SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2302.01757 · NeurIPS · 2023

RS-Del: Edit Distance Robustness Certificates for Sequence Classifiers via Randomized Deletion

Olga Ohrimenko, Benjamin Rubinstein, Zhuoqun Huang, Neil Marchant, Keane Lucas, Lujo Bauer

arXiv · PDF · Open in the Atlas

Code that ran

11 of 14 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; 3 honoured the contract we drafted, 0 violated it and 6 have no contract verdict; 2 failures of our instrument; 3 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/2302.01757.json · badge: /api/ran/2302.01757.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
dovermore/randomized-deletion canonical (classifier label)
classifier basis not recorded on the link
7 ran · 2 no recorded run
lions-epfl/lipslev community
Papers with Code archive
4 ran · 1 no recorded run
FunctionStatusWhere it lives
BinLCB Ran
ran, honoured the contract we drafted
lions-epfl/lipslev/verification.py
code served (permissive licence) · get_code("92f5153e34454ae1")
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 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 scipy.stats as stats

# BinLCB computes a lower confidence bound for a binomial proportion
# It takes:
# - k (int): number of successes
# - n (int): number of trials
# - alpha (float): confidence level

# The function uses stats.beta.ppf which requires numeric inputs
# It also checks if k == 0

ARGS = {
    "k": 5,
    "n": 20,
    "alpha": 0.05
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
scipy — 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
apply_del Ran
ran on a synthesized input
dovermore/randomized-deletion/src/torchmalware/certification/deletion_mech.py
code served (permissive licence) · get_code("9579b4874f24ed68")
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 3, dtype family float, shape (2, 3, 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 torch

torch.manual_seed(0)

# Create a batch of binaries with shape (batch_size, num_bytes, embed_dim)
# batch_size=2, num_bytes=4, embed_dim=3
binaries = torch.randn(2, 4, 3)

# Create indices tuple for indexing into the binaries
# indices should be a tuple of (batch_indices, byte_indices)
# We'll delete some bytes from the binaries
batch_indices = torch.tensor([0, 0, 1])
byte_indices = torch.tensor([1, 3, 2])
indices = (batch_indices, byte_indices)

# pad_value is optional - can be None or a tensor
pad_value = torch.zeros(3)

ARGS = {
    "binaries": binaries,
    "indices": indices,
    "pad_value": pad_value,
}
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 3, dtype family float, shape (2, 3, 3) — shape (2, 3, 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
apply_mask Ran
ran on a synthesized input
dovermore/randomized-deletion/src/torchmalware/certification/masking_mech.py
code served (permissive licence) · get_code("a89350a23ca96ab6")
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 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 torch

torch.manual_seed(0)

# binaries: A padded batch of binaries with shape (batch_size, max_byte_addresses, embedding_dim)
binaries = torch.randn(2, 4, 3)

# indices: A tuple of LongTensors for advanced indexing
# First tensor indexes batch dimension, second indexes byte address dimension
batch_indices = torch.LongTensor([0, 1, 0])
address_indices = torch.LongTensor([1, 2, 3])
indices = (batch_indices, address_indices)

# mask_value: Fill value for masked entries
# Should have shape compatible with the embedding dimension
mask_value = torch.zeros(3)

ARGS = {
    "binaries": binaries,
    "indices": indices,
    "mask_value": mask_value,
}
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 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
char_tokenizer Ran
ran, honoured the contract we drafted
lions-epfl/lipslev/verification.py
code served (permissive licence) · get_code("5a66080978a0cf31")
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 (5,) 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)

# Analyzing the function:
# - S: a string that will be iterated character by character
# - to_id: a dictionary mapping lowercase characters to IDs (accessed via .keys() and indexing)

S = "hello"

to_id = {
    'h': 0,
    'e': 1,
    'l': 2,
    'o': 3,
    'w': 4,
}

ARGS = {
    'S': S,
    'to_id': to_id,
}
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 (5,) — shape (5,) 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
collate_pad Ran
ran on a synthesized input
dovermore/randomized-deletion/src/torchmalware/utils.py
code served (permissive licence) · get_code("e095e86d36ac3bde")
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 3, dtype family float, shape (2, 2, 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 torch
import collections

torch.manual_seed(0)

# Test case 1: batch of tensors with equal sizes (should stack)
batch_equal_tensors = [
    torch.randn(2, 3),
    torch.randn(2, 3),
]

# Test case 2: batch of tensors with variable sizes (should pad)
batch_variable_tensors = [
    torch.randn(2, 3),
    torch.randn(3, 3),
]

# Test case 3: batch of numpy arrays
import numpy as np
batch_numpy = [
    np.array([1.0, 2.0, 3.0]),
    np.array([4.0, 5.0, 6.0]),
]

# Test case 4: batch of floats
batch_floats = [1.5, 2.5, 3.5]

# Test case 5: batch of ints
batch_ints = [1, 2, 3]

# Test case 6: batch of strings
batch_strings = ["hello", "world"]

# Test case 7: batch of dicts
batch_dicts = [
    {"a": torch.randn(2, 3), "b": torch.randn(2, 3)},
    {"a": torch.randn(2, 3), "b": torch.randn(2, 3)},
]

# Test case 8: batch of lists (sequences)
batch_lists = [
    [torch.randn(2, 3), torch.randn(2, 3)],
    [torch.randn(2, 3), torch.randn(2, 3)],
]

# Test case 9: batch with namedtuple
Point = collections.namedtuple('Point', ['x', 'y'])
batch_namedtuple = [
    Point(torch.randn(2, 3), torch.randn(2, 3)),
    Point(torch.randn(2, 3), torch.randn(2, 3)),
]

# Use the simplest case: batch of equal-sized tensors with stack=True
ARGS = {
    "batch": batch_equal_tensors,
    "stack": True,
}
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 3, dtype family float, shape (2, 2, 3) — shape (2, 2, 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
sample_del Ran
ran on a synthesized input
dovermore/randomized-deletion/src/torchmalware/certification/deletion_mech.py
code served (permissive licence) · get_code("66029f47f06157a8")
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 2, dtype family int; 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

torch.manual_seed(0)

# Small batch of binaries: shape (batch_size=2, length=4, embedding_dim=3)
binaries = torch.randn(2, 4, 3)

# Probability of deletion
p_del = 0.3

# Optional: byte_chunks tensor (shape: batch_size x length)
# Labels contiguous chunks; 0 means padding/unlabeled
byte_chunks = torch.tensor([
    [0, 1, 1, 0],
    [0, 2, 2, 3]
], dtype=torch.int64)

# Optional: minimum number of chunks to retain
min_chunks = 1

# Optional: pad_value (should match embedding dimension)
pad_value = torch.zeros(3, dtype=torch.float32)

ARGS = {
    "binaries": binaries,
    "p_del": p_del,
    "byte_chunks": byte_chunks,
    "min_chunks": min_chunks,
    "pad_value": pad_value,
}
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 2, dtype family int — 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
sample_deletes Ran
ran; our contract draft was wrong, not the code
lions-epfl/lipslev/verification.py
code served (permissive licence) · get_code("cc4f7f6dc845e532")
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 list, 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 numpy as np

ARGS = {
    'S': 'hello world',
    'n': 3,
    'p_del': 0.5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list 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
sample_mask Ran
ran on a synthesized input
dovermore/randomized-deletion/src/torchmalware/certification/masking_mech.py
code served (permissive licence) · get_code("fc750dd2d5bc8304")
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 2, dtype family int; 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

torch.manual_seed(0)

# Create a simple batch of binaries
# Shape: (batch_size=2, sequence_length=5, embedding_dim=3)
binaries = torch.randn(2, 5, 3)

# Create byte_chunks tensor
# Shape: (batch_size=2, sequence_length=5)
# Values: chunk IDs (0 for padding/unlabeled, positive integers for chunks)
byte_chunks = torch.tensor([
    [0, 1, 1, 2, 0],
    [1, 1, 0, 2, 2]
], dtype=torch.int64)

# Mask fraction: fraction of chunks to mask
mask_fraction = 0.4

# Minimum chunks to retain
min_chunks = 1

# Pad value: zeros with embedding_dim=3
pad_value = torch.zeros(3)

ARGS = {
    "binaries": binaries,
    "mask_fraction": mask_fraction,
    "byte_chunks": byte_chunks,
    "min_chunks": min_chunks,
    "pad_value": pad_value,
}
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 2, dtype family int — 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
split Ran
ran; our contract draft was wrong, not the code
dovermore/randomized-deletion/src/repeat_forward_exp.py
code served (permissive licence) · get_code("ffb145c4a0426f6d")
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 list. 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 numpy as np

torch.manual_seed(0)

# The function split(a, n) divides a sequence 'a' into 'n' roughly equal parts
# It uses len(a) and slicing operations on 'a'
# 'a' should be a sequence (list, tuple, or tensor)
# 'n' should be an integer representing the number of splits

# Create a simple list as the sequence to split
a = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

# Create an integer for the number of splits
n = 3

ARGS = {
    "a": a,
    "n": n
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list — a list 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
verify_rsdel Ran
ran, honoured the contract we drafted
lions-epfl/lipslev/verification.py
code served (permissive licence) · get_code("b31b3a42c591bfcd")
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 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 torch.nn as nn
import torch.nn.functional as F
import numpy as np
import pandas as pd
from torch.nn.utils.rnn import pad_sequence
import scipy.stats as stats

torch.manual_seed(0)
np.random.seed(0)

# Build a minimal model that matches the function's usage
class MinimalModel(nn.Module):
    def __init__(self, vocab_size=287, hidden_dim=8, num_classes=2):
        super().__init__()
        self.embedding = nn.Embedding(vocab_size, hidden_dim)
        self.lstm = nn.LSTM(hidden_dim, hidden_dim, batch_first=True)
        self.classifier = nn.Linear(hidden_dim, num_classes)
    
    def forward(self, x):
        # x shape: (batch_size, seq_len)
        embedded = self.embedding(x)  # (batch_size, seq_len, hidden_dim)
        lstm_out, (h, c) = self.lstm(embedded)  # h: (1, batch_size, hidden_dim)
        # Use last hidden state for classification
        logits = self.classifier(h[-1])  # (batch_size, num_classes)
        return logits, lstm_out

model = MinimalModel(vocab_size=287, hidden_dim=8, num_classes=2)

# Create a character-to-id mapping
to_id = {chr(i): i - 32 for i in range(32, 127)}  # ASCII printable chars
to_id.update({'a': 1, 'b': 2, 'c': 3, 'd': 4, 'e': 5, ' ': 0})

# Create a minimal dataset
dataset = [
    {'sentence': 'hello world', 'label': 0},
    {'sentence': 'test example', 'label': 1},
]

# Device
device = 'cpu'

# Output folder and name
output_folder = '/tmp/results_verif'
output_name = 'test'

# Pad value
pad = 286

# Max number of samples to process
max_n = 2

# Synonyms (optional, not used in function body)
synonyms = None

# Tokenizer function
def char_tokenizer(S, to_id):
    l = []
    for c in S:
        if c.lower() in to_id.keys():
            l.append(to_id[c.lower()])
    return torch.tensor(l)

tokenizer = char_tokenizer

# Resume flag
resume = False

# Deletion probability
p_del = 0.9

# Number of predictions for initial sampling
n_pred = 10

# Number of samples for bound computation
n_bnd = 20

ARGS = {
    'model': model,
    'to_id': to_id,
    'dataset': dataset,
    'device': device,
    'output_folder': output_folder,
    'output_name': output_name,
    'pad': pad,
    'max_n': max_n,
    'synonyms': synonyms,
    'tokenizer': tokenizer,
    'resume': resume,
    'p_del': p_del,
    'n_pred': n_pred,
    'n_bnd': n_bnd,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, pandas, scipy, torch, tqdm — 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
vrange Ran
ran on a synthesized input
dovermore/randomized-deletion/src/torchmalware/utils.py
code served (permissive licence) · get_code("431961661790790c")
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 (8,) 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)

# Create small tensor arguments for starts and ends
starts = torch.tensor([0, 5, 10])
ends = torch.tensor([3, 8, 12])

ARGS = {
    "starts": starts,
    "ends": ends,
}
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 (8,) — shape (8,) 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
collate_fn Not yet run
unverified (harvested, no recorded run)
lions-epfl/lipslev/verification.py
code served (permissive licence) · get_code("0073f606a0fc72e1")
How it ran

no invocation recorded.

get_path Not yet run
unverified (harvested, no recorded run)
dovermore/randomized-deletion/src/preprocess_pe.py
code served (permissive licence) · get_code("924f45965b43b3de")
How it ran

no invocation recorded.

inv_collate_pad Not yet run
unverified (harvested, no recorded run)
dovermore/randomized-deletion/src/torchmalware/utils.py
code served (permissive licence) · get_code("7004df60ba05e77f")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 13 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
split 1 other paper
1 later
2412.11253 (2024-12)
Are Expressive Models Truly Necessary for Offline RL?

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

Randomized smoothing is a leading approach for constructing classifiers that are certifiably robust against adversarial examples. Existing work on randomized smoothing has focused on classifiers with continuous inputs, such as images, where ℓ p -norm bounded adversaries are commonly studied. However, there has been limited work for classifiers with discrete or variable-size inputs, such as for source code, which require different threat models and smoothing mechanisms. In this work, we adapt randomized smoothing for discrete sequence classifiers to provide certified robustness against edit distance-bounded adversaries. Our proposed smoothing mechanism randomized deletion (RS-Del) applies random deletion edits, which are (perhaps surprisingly) sufficient to confer robustness against adversarial deletion, insertion and substitution edits. Our proof of certification deviates from the established Neyman-Pearson approach, which is intractable in our setting, and is instead organized around longest common subsequences. We present a case study on malware detection-a binary classification problem on byte sequences where classifier evasion is a well-established threat model. When applied to the popular MalConv malware detection model, our smoothing mechanism RS-Del achieves a certified accuracy of 91% at an edit distance radius of 128 bytes.

For agents

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

get_harvested_code_for_paper("2302.01757")
get_code_for_paper("2302.01757")
have("2302.01757")

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

curl https://syntology.ai/api/ran/2302.01757.json

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

[![Syntology run record](https://syntology.ai/api/ran/2302.01757.svg)](https://syntology.ai/paper/2302.01757)

Connect an agent — have() is free.