SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2206.14754 · 2022

Distilling Model Failures as Directions in Latent Space

arXiv · PDF · Open in the Atlas

Code that ran

6 of 7 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; 0 honoured the contract we drafted, 0 violated it and 6 have no contract verdict; 0 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/2206.14754.json · badge: /api/ran/2206.14754.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
madrylab/failure-directions canonical (classifier label)
classifier basis not recorded on the link
6 ran · 1 no recorded run
FunctionStatusWhere it lives
ffcv_read_check_override_config Ran
ran on a synthesized input
madrylab/failure-directions/failure_directions/src/config_parsing.py
code served (permissive licence) · get_code("0800a7a994ffeaf5")
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 dict. 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
from collections.abc import Iterable

torch.manual_seed(0)

# Create a minimal config dict with all required and default keys
config = {
    'train_beton': '/tmp/train.beton',
    'test_beton': '/tmp/test.beton',
    'imgsz': 224,
    'arch': 'resnet18',
    'arch_type': 'resnet',
    'num_classes': 10,
    'batch_size': 256,
    'num_workers': 2,
    'mean': [0.485, 0.456, 0.406],
    'std': [0.229, 0.224, 0.225],
    'os_cache': True,
    'quasi_random': False,
    'train_img_decoder': 'random_resized_crop',
    'val_img_decoder': 'center_crop',
    'train_aug': 'auto_augment',
    'val_aug': None,
    'loss_vec_file': '/tmp/loss_vec.npy',
    'shuffle': False,
    'drop_last': False,
    'indices_file': '/tmp/indices.npy',
    'val_beton': '/tmp/val.beton',
    'unlabeled_beton': '/tmp/unlabeled.beton',
    'loss_upweight': 10,
    'bce': True,
    'cmnist': True,
    'training': {
        'epochs': 50,
        'optimizer': {'momentum': 0.95, 'weight_decay': 0.001},
        'lr': 0.05,
        'lr_scheduler': {'type': 'cosine_lr', 'step_size': 25, 'gamma': 0.05}
    }
}

# Create override_args as an iterable of (key, value) tuples
override_args = [
    ('batch_size', 512),
    ('num_workers', 4),
]

ARGS = {
    'config': config,
    'override_args': override_args,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
get_pairwise_squared_dists Ran
ran on a synthesized input
madrylab/failure-directions/failure_directions/src/dist_shift_utils.py
code served (permissive licence) · get_code("f9c7d0f4c66bf055")
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 torch

torch.manual_seed(0)

# Create a small tensor of latents with shape (2, 4)
# 2 latent vectors, each with dimension 4
latents = torch.randn(2, 4)

ARGS = {
    "latents": latents
}
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 — 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
inv_norm Ran
ran on a synthesized input
madrylab/failure-directions/failure_directions/src/ffcv_utils.py
code served (permissive licence) · get_code("e6ee1c111aeef119")
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 Compose. 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
from torchvision import transforms

torch.manual_seed(0)

# The function inv_norm takes two parameters:
# - ds_mean: used in the Normalize call, expects an iterable of mean values
# - ds_std: checked for None, and if not None, used in the Normalize calls

# Looking at the function body:
# - ds_std is checked against None
# - If ds_std is not None, both ds_mean and ds_std are iterated over
# - They're used in list comprehensions with [x for x in ds_std] and [x for x in ds_mean]
# - The function expects 3 channels (hardcoded [0.0, 0.0, 0.0] and [1.0, 1.0, 1.0])

# Valid arguments:
ds_mean = [0.485, 0.456, 0.406]  # Example ImageNet mean values (3 channels)
ds_std = [0.229, 0.224, 0.225]   # Example ImageNet std values (3 channels)

ARGS = {
    "ds_mean": ds_mean,
    "ds_std": ds_std
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Compose — a Compose output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Compose 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
pairwise_exp_kernel Ran
ran on a synthesized input
madrylab/failure-directions/failure_directions/src/dist_shift_utils.py
code served (permissive licence) · get_code("b994e6794b2055c4")
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, 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)

# pairwise_exp_kernel takes:
# - pairwise_squared_dists: a tensor of pairwise squared distances
# - sigma: a scaling parameter (scalar)

# Create a small pairwise squared distances matrix
# Shape: (2, 3) to have distinct dimensions
pairwise_squared_dists = torch.tensor([
    [0.0, 1.5, 2.3],
    [1.5, 0.0, 0.8]
], dtype=torch.float32)

# Create a sigma parameter (scalar)
sigma = torch.tensor(0.5)

ARGS = {
    "pairwise_squared_dists": pairwise_squared_dists,
    "sigma": sigma
}
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 2, dtype family float, shape (2, 3) — shape (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
perm_2d_tensor Ran
ran on a synthesized input
madrylab/failure-directions/failure_directions/src/dist_shift_utils.py
code served (permissive licence) · get_code("74b176ef329f93cb")
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 (4, 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)

# Create a small 2D tensor (4x4 for square matrix permutation)
mat = torch.randn(4, 4)

# Create a permutation as a tensor of indices
perm = torch.tensor([2, 0, 3, 1])

ARGS = {
    "mat": mat,
    "perm": perm
}
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 (4, 4) — shape (4, 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
svm_read_check_override_config Ran
ran on a synthesized input
madrylab/failure-directions/failure_directions/src/config_parsing.py
code served (permissive licence) · get_code("2ad398a336dccf5d")
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 dict. 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

# Construct valid arguments for svm_read_check_override_config
config = {
    'method': 'SVM',
    'embedding': 'clip',
    'whiten': False,
    'svm_args': {
        'split_and_search': False,
        'balanced': False,
        'cv': 5
    }
}

override_args = [
    ('method', 'SVM_Linear'),
    ('whiten', True)
]

ARGS = {
    'config': config,
    'override_args': override_args
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
create_dataset_split Not yet run
unverified (harvested, no recorded run)
madrylab/failure-directions/failure_directions/src/ds_utils.py
code served (permissive licence) · get_code("ce1c8775425db385")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 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 7 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

Existing methods for isolating hard subpopulations and spurious correlations in datasets often require human intervention. This can make these methods labor-intensive and dataset-specific. To address these shortcomings, we present a scalable method for automatically distilling a model's failure modes. Specifically, we harness linear classifiers to identify consistent error patterns, and, in turn, induce a natural representation of these failure modes as directions within the feature space. We demonstrate that this framework allows us to discover and automatically caption challenging subpopulations within the training dataset. Moreover, by combining our framework with off-the-shelf diffusion models, we can generate images that are especially challenging for the analyzed model, and thus can be used to perform synthetic data augmentation that helps remedy the model's failure modes. Code available at https://github.com/MadryLab/failure-directions

For agents

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

get_harvested_code_for_paper("2206.14754")
get_code_for_paper("2206.14754")
have("2206.14754")

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

curl https://syntology.ai/api/ran/2206.14754.json

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

[![Syntology run record](https://syntology.ai/api/ran/2206.14754.svg)](https://syntology.ai/paper/2206.14754)

Connect an agent — have() is free.