SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2111.03602 · NeurIPS · 2021

NAS-Bench-x11 and the Power of Learning Curves

Frank Hutter, Shen Yan, Colin White, Yash Savani

arXiv · PDF · Open in the Atlas

Code that ran

4 of 6 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 2 failures of our instrument; 2 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/2111.03602.json · badge: /api/ran/2111.03602.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
automl/nas-bench-x11 canonical (classifier label)
classifier basis: classifier_2025
4 ran · 2 no recorded run
FunctionStatusWhere it lives
ParametricModel Ran
ran on a synthesized input · constructed an object; no result was computed
automl/nas-bench-x11/naslib/predictors/lce/lce.py
code served (permissive licence) · get_code("36c51e35267ab6c9")
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 an instance of its own class, ParametricModel: the run constructed the object and did not call it, so no result was computed. 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

class SimpleModel:
    """A simple parametric model that takes parameters and returns a callable."""
    def __call__(self, parameters):
        """
        Returns a function that takes x and returns predictions.
        parameters: array of coefficients [a, b, c, ...] where last one is often variance
        """
        def predict_fn(x):
            # Simple linear model: y = parameters[0] * x + parameters[1] + noise_param
            # The last parameter is treated as variance/noise scale
            x_arr = np.asarray(x)
            result = parameters[0] * x_arr + parameters[1]
            return result
        return predict_fn

# Create test data
np.random.seed(42)
x_data = np.random.randn(10)
y_data = 2.0 * x_data + 1.0 + 0.1 * np.random.randn(10)

ARGS = {
    "model_class": SimpleModel(),
    "degrees_freedom": 2,
    "name": "test_model",
    "bounds": None
}
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 ParametricModel — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ParametricModel 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
Predictor Ran
ran on a synthesized input · constructed an object; no result was computed
automl/nas-bench-x11/naslib/predictors/lce/lce.py
code served (permissive licence) · get_code("962e3193f835c354")
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 an instance of its own class, Predictor: the run constructed the object and did not call it, so no result was computed. 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)

class StandInArchitecture:
    """Stand-in class to represent a neural network architecture"""
    def __init__(self, arch_id):
        self.arch_id = arch_id
        self.params = {"layers": 3, "units": 64}

# Create small tensor data for training and testing
xtrain = [StandInArchitecture(i) for i in range(2)]
ytrain = [0.85, 0.90]

xtest = [StandInArchitecture(i) for i in range(2, 4)]

info_train = {
    "partial_lc": [0.5, 0.7, 0.85],
    "epochs": 20,
    "architecture_hyperparams": {"lr": 0.001, "batch_size": 32}
}

info_test = {
    "partial_lc": [0.4, 0.6, 0.8],
    "epochs": 20,
    "architecture_hyperparams": {"lr": 0.001, "batch_size": 32}
}

hyperparams = {
    "predictor_lr": 0.01,
    "predictor_epochs": 100,
    "regularization": 0.001
}

ARGS = {
    "ss_type": "reinforce",
    "encoding_type": "path"
}
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 Predictor — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Predictor 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
construct_parametric_model Ran
ran; our contract draft was wrong, not the code
automl/nas-bench-x11/naslib/predictors/lce/lce.py
code served (permissive licence) · get_code("6795115c8f12bbd6")
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 ParametricModel. 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
from scipy.optimize import minimize


class SimpleModel:
    """A simple parametric model that takes parameters and returns predictions"""
    def __init__(self, params):
        self.params = params
    
    def __call__(self, x):
        # Simple linear model: y = params[0] * x + params[1]
        # params[-1] is treated as noise variance
        if isinstance(x, list):
            x = np.array(x)
        return self.params[0] * x + self.params[1]


class ParametricModel:

    def __init__(self, model_class, degrees_freedom, name, bounds=None):
        self.model_class = model_class
        self.degrees_freedom = degrees_freedom
        self.name = name
        self.bounds = bounds

    def fit(self, x):
        self.x = x
        self.params = optimize_model_class(self.model_class, self.degrees_freedom, list(range(1, x.shape[0] + 1)), x, bounds=self.bounds)

    def get_params(self):
        return self.params

    def set_params(self, params):
        self.params = params

    def predict(self, x, params=None):
        if params is not None:
            return self.model_class(params)(x)
        return self.model_class(self.params)(x)


def optimize_model_class(fn_class, deg_freedom, x, y, bounds=None):

    def likelihood(parameters):
        return len(x) / 2 * np.log(2 * np.pi) + len(x) / 2 * np.log(parameters[-1] ** 2) + 1 / (2 * parameters[-1] ** 2) * sum((y - fn_class(parameters)(x)) ** 2)
    opt_model = minimize(likelihood, np.array([1] * (deg_freedom + 1)), method='L-BFGS-B', bounds=bounds)
    return opt_model['x']


def construct_parametric_model(model_config, model_name):
    return ParametricModel(model_config[model_name]['model'], model_config[model_name]['deg_freedom'], model_name, model_config[model_name]['bounds'])


ARGS = {
    'model_config': {
        'linear_model': {
            'model': SimpleModel,
            'deg_freedom': 2,
            'bounds': [(-10, 10), (-10, 10), (0.1, 10)]
        }
    },
    'model_name': 'linear_model'
}
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 ParametricModel — a ParametricModel output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ParametricModel 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
optimize_model_class Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
automl/nas-bench-x11/naslib/predictors/lce/lce.py
code served (permissive licence) · get_code("1eddae36959d2bea")
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 (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
from scipy.optimize import minimize

# Define a simple callable class that acts as a model factory
class SimplePolynomialModel:
    """A simple model that takes parameters and returns a callable function."""
    def __init__(self, parameters):
        self.parameters = parameters
    
    def __call__(self, x):
        # Simple polynomial model: y = p[0] + p[1]*x + p[2]*x^2 + ... + p[n-1]*x^n
        # Last parameter is treated as noise variance
        result = self.parameters[0]
        for i in range(1, len(self.parameters) - 1):
            result = result + self.parameters[i] * (x ** i)
        return result

# Generate small synthetic data
np.random.seed(0)
x = np.linspace(0, 1, 5)
y = 2.0 + 3.0 * x + np.random.normal(0, 0.1, size=len(x))

# Define the function class that will be used
fn_class = SimplePolynomialModel

# Degrees of freedom (number of parameters excluding the noise variance)
deg_freedom = 2

# Bounds for the parameters (optional)
bounds = [(-10, 10), (-10, 10), (0.01, 10)]  # bounds for [intercept, slope, noise_variance]

ARGS = {
    'fn_class': fn_class,
    'deg_freedom': deg_freedom,
    'x': x,
    'y': y,
    'bounds': bounds
}
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 1, dtype family float, shape (3,) — shape (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
LCEPredictor Not yet run
unverified (harvested, no recorded run)
automl/nas-bench-x11/naslib/predictors/lce/lce.py
code served (permissive licence) · get_code("39838ad0aeb9ef75")
How it ran

no invocation recorded.

ParametricEnsemble Not yet run
unverified (harvested, no recorded run)
automl/nas-bench-x11/naslib/predictors/lce/lce.py
code served (permissive licence) · get_code("b6361e24f10c28f1")
How it ran

no invocation recorded.

Also present in other papers' code

Of 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 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 6 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

While early research in neural architecture search (NAS) required extreme computational resources, the recent releases of tabular and surrogate benchmarks have greatly increased the speed and reproducibility of NAS research. However, two of the most popular benchmarks do not provide the full training information for each architecture. As a result, on these benchmarks it is not possible to run many types of multi-fidelity techniques, such as learning curve extrapolation, that require evaluating architectures at arbitrary epochs. In this work, we present a method using singular value decomposition and noise modeling to create surrogate benchmarks, NAS-Bench-111, NAS-Bench-311, and NAS-Bench-NLP11, that output the full training information for each architecture, rather than just the final validation accuracy. We demonstrate the power of using the full training information by introducing a learning curve extrapolation framework to modify single-fidelity algorithms, showing that it leads to improvements over popular single-fidelity algorithms which claimed to be state-of-the-art upon release. Our code and pretrained models are available at https://github.com/automl/nas-bench-x11.

For agents

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

get_harvested_code_for_paper("2111.03602")
get_code_for_paper("2111.03602")
have("2111.03602")

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

curl https://syntology.ai/api/ran/2111.03602.json

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

[![Syntology run record](https://syntology.ai/api/ran/2111.03602.svg)](https://syntology.ai/paper/2111.03602)

Connect an agent — have() is free.