it and 9 have no contract verdict; 2 failures of our instrument; 1 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/1608.04644.json · badge: /api/ran/1608.04644.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
carlini/nn_robust_attacks canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
PerryXDeng/adversarial_mnist_attacks community (classifier label)
classifier basis not recorded on the link
6 ran · 0 no recorded run
kkew3/pytorch-cw2 community (classifier label)
classifier basis not recorded on the link
2 ran · 1 no recorded run
Bhushan-Jagtap-2013/Adversarial_Attack_on_RNN extension (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
zetongqi/attacking_neural_networks reimplementation (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
jfc43/pixel-discretization community (classifier label)
classifier basis not recorded on the link
1 ran · 0 no recorded run
FunctionStatusWhere it lives
activation Ran
ran on a synthesized input
PerryXDeng/adversarial_mnist_attacks/neural_network.py
code served (permissive licence) · get_code("cc16b0d5bc15bb67")
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 ndarray, ndim 2, dtype family float, shape (3, 2) 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
import scipy.special as ss

def sigmoid(z):
    """
  sigmoid/softmax activation, uses ss.expit to avoid numeric instability
  :param z: vectorized values to be activated
  :return: vectorized activated values
  """
    return ss.expit(z)

def activation(prev, weights, bias):
    """
  a neural network layer
  :param prev: activations of previous layer
  :param weights: weights of this layer
  :param bias: biases of this layer
  :return: activations of this layer
  """
    prev_copy = np.r_[np.ones(prev.shape[1])[np.newaxis], prev]
    weights_copy = np.c_[bias, weights]
    return sigmoid(np.matmul(weights_copy, prev_copy))


# Setup code to construct valid arguments
# prev: activations of previous layer - shape (num_neurons_prev, batch_size)
prev = np.random.randn(5, 2).astype(np.float32)

# weights: weights matrix - shape (num_neurons_current, num_neurons_prev)
weights = np.random.randn(3, 5).astype(np.float32)

# bias: bias vector - shape (num_neurons_current,)
bias = np.random.randn(3).astype(np.float32)

ARGS = {
    "prev": prev,
    "weights": weights,
    "bias": bias
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 2) — shape (3, 2) 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
cross_entropy Ran
ran on a synthesized input
PerryXDeng/adversarial_mnist_attacks/neural_network.py
code served (permissive licence) · get_code("39df9d693fbdc4de")
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 float32, 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 numpy as np

# cross_entropy expects:
# - h: vectorized hypothesis value (probabilities, shape matching y)
# - y: vectorized actual value (one-hot or probability labels, same shape as h)
# Both should be numpy arrays suitable for element-wise operations

# Create small hypothesis values (predicted probabilities)
# Shape: (2, 4) - batch of 2, 4 classes
h = np.array([
    [0.7, 0.2, 0.05, 0.05],
    [0.1, 0.1, 0.7, 0.1]
], dtype=np.float32)

# Create corresponding ground truth labels (one-hot encoded)
# Shape: (2, 4) - same shape as h
y = np.array([
    [1.0, 0.0, 0.0, 0.0],
    [0.0, 0.0, 1.0, 0.0]
], dtype=np.float32)

ARGS = {
    'h': h,
    'y': y
}
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 float32, 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
denormalize Ran
ran on a synthesized input
PerryXDeng/adversarial_mnist_attacks/data_preparation.py
code served (permissive licence) · get_code("6c0b9095663ceff6")
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 ndarray, ndim 3, dtype family int, shape (2, 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 numpy as np

# Create a normalized input array (values typically in range [-1, 1] or [0, 1])
# The function appears to denormalize by scaling by 255 and adding 255/2
normalized_input = np.array([
    [[0.0, 0.5, -0.5, 1.0],
     [0.2, -0.2, 0.8, -0.8]],
    [[0.1, 0.3, -0.1, 0.9],
     [-0.3, 0.7, -0.7, 0.4]]
], dtype=np.float32)

ARGS = {
    "x": normalized_input
}
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 ndarray, ndim 3, dtype family int, shape (2, 2, 4) — shape (2, 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
generate_data Ran
ran; our contract draft was wrong, not the code
Bhushan-Jagtap-2013/Adversarial_Attack_on_RNN/test_attack.py
code served (permissive licence) · get_code("ae7a589f65750d71")
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. 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 numpy as np
import random

random.seed(0)
np.random.seed(0)

class FakeData:
    def __init__(self):
        num_samples = 10
        num_classes = 5
        image_shape = (28, 28, 1)
        
        self.test_data = np.random.randn(num_samples, *image_shape).astype(np.float32)
        self.test_labels = np.eye(num_classes)[np.random.randint(0, num_classes, num_samples)]

data = FakeData()

ARGS = {
    "data": data,
    "samples": 2,
    "targeted": True,
    "start": 0,
    "inception": False,
}
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 tuple — a tuple output has no array shape of its own; none is 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
generate_data Ran
ran; our contract draft was wrong, not the code
carlini/nn_robust_attacks/test_attack.py
code served (permissive licence) · get_code("33c3e5273c448151")
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. 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 numpy as np
import random

random.seed(0)
np.random.seed(0)

class MockData:
    def __init__(self):
        # num_samples=5, image_dim=4, num_classes=3
        self.test_data = np.random.randn(5, 4).astype(np.float32)
        # One-hot encoded labels
        self.test_labels = np.eye(3)[np.array([0, 1, 2, 0, 1])]

data = MockData()

ARGS = {
    "data": data,
    "samples": 2,
    "targeted": True,
    "start": 0,
    "inception": False,
}
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 tuple — a tuple output has no array shape of its own; none is 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_cuda_state Ran
ran on a synthesized input
kkew3/pytorch-cw2/runutils.py
code served (permissive licence) · get_code("ab527cad8796ffd2")
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 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 torch
import torch.nn as nn

torch.manual_seed(0)

# Test case 1: nn.Module with parameters (on CPU)
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
model_with_params = SimpleModel()

# Test case 2: tensor (on CPU)
tensor_cpu = torch.randn(2, 4)

# Test case 3: nn.Module with no parameters
class EmptyModel(nn.Module):
    def __init__(self):
        super().__init__()

model_no_params = EmptyModel()

# Use the model with parameters as the primary test object
ARGS = {
    'obj': model_with_params
}
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 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
load_image Ran
ran, honoured the contract we drafted
zetongqi/attacking_neural_networks/CW_l2attack.py
pointer only (licence: NONE) · get_code("0cb12596aa3c0436")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, zetongqi/attacking_neural_networks 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.

normalize Ran
ran on a synthesized input
PerryXDeng/adversarial_mnist_attacks/data_preparation.py
code served (permissive licence) · get_code("f33c321f1f8ee09c")
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 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 numpy as np

torch.manual_seed(0)

# The function normalize takes a single parameter 'x' and performs arithmetic operations on it
# It expects x to be a numeric type that supports subtraction and division
# We'll create a small tensor as input

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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 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
predict Ran
ran on a synthesized input
kkew3/pytorch-cw2/runutils.py
code served (permissive licence) · get_code("6189cef972a4615e")
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 Tensor, ndim 1, dtype family int, shape (2,) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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.autograd import Variable

torch.manual_seed(0)

# Create a simple neural network that takes inputs and outputs class predictions
class SimpleNet(nn.Module):
    def __init__(self):
        super(SimpleNet, self).__init__()
        self.fc1 = nn.Linear(4, 8)
        self.fc2 = nn.Linear(8, 3)  # 3 output classes
    
    def forward(self, x):
        x = torch.relu(self.fc1(x))
        x = self.fc2(x)
        return x

# Create the network
net = SimpleNet()

# Create input tensor: batch size 2, 4 features (matching input dimension)
# The function expects [B x C x W x H] but will work with any compatible shape
inputs = torch.randn(2, 4)

ARGS = {
    'net': net,
    'inputs': inputs
}
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 Tensor, ndim 1, dtype family int, shape (2,) — shape (2,) 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
preprocess Ran
ran on a synthesized input
jfc43/pixel-discretization/data-specific-discretization/ImageNet/util.py
code served (permissive licence) · get_code("544bb80adbb9ade1")
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 ndarray, ndim 3, dtype family float, shape (2, 4, 3) 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

# Set random seed for reproducibility
np.random.seed(0)

# Create small sample data
# images0: batch of images, each image is a 3D array (height, width, 3 channels)
# Shape: (2, 4, 3) - 2 images, each 4x3 pixels with 3 color channels
images0 = np.random.randint(0, 256, size=(2, 4, 3)).astype(np.uint8)

# codes: a palette of color codes to quantize to
# Shape: (5, 3) - 5 colors, each with 3 channels (RGB)
codes = np.array([
    [0, 0, 0],
    [255, 0, 0],
    [0, 255, 0],
    [0, 0, 255],
    [255, 255, 255]
], dtype=float)

ARGS = {
    "images0": images0,
    "codes": codes
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (2, 4, 3) — shape (2, 4, 3) 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
sigmoid Ran
ran on a synthesized input
PerryXDeng/adversarial_mnist_attacks/neural_network.py
code served (permissive licence) · get_code("cf589faab9919534")
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 ndarray, ndim 1, dtype family float, shape (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 numpy as np
import scipy.special as ss

torch.manual_seed(0)

# The sigmoid function takes a single parameter 'z' which should be
# vectorized values (array-like). It uses ss.expit for computation.
# We'll create a small numpy array and a torch tensor as test cases.

z_numpy = np.array([0.0, 1.0, -1.0, 2.0])
z_torch = torch.tensor([0.0, 1.0, -1.0, 2.0])

# Using numpy array as the primary argument
ARGS = {
    "z": z_numpy
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (4,) — shape (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
vectorized_label Ran
ran on a synthesized input
PerryXDeng/adversarial_mnist_attacks/data_preparation.py
code served (permissive licence) · get_code("aca0af8d07cfa276")
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 ndarray, ndim 2, dtype family float, shape (10, 1) 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

# The function takes a scalar label and converts it to a one-hot vector
# The label should be an integer between 0 and 9 (since the output vector has size 10)

ARGS = {
    "label": 5
}
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 ndarray, ndim 2, dtype family float, shape (10, 1) — shape (10, 1) 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
make_cuda_consistent Not yet run
unverified (harvested, no recorded run)
kkew3/pytorch-cw2/runutils.py
code served (permissive licence) · get_code("9a04b87a21ee7dbd")
How it ran

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 12 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 11 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
generate_data 1 other paper
1 dated earlier
1604.08275 (2016-04)
Crafting Adversarial Input Sequences for Recurrent Neural Networks

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 networks provide state-of-the-art results for most machine learning tasks. Unfortunately, neural networks are vulnerable to adversarial examples: given an input x and any target classification t, it is possible to find a new input x that is similar to x but classified as t. This makes it difficult to apply neural networks in security-critical areas. Defensive distillation is a recently proposed approach that can take an arbitrary neural network, and increase its robustness, reducing the success rate of current attacks' ability to find adversarial examples from 95% to 0.5%. In this paper, we demonstrate that defensive distillation does not significantly increase the robustness of neural networks by introducing three new attack algorithms that are successful on both distilled and undistilled neural networks with 100% probability. Our attacks are tailored to three distance metrics used previously in the literature, and when compared to previous adversarial example generation algorithms, our attacks are often much more effective (and never worse). Furthermore, we propose using high-confidence adversarial examples in a simple transferability test we show can also be used to break defensive distillation. We hope our attacks will be used as a benchmark in future defense attempts to create neural networks that resist adversarial examples.

For agents

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

get_harvested_code_for_paper("1608.04644")
get_code_for_paper("1608.04644")
have("1608.04644")

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

curl https://syntology.ai/api/ran/1608.04644.json

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

[![Syntology run record](https://syntology.ai/api/ran/1608.04644.svg)](https://syntology.ai/paper/1608.04644)

Connect an agent — have() is free.