SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2003.14058 · 2020

MTL-NAS: Task-Agnostic Neural Architecture Search towards General-Purpose Multi-Task Learning

arXiv · PDF · Open in the Atlas

Code that ran

9 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; 0 honoured the contract we drafted, 0 violated it and 7 have no contract verdict; 2 failures of our instrument; 4 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/2003.14058.json · badge: /api/ran/2003.14058.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
bhpfelix/MTLNAS canonical (classifier label)
classifier basis: census
9 ran · 4 no recorded run
FunctionStatusWhere it lives
batch_norm Ran
ran; our contract draft was wrong, not the code
bhpfelix/MTLNAS/core/models/common_layers.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from ethanygao/aux-nas, is pointer only (licence: NONE) · get_code("56af5b6a5714675c")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from ethanygao/aux-nas, and GitHub detected no licence in that repository 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.

compute_angle Ran
ran on a synthesized input
bhpfelix/MTLNAS/core/utils/metrics.py
code served (permissive licence) · get_code("27f37e43a9a8bf2a")
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 (38,) 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
import numpy as np

torch.manual_seed(0)

# Create prediction tensor: (batch=2, channels=3, height=4, width=5)
prediction = torch.randn(2, 3, 4, 5)

# Create gt tensor: (batch=2, channels=3, height=4, width=5)
# Need to ensure some pixels have all three channels equal to ignore_label
gt = torch.randint(0, 10, (2, 3, 4, 5)).float()

# Set some pixels to ignore_label (value 255) across all 3 channels
ignore_label = 255
gt[0, :, 0, 0] = ignore_label  # Set first pixel of first batch to ignore_label across all channels
gt[1, :, 1, 1] = ignore_label  # Set a pixel in second batch to ignore_label across all channels

# ignore_label value
ignore_label = 255

ARGS = {
    "prediction": prediction,
    "gt": gt,
    "ignore_label": ignore_label,
}
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 ndarray, ndim 1, dtype family float, shape (38,) — shape (38,) 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
get_nddr_bn Ran
ran on a synthesized input
bhpfelix/MTLNAS/core/models/common_layers.py
code served (permissive licence) · get_code("10b8ad22843e7d02")
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 function. 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)

class Config:
    class MODEL:
        NDDR_BN_TYPE = 'default'
        BATCH_NORM_MOMENTUM = 0.05

cfg = Config()

ARGS = {
    'cfg': cfg
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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_tasks Ran
ran on a synthesized input
bhpfelix/MTLNAS/core/tasks.py
code served (permissive licence) · get_code("23e12e7e82ed9859")
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. 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 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 numpy as np

torch.manual_seed(0)

class MockConfig:
    def __init__(self):
        self.TASK = 'pixel'
        self.DATASET = 'nyu_v2'
        self.MODEL = MockModel()

class MockModel:
    def __init__(self):
        self.NET1_CLASSES = 13

cfg = MockConfig()

ARGS = {
    'cfg': cfg
}
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 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
normal_loss Ran
ran on a synthesized input
bhpfelix/MTLNAS/core/utils/losses.py
code served (permissive licence) · get_code("408334df60d6a716")
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 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.functional as F

torch.manual_seed(0)

# prediction: output of CNN, Float type, shape [N, C, H, W]
# The function interpolates it to match gt's spatial dimensions
# gt: groundtruth, Float type, shape [N, C, H, W] where C=3 (normals)
# After interpolation and reshaping, both become [N*H*W, 3]

# Create small tensors with distinct dimensions:
# Batch size: 2
# Channels (normals): 3
# Height: 4
# Width: 5
# Prediction has different spatial dims to test interpolation

N, C, H, W = 2, 3, 4, 5
prediction_H, prediction_W = 2, 3

# prediction: [2, 3, 2, 3] - will be interpolated to [2, 3, 4, 5]
prediction = torch.randn(N, C, prediction_H, prediction_W, dtype=torch.float32)

# gt: [2, 3, 4, 5] - contains normal vectors
# Most values should be valid (not equal to ignore_label=255)
# Some should be marked as ignore (all channels = 255)
gt = torch.randn(N, C, H, W, dtype=torch.float32) * 10

# Mark some pixels to be ignored by setting all channels to ignore_label
gt[0, :, 0, 0] = 255.0  # First position in first batch: all channels = 255
gt[1, :, 1, 1] = 255.0  # One position in second batch: all channels = 255

# Ensure most pixels have valid values (not all channels = 255)
# Keep the random values for other pixels (already set above)

ARGS = {
    'prediction': prediction,
    'gt': gt,
    'ignore_label': 255
}
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 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
normalize Ran
ran on a synthesized input
bhpfelix/MTLNAS/core/utils/visualization.py
code served (permissive licence) · get_code("a3dc17b381d3804c")
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 (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

# The function normalize takes a single parameter 'x' which should be a numpy array
# It computes the L2 norm along axis 0 and normalizes by dividing element-wise

# Create a small numpy array with distinct dimensions for testing
# Using shape (4, 3) - 4 rows, 3 columns
x = np.random.randn(4, 3).astype(np.float32)

ARGS = {
    'x': x
}
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 (4, 3) — shape (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
poly Ran
ran; our contract draft was wrong, not the code
bhpfelix/MTLNAS/core/utils/losses.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from ethanygao/aux-nas, is pointer only (licence: NONE) · get_code("da5bbab470adc6d7")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from ethanygao/aux-nas, and GitHub detected no licence in that repository 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.

process_image Ran
ran on a synthesized input
bhpfelix/MTLNAS/core/utils/visualization.py
code served (permissive licence) · get_code("e85c27ee7db9da35")
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 4, dtype family int, shape (2, 3, 4, 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
import numpy as np

torch.manual_seed(0)

# The function expects:
# - image: a torch tensor that has .cpu() method and can be converted to numpy
# - image_mean: a numpy array that supports indexing with [:, None, None]

# Create a small image tensor (batch_size=2, channels=3, height=4, width=5)
image = torch.randn(2, 3, 4, 5)

# Create image_mean as a numpy array with shape (3,) for 3 channels
# The function indexes it as image_mean[:, None, None] to broadcast
image_mean = np.array([0.5, 0.4, 0.3])

ARGS = {
    "image": image,
    "image_mean": image_mean
}
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 ndarray, ndim 4, dtype family int, shape (2, 3, 4, 5) — shape (2, 3, 4, 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
seg_loss Ran
ran on a synthesized input
bhpfelix/MTLNAS/core/utils/losses.py
code served (permissive licence) · get_code("2bc85b26189ce633")
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 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)

# Create prediction tensor: (batch=2, channels=4, height=3, width=5)
# This will be interpolated to match gt spatial dimensions
prediction = torch.randn(2, 4, 3, 5)

# Create ground truth tensor: (batch=2, 1, height=6, width=7)
# The function will squeeze(1) to get (batch=2, height=6, width=7)
# which are the target class indices for cross_entropy
gt = torch.randint(0, 4, (2, 1, 6, 7))

# Set some values to ignore_label to test that functionality
gt[0, 0, 0, 0] = 255

ignore_label = 255

ARGS = {
    "prediction": prediction,
    "gt": gt,
    "ignore_label": ignore_label,
}
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 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
compute_hist Not yet run
unverified (harvested, no recorded run)
bhpfelix/MTLNAS/core/utils/metrics.py
code served (permissive licence) · get_code("ca9ac840bc586a01")
How it ran

no invocation recorded.

count_correct Not yet run
unverified (harvested, no recorded run)
bhpfelix/MTLNAS/core/utils/metrics.py
code served (permissive licence) · get_code("be1d4c78c6570d4f")
How it ran

no invocation recorded.

get_nddr Not yet run
unverified (harvested, no recorded run)
bhpfelix/MTLNAS/core/models/common_layers.py
code served (permissive licence) · get_code("fc79b5b14193af1e")
How it ran

no invocation recorded.

process_seg_label Not yet run
unverified (harvested, no recorded run)
bhpfelix/MTLNAS/core/utils/visualization.py
code served (permissive licence) · get_code("40e5c245b7465f9b")
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, 2 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 2 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
batch_norm 1 other paper
1 later
2405.05695 (2024-05)
Aux-NAS: Exploiting Auxiliary Labels with Negligibly Extra Inference Cost
poly 1 other paper
1 later
2405.05695 (2024-05)
Aux-NAS: Exploiting Auxiliary Labels with Negligibly Extra Inference Cost

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

We propose to incorporate neural architecture search (NAS) into general-purpose multi-task learning (GP-MTL). Existing NAS methods typically define different search spaces according to different tasks. In order to adapt to different task combinations (i.e., task sets), we disentangle the GP-MTL networks into single-task backbones (optionally encode the task priors), and a hierarchical and layerwise features sharing/fusing scheme across them. This enables us to design a novel and general task-agnostic search space, which inserts cross-task edges (i.e., feature fusion connections) into fixed single-task network backbones. Moreover, we also propose a novel single-shot gradient-based search algorithm that closes the performance gap between the searched architectures and the final evaluation architecture. This is realized with a minimum entropy regularization on the architecture weights during the search phase, which makes the architecture weights converge to near-discrete values and therefore achieves a single model. As a result, our searched model can be directly used for evaluation without (re-)training from scratch. We perform extensive experiments using different single-task backbones on various task sets, demonstrating the promising performance obtained by exploiting the hierarchical and layerwise features, as well as the desirable generalizability to different i) task sets and ii) single-task backbones. The code of our paper is available at https://github.com/bhpfelix/MTLNAS.

For agents

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

get_harvested_code_for_paper("2003.14058")
get_code_for_paper("2003.14058")
have("2003.14058")

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

curl https://syntology.ai/api/ran/2003.14058.json

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

[![Syntology run record](https://syntology.ai/api/ran/2003.14058.svg)](https://syntology.ai/paper/2003.14058)

Connect an agent — have() is free.