SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.25839 · 2026

A Compression Perspective on Simplicity Bias

Sarthak Mittal, Guillaume Lajoie, Mizu Nishikawa-Toomey, Eric Elmoznino, Dhanya Sridhar, Tejas Kasetty, Leo Gagnon, Tom Marty

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; 2 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 9 failures of our instrument; 1 no recorded run), measured 2026-11-22; 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/2603.25839.json · badge: /api/ran/2603.25839.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
3rdCore/complicity canonical (classifier label)
classifier basis: census
12 ran · 1 no recorded run
FunctionStatusWhere it lives
Classifier Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/networks.py
code served (permissive licence) · get_code("ce7990d7ad5821ff")
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 Linear. 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)

ARGS = {
    "in_features": 16,
    "out_features": 4,
    "is_nonlinear": False,
}
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 Linear — a Linear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Linear 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
Featurizer Ran
ran on a synthesized input
3rdCore/complicity/source/networks.py
code served (permissive licence) · get_code("ec18af6058c44a7b")
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 SimpleMLP. 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

torch.manual_seed(0)

# Test case 1: images with simple_mlp (3-channel 32x32 image)
ARGS = {
    'data_type': 'images',
    'input_shape': (3, 32, 32),
    'hparams': {
        'image_arch': 'simple_mlp',
        'n_outputs': 256,
        'mlp_width': 128,
        'mlp_depth': 3,
        'mlp_dropout': 0.1
    }
}
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 SimpleMLP — a SimpleMLP output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleMLP 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
binary_metrics Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
3rdCore/complicity/source/utils/eval_helper.py
code served (permissive licence) · get_code("3107c7b5cd6c8ce8")
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 numpy as np

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

# Create simple binary classification targets and predictions
targets = np.array([0, 1, 0, 1, 1, 0, 1, 0])
preds = np.array([0, 1, 0, 1, 0, 0, 1, 1])

# Define ARGS dictionary with all parameters
ARGS = {
    'targets': targets,
    'preds': preds,
    'label_set': [0, 1],
    'return_arrays': False
}
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 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
build_group_index Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/utils/misc.py
code served (permissive licence) · get_code("e6cb0453a0e5d9e0")
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 defaultdict. 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 collections import defaultdict

torch.manual_seed(0)

# Create a simple dataset class with labels and attributes
class SimpleDataset:
    def __init__(self, size=5, include_attrs=True):
        self.size = size
        self.labels = torch.tensor([0, 1, 0, 1, 0])[:size]
        if include_attrs:
            self.attributes = torch.tensor([0, 1, 1, 0, 1])[:size]
        else:
            self.attributes = None
    
    def __len__(self):
        return self.size
    
    def __getitem__(self, idx):
        # Return a 5-tuple: (x, metadata, y, a, extra)
        x = torch.randn(4)
        metadata = "sample"
        y = self.labels[idx]
        a = self.attributes[idx] if self.attributes is not None else 0
        extra = None
        return x, metadata, y, a, extra

# Test case 1: Dataset with labels and attributes attributes
ds1 = SimpleDataset(size=5, include_attrs=True)

ARGS = {
    "ds": ds1
}
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 defaultdict — a defaultdict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a defaultdict 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
denormalize_cmnist Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/utils/misc.py
code served (permissive licence) · get_code("cdf8a449ec6c94cf")
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 4, dtype family float, shape (2, 3, 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

torch.manual_seed(0)

# Create a small normalized CMNIST image tensor
# Shape: (batch=2, channels=3, height=4, width=4)
# Values in range [0, 1] as if they were normalized
img_tensor = torch.rand(2, 3, 4, 4)

ARGS = {
    "img_tensor": img_tensor
}
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 4, dtype family float, shape (2, 3, 4, 4) — shape (2, 3, 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
generate_watermark_banks Ran
ran, honoured the contract we drafted
3rdCore/complicity/source/datasets.py
code served (permissive licence) · get_code("829d3ac448ab1860")
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 3, 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 numpy as np

ARGS = {
    "n_bits": 8,
    "bank_size": 4,
    "seed": 42,
}
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, ndim 3, 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
get_adam_optim Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/algorithms.py
code served (permissive licence) · get_code("4268459b5b2f591b")
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 Adam. 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

torch.manual_seed(0)

# Create a minimal network with parameters
class SimpleNetwork(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

network = SimpleNetwork()
lr = 0.001
weight_decay = 1e-5

ARGS = {
    "network": network,
    "lr": lr,
    "weight_decay": weight_decay
}
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 Adam — a Adam output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Adam 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_adamw_optim Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/algorithms.py
code served (permissive licence) · get_code("96cdeeec1850a33e")
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 AdamW. 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 torch.optim import AdamW

torch.manual_seed(0)

class SimpleNetwork(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.bias = nn.Parameter(torch.randn(2))
        self.layer_norm = nn.LayerNorm(3)
    
    def forward(self, x):
        return self.layer_norm(self.linear1(x))

network = SimpleNetwork()
lr = 0.001
weight_decay = 0.01

ARGS = {
    'network': network,
    'lr': lr,
    'weight_decay': weight_decay
}
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 AdamW — a AdamW output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a AdamW 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_base_dataset Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/utils/misc.py
code served (permissive licence) · get_code("871300bf829619b4")
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 SimpleDataset. 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
from torch.utils.data import Subset, Dataset

torch.manual_seed(0)

class SimpleDataset(Dataset):
    def __init__(self, size=10):
        self.size = size
        self.data = torch.randn(size, 4)
        self.labels = torch.randint(0, 2, (size,))
    
    def __len__(self):
        return self.size
    
    def __getitem__(self, idx):
        return self.data[idx], self.labels[idx]

base_dataset = SimpleDataset(size=10)
subset1 = Subset(base_dataset, indices=[0, 1, 2, 3])
subset2 = Subset(subset1, indices=[0, 1])

ARGS = {
    "ds": subset2
}
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 SimpleDataset — a SimpleDataset 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_mean_log_loss_and_accuracy Ran
ran, honoured the contract we drafted
3rdCore/complicity/source/utils/notebook_helpers.py
code served (permissive licence) · get_code("20ffd34ae1886ee6")
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 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

torch.manual_seed(0)

class SimpleAlgoModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def predict(self, x, y, return_loss=False):
        logits = self.linear(x)
        if return_loss:
            loss_fn = nn.CrossEntropyLoss()
            loss = loss_fn(logits, y)
            return logits, loss
        return logits
    
    def eval(self):
        super().eval()
        return self

class SimpleEvalLoader:
    def __init__(self):
        self.data = [
            (0, torch.randn(2, 4), torch.randint(0, 3, (2,)), torch.randn(2), torch.randn(2), torch.randn(2)),
            (1, torch.randn(2, 4), torch.randint(0, 3, (2,)), torch.randn(2), torch.randn(2), torch.randn(2)),
        ]
    
    def __iter__(self):
        return iter(self.data)

algo_instance = SimpleAlgoModule()
eval_loader = SimpleEvalLoader()
device = 'cpu'

ARGS = {
    'algo_instance': algo_instance,
    'eval_loader': eval_loader,
    'device': device,
}
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 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
get_sgd_optim Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/algorithms.py
code served (permissive licence) · get_code("10a36b98f3cf81cf")
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 SGD. 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

torch.manual_seed(0)

# Create a minimal neural network
class SimpleNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Instantiate the network
network = SimpleNet()

# Define hyperparameters
lr = 0.01
weight_decay = 0.0001

# Construct arguments dict
ARGS = {
    "network": network,
    "lr": lr,
    "weight_decay": weight_decay
}
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 SGD — a SGD output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SGD 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
predict_on_set Ran
ran; our contract draft was wrong, not the code
3rdCore/complicity/source/utils/eval_helper.py
code served (permissive licence) · get_code("37040099f67bad05")
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 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)

# Create a minimal algorithm (model) that satisfies the function's usage
class MinimalAlgorithm(nn.Module):
    def __init__(self, input_dim=4, output_dim=2):
        super().__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim
        self.linear = nn.Linear(input_dim, output_dim)
    
    def predict(self, x):
        # Returns logits with shape (batch_size, output_dim)
        return self.linear(x)
    
    def eval(self):
        super().eval()

# Create a minimal dataset class
class MinimalDataset:
    def __init__(self, num_labels=2):
        self.num_labels = num_labels

# Create minimal batches - the function unpacks batches with 4, 5, or 6 elements
# Format: (_, x, y, a, [env], [digit])
def create_batch(batch_size=2, input_dim=4, num_labels=2, variant=4):
    """Create a batch tuple with variant number of elements (4, 5, or 6)"""
    _ = torch.tensor([0])  # Placeholder first element
    x = torch.randn(batch_size, input_dim)  # Input features
    y = torch.randint(0, num_labels, (batch_size,)).numpy()  # Labels
    a = torch.randint(0, 2, (batch_size,)).numpy()  # Attributes
    
    if variant == 4:
        return (_, x, y, a)
    elif variant == 5:
        digit = torch.randint(0, 10, (batch_size,)).numpy()
        return (_, x, y, a, digit)
    else:  # variant == 6
        env = torch.randint(0, 3, (batch_size,)).numpy()
        digit = torch.randint(0, 10, (batch_size,)).numpy()
        return (_, x, y, a, env, digit)

# Create a data loader (list of batches)
class MinimalDataLoader:
    def __init__(self, dataset, num_batches=2, batch_size=2, variant=4):
        self.dataset = dataset
        self.num_batches = num_batches
        self.batch_size = batch_size
        self.variant = variant
    
    def __iter__(self):
        for _ in range(self.num_batches):
            yield create_batch(self.batch_size, input_dim=4, num_labels=self.dataset.num_labels, variant=self.variant)

# Set up arguments
algorithm = MinimalAlgorithm(input_dim=4, output_dim=2)
dataset = MinimalDataset(num_labels=2)
loader = MinimalDataLoader(dataset, num_batches=2, batch_size=2, variant=4)
device = torch.device('cpu')

ARGS = {
    'algorithm': algorithm,
    'loader': loader,
    'device': device
}
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 tuple, ndim 2, 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
get_dataset_class Not yet run
unverified (harvested, no recorded run)
3rdCore/complicity/source/datasets.py
code served (permissive licence) · get_code("5b0092cb7767b24c")
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 13 distinct functions with an identity, 3 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 3 later) and 10 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
Classifier 24 other papers
23 dated earlier, 1 later · 4 only as an assembly brick
2106.02266 (2021-06)
SAND-mask: An Enhanced Gradient Masking Strategy for the Discovery of Invariances in Domain Generalization
2111.12853 (2021-11)
Domain Prompt Learning for Efficiently Adapting CLIP to Unseen Domains
2205.09739 (2022-05)
Diverse Weight Averaging for Out-of-Distribution Generalization
and 21 more not named here
get_dataset_class 3 other papers
2 dated earlier, 1 later
2302.12254 (2023-02)
Change is Hard: A Closer Look at Subpopulation Shift
2403.07241 (2024-03)
Calibrating Multi-modal Representations: A Pursuit of Group Robustness without Annotations
2606.00082 (2026-06)
Hoeffding Concept Bottleneck Models with Applications to Overhead Images
binary_metrics 1 other paper
1 later
2606.00082 (2026-06)
Hoeffding Concept Bottleneck Models with Applications to Overhead Images

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

Deep neural networks exhibit a simplicity bias, a well-documented tendency to favor simple functions over complex ones. In this work, we cast new light on this phenomenon through the lens of the Minimum Description Length principle, formalizing supervised learning as a problem of optimal two-part lossless compression. Our theory explains how simplicity bias governs feature selection in neural networks through a fundamental trade-off between model complexity (the cost of describing the hypothesis) and predictive power (the cost of describing the data). Our framework predicts that as the amount of available training data increases, learners transition through qualitatively different features -from simple spurious shortcuts to complex features -only when the reduction in data encoding cost justifies the increased model complexity. Consequently, we identify distinct data regimes where increasing data promotes robustness by ruling out trivial shortcuts, and conversely, regimes where limiting data can act as a form of complexity-based regularization, preventing the learning of unreliable complex environmental cues. We validate our theory on a semi-synthetic benchmark showing that the feature selection of neural networks follows the same trajectory of solutions as optimal two-part compressors.

For agents

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

get_harvested_code_for_paper("2603.25839")
get_code_for_paper("2603.25839")
have("2603.25839")

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

curl https://syntology.ai/api/ran/2603.25839.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.25839.svg)](https://syntology.ai/paper/2603.25839)

Connect an agent — have() is free.