SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2502.11986 · ICLR · 2025

Selective Task Group Updates for Multi-Task Optimization

Kuk-Jin Yoon, Wooseong Jeong

arXiv · PDF · Open in the Atlas

Code that ran

6 of 16 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, 1 violated it and 2 have no contract verdict; 3 failures of our instrument; 10 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/2502.11986.json · badge: /api/ran/2502.11986.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
wooseong97/sel-update-mtl canonical (classifier label)
classifier basis: classifier_2025
6 ran · 10 no recorded run
FunctionStatusWhere it lives
MetricLogger Ran
ran on a synthesized input · constructed an object; no result was computed
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("f5627726b4ae2bcc")
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, MetricLogger: the run constructed the object and did not call it, so no result was computed. 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 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 time
from collections import defaultdict, deque

torch.manual_seed(0)

ARGS = {
    'delimiter': '\t'
}
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 MetricLogger — 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 MetricLogger 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
SmoothedValue Ran
ran on a synthesized input · constructed an object; no result was computed
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("79d5ce6f2d2f4574")
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, SmoothedValue: the run constructed the object and did not call it, so no result was computed. 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 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 collections import defaultdict, deque

torch.manual_seed(0)

ARGS = {
    'window_size': 20,
    'fmt': '{median:.4f} ({global_avg:.4f})'
}
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 SmoothedValue — 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 SmoothedValue 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_loss Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("32835f4d0f551b1b")
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)

# Test case 1: classification task
outputs_class = torch.randn(2, 1)
targets_class = torch.randn(2, 1)
task_class = 'class_task'

# Test case 2: semantic segmentation task
outputs_seg = torch.randn(2, 5, 4, 4)  # batch, num_classes, height, width
targets_seg = torch.randint(0, 5, (2, 4, 4))  # batch, height, width with class indices
task_seg = 'segment_semantic'

# Test case 3: normal prediction task
outputs_normal = torch.randn(2, 3, 4, 4)
targets_normal = torch.randn(2, 3, 4, 4)
task_normal = 'normal'

# Test case 4: depth task
outputs_depth = torch.randn(2, 1, 4, 4)
targets_depth = torch.randn(2, 4, 4)
task_depth = 'depth'

# Test case 5: keypoint task
outputs_keypoint = torch.randn(2, 1, 4, 4)
targets_keypoint = torch.randn(2, 4, 4)
task_keypoint = 'keypoint'

# Test case 6: edge task
outputs_edge = torch.randn(2, 1, 4, 4)
targets_edge = torch.randn(2, 4, 4)
task_edge = 'edge'

# Test case 7: default MSE task
outputs_default = torch.randn(2, 1, 4, 4)
targets_default = torch.randn(2, 4, 4)
task_default = 'unknown_task'

ARGS = {
    'outputs': outputs_class,
    'targets': targets_class,
    'task': task_class
}
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
get_metric Ran
ran, violated the contract we drafted
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("4799ae2084413fda")
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 float, 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)

# Mock function for compute_depth_errors since it's called but not defined
def compute_depth_errors(outputs, targets):
    return F.mse_loss(outputs, targets)

# Test case 1: classification task
outputs_class = torch.randn(2, 4)  # batch_size=2, num_classes=4
targets_class = torch.zeros(2, 4)
targets_class[0, 1] = 1.0
targets_class[1, 2] = 1.0

# Test case 2: depth_euclidean task
outputs_depth = torch.randn(2, 3, 4, 5)  # batch_size=2, channels=3, height=4, width=5
targets_depth = torch.randn(2, 3, 4, 5)

# Test case 3: curvature task
outputs_curv = torch.randn(2, 6, 7, 8)  # batch_size=2, channels=6, height=7, width=8
targets_curv = torch.randn(2, 6, 7, 8)

# Test case 4: generic task (2D tensors)
outputs_generic = torch.randn(2, 9)
targets_generic = torch.randn(2, 9)

ARGS = {
    "outputs": outputs_class,
    "targets": targets_class,
    "task": "class"
}
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 float, 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
get_output Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("a64ab65acb301c2d")
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, 5, 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 torch.nn.functional as F

torch.manual_seed(0)

# Test case 1: normals task
output_normals = torch.randn(2, 4, 3, 5)
task_normals = 'normals'

# Test case 2: semseg task
output_semseg = torch.randn(2, 8, 3, 5)
task_semseg = 'semseg'

# Test case 3: human_parts task
output_human_parts = torch.randn(2, 6, 3, 5)
task_human_parts = 'human_parts'

# Test case 4: edge task
output_edge = torch.randn(2, 1, 3, 5)
task_edge = 'edge'

# Test case 5: sal task
output_sal = torch.randn(2, 2, 3, 5)
task_sal = 'sal'

# Test case 6: depth task
output_depth = torch.abs(torch.randn(2, 1, 3, 5))
task_depth = 'depth'

# Use normals as the primary test case
ARGS = {
    'output': output_normals,
    'task': task_normals
}
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, 5, 4) — shape (2, 3, 5, 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
to_cuda Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("48fca14a672a5108")
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. 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 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)

# The function accepts a single parameter 'batch' which can be:
# - A dict (with special handling for 'meta' key)
# - A torch.Tensor
# - A list
# - Other types (returned as-is)

# Create test data structures that exercise all code paths
tensor_1 = torch.randn(2, 4)
tensor_2 = torch.randn(3, 5)

# Test case 1: dict with tensors and 'meta' key
batch_dict = {
    'data': tensor_1,
    'labels': tensor_2,
    'meta': {'info': 'should_not_be_moved'}
}

# Test case 2: nested structures
batch_nested = {
    'tensors': [tensor_1, tensor_2],
    'single': tensor_1,
    'meta': {'preserved': True}
}

# Test case 3: list of tensors
batch_list = [tensor_1, tensor_2, tensor_1]

# Test case 4: simple tensor
batch_tensor = tensor_1

# For demonstration, we'll use the nested dict case as the primary test
ARGS = {
    'batch': batch_nested
}
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
PerformanceMeter Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("05242b3a981049cf")
How it ran

no invocation recorded.

evaluate Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("cc95be4bcb3a1abd")
How it ran

no invocation recorded.

get_single_task_meter Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("6a216e80da829b4d")
How it ran

no invocation recorded.

mkdir_if_missing Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("3cc281449dc70540")
How it ran

no invocation recorded.

save_model_pred_for_one_task Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("c5851b0ba58c9bd3")
How it ran

no invocation recorded.

tb_update_perf Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("3d53794e86af79d4")
How it ran

no invocation recorded.

test_phase Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("ceea43c4d7d88aca")
How it ran

no invocation recorded.

train_phase_affinity Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("4e9930eb324564f2")
How it ran

no invocation recorded.

update_affinity Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("6e881abbc3437293")
How it ran

no invocation recorded.

update_tb Not yet run
unverified (harvested, no recorded run)
wooseong97/sel-update-mtl/sel-update-mtl/utils/train_utils.py
code served (permissive licence) · get_code("07aee3dd8e23ca9c")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 3 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 2 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
MetricLogger
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2607.02447 (2026-07)
Understanding the Robustness of Distributed Self-Supervised Learning Frameworks Against Non-IID Data
SmoothedValue
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2607.02447 (2026-07)
Understanding the Robustness of Distributed Self-Supervised Learning Frameworks Against Non-IID Data
get_single_task_meter
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2203.07997 (2022-03)
Inverted Pyramid Multi-task Transformer for Dense Scene Understanding

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

Multi-task learning enables the acquisition of task-generic knowledge by training multiple tasks within a unified architecture. However, training all tasks together in a single architecture can lead to performance degradation, known as negative transfer, which is a main concern in multi-task learning. Previous works have addressed this issue by optimizing the multi-task network through gradient manipulation or weighted loss adjustments. However, their optimization strategy focuses on addressing task imbalance in shared parameters, neglecting the learning of task-specific parameters. As a result, they show limitations in mitigating negative transfer, since the learning of shared space and task-specific information influences each other during optimization. To address this, we propose a different approach to enhance multi-task performance by selectively grouping tasks and updating them for each batch during optimization. We introduce an algorithm that adaptively determines how to effectively group tasks and update them during the learning process. To track inter-task relations and optimize multi-task networks simultaneously, we propose proximal inter-task affinity, which can be measured during the optimization process. We provide a theoretical analysis on how dividing tasks into multiple groups and updating them sequentially significantly affects multi-task performance by enhancing the learning of task-specific parameters. Our methods substantially outperform previous multi-task optimization approaches and are scalable to different architectures and various numbers of tasks. Our implementation is publicly available at https://github.com/wooseong97/sel-update-mtl

For agents

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

get_harvested_code_for_paper("2502.11986")
get_code_for_paper("2502.11986")
have("2502.11986")

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

curl https://syntology.ai/api/ran/2502.11986.json

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

[![Syntology run record](https://syntology.ai/api/ran/2502.11986.svg)](https://syntology.ai/paper/2502.11986)

Connect an agent — have() is free.