SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2511.22305 · NeurIPS · 2025

FLUX: Efficient Descriptor-Driven Clustered Federated Learning under Arbitrary Distribution Shifts

Pietro Barbiero, Nicholas Lane, Dario Fenoglio, Martin Gjoreski, Marc Langheinrich, Mohan Li

arXiv · PDF · Open in the Atlas

Code that ran

3 of 23 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 0 have no contract verdict; 3 failures of our instrument; 20 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/2511.22305.json · badge: /api/ran/2511.22305.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
alfredoLimo/ANDA canonical (classifier label)
classifier basis: census
3 ran · 20 no recorded run
FunctionStatusWhere it lives
merge_data Ran
ran; our contract draft was wrong, not the code
alfredoLimo/ANDA/utils.py
code served (permissive licence) · get_code("891ad5c80364290a")
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 list. 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 sample client data dictionaries
# Each client has train/test features and labels
client_1_data = {
    'train_features': torch.randn(2, 4),  # 2 samples, 4 features
    'train_labels': torch.randint(0, 2, (2,)),  # 2 labels
    'test_features': torch.randn(2, 4),  # 2 test samples, 4 features
    'test_labels': torch.randint(0, 2, (2,))  # 2 test labels
}

client_2_data = {
    'train_features': torch.randn(2, 4),  # 2 samples, 4 features
    'train_labels': torch.randint(0, 2, (2,)),  # 2 labels
    'test_features': torch.randn(2, 4),  # 2 test samples, 4 features
    'test_labels': torch.randint(0, 2, (2,))  # 2 test labels
}

# Create list of client data
data = [client_1_data, client_2_data]

ARGS = {
    'data': data
}
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 list — a list 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
rotate_dataset Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
alfredoLimo/ANDA/utils.py
code served (permissive licence) · get_code("f25dc7f07588f5ef")
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, 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
from torchvision import transforms

torch.manual_seed(0)

# Create a small dataset of tiny images
# The function expects a tensor of shape (N,) where each element is an image tensor
# From the code: transforms.ToPILImage()(img_tensor) converts each element
# and transforms.ToTensor()(rotated_img).squeeze(0) suggests images are 2D after squeeze
# So we need image tensors of shape (C, H, W) where C=3 (RGB), H=4, W=3

# Create 2 sample images with shape (3, 4, 3) - 3 channels, 4 height, 3 width
img1 = torch.rand(3, 4, 3)
img2 = torch.rand(3, 4, 3)

# Stack them into a dataset of shape (2,) containing image tensors
dataset = torch.stack([img1, img2])

# Create corresponding rotation degrees
degrees = [45, 90]

ARGS = {
    'dataset': dataset,
    'degrees': degrees
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — 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, 3) — shape (2, 3, 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
split_trND_teDR_Py Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
alfredoLimo/ANDA/split_fn_trND_teDR.py
code served (permissive licence) · get_code("3e9ecc25052592ac")
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 list, 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 numpy as np

torch.manual_seed(0)
np.random.seed(0)

# Create small sample tensors
# train_features: (8, 4) - 8 samples, 4 features
train_features = torch.randn(8, 4)
# train_labels: (8,) - 8 labels, 3 classes (0, 1, 2)
train_labels = torch.tensor([0, 1, 2, 0, 1, 2, 0, 1])
# test_features: (6, 4) - 6 samples, 4 features
test_features = torch.randn(6, 4)
# test_labels: (6,) - 6 labels, 3 classes (0, 1, 2)
test_labels = torch.tensor([0, 1, 2, 0, 1, 2])

ARGS = {
    'train_features': train_features,
    'train_labels': train_labels,
    'test_features': test_features,
    'test_labels': test_labels,
    'client_number': 2,
    'scaling_low': 0.5,
    'scaling_high': 0.8,
    'reverse_test': False,
    'verbose': False,
}
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 list, ndim 1, dtype family float — a list 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
load_full_datasets Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/utils.py
code served (permissive licence) · get_code("c22f37ca60d19d5c")
How it ran

no invocation recorded.

load_split_datasets Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/anda.py
code served (permissive licence) · get_code("8a90e00fbc6438a2")
How it ran

no invocation recorded.

load_split_datasets_dynamic Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/anda.py
code served (permissive licence) · get_code("1ac45089a2626181")
How it ran

no invocation recorded.

split_feature_label_skew Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn.py
code served (permissive licence) · get_code("16b64a18e3b128ea")
How it ran

no invocation recorded.

split_feature_skew Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn.py
code served (permissive licence) · get_code("b3f9f679b86f0f1a")
How it ran

no invocation recorded.

split_label_skew Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn.py
code served (permissive licence) · get_code("34f5c3f4ed761f12")
How it ran

no invocation recorded.

split_trDA_teDR_Px Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDA_teDR.py
code served (permissive licence) · get_code("1f69dcd04d3f2669")
How it ran

no invocation recorded.

split_trDA_teDR_Py Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDA_teDR.py
code served (permissive licence) · get_code("5ac81bec1fce01f8")
How it ran

no invocation recorded.

split_trDA_teDR_Py_x Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDA_teDR.py
code served (permissive licence) · get_code("ea6e92d7f70005eb")
How it ran

no invocation recorded.

split_trDA_teND_Px Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDA_teND.py
code served (permissive licence) · get_code("df47519878157bdc")
How it ran

no invocation recorded.

split_trDA_teND_Py Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDA_teND.py
code served (permissive licence) · get_code("1f13b30ca7272d28")
How it ran

no invocation recorded.

split_trDA_teND_Py_x Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDA_teND.py
code served (permissive licence) · get_code("954e8bca36512ae7")
How it ran

no invocation recorded.

split_trDR_teDR_Px Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDR_teDR.py
code served (permissive licence) · get_code("52b9d39b5ebfd351")
How it ran

no invocation recorded.

split_trDR_teDR_Py Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDR_teDR.py
code served (permissive licence) · get_code("8b5656660ffeffc5")
How it ran

no invocation recorded.

split_trDR_teDR_Py_x Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDR_teDR.py
code served (permissive licence) · get_code("e62892cf1a395368")
How it ran

no invocation recorded.

split_trDR_teND_Px Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDR_teND.py
code served (permissive licence) · get_code("3b2a66cafa3dde80")
How it ran

no invocation recorded.

split_trDR_teND_Py Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDR_teND.py
code served (permissive licence) · get_code("fe9c2fef71e10aed")
How it ran

no invocation recorded.

split_trDR_teND_Py_x Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trDR_teND.py
code served (permissive licence) · get_code("3dc06bc3aaf65d01")
How it ran

no invocation recorded.

split_trND_teDR_Px Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trND_teDR.py
code served (permissive licence) · get_code("843c09ae939ba769")
How it ran

no invocation recorded.

split_trND_teDR_Py_x Not yet run
unverified (harvested, no recorded run)
alfredoLimo/ANDA/split_fn_trND_teDR.py
code served (permissive licence) · get_code("2c85ea3b5c8a86a3")
How it ran

no invocation recorded.

Also present in other papers' code

Of 23 samples lifted from this paper's linked repositories, 23 carry a function identity and 0 do not. Of the 23 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 23 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

Federated Learning (FL) enables collaborative model training across multiple clients while preserving data privacy. Traditional FL methods often use a global model to fit all clients, assuming that clients' data are independent and identically distributed (IID). However, when this assumption does not hold, the global model accuracy may drop significantly, limiting FL applicability in real-world scenarios. To address this gap, we propose FLUX, a novel clustering-based FL (CFL) framework that addresses the four most common types of distribution shifts during both training and test time. To this end, FLUX leverages privacy-preserving client-side descriptor extraction and unsupervised clustering to ensure robust performance and scalability across varying levels and types of distribution shifts. Unlike existing CFL methods addressing non-IID client distribution shifts, FLUX i) does not require any prior knowledge of the types of distribution shifts or the number of client clusters, and ii) supports test-time adaptation, enabling unseen and unlabeled clients to benefit from the most suitable cluster-specific models. Extensive experiments across four standard benchmarks, two real-world datasets and ten state-of-theart baselines show that FLUX improves performance and stability under diverse distribution shifts-achieving an average accuracy gain of up to 23 percentage points over the best-performing baselines-while maintaining computational and communication overhead comparable to FedAvg.

For agents

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

get_harvested_code_for_paper("2511.22305")
get_code_for_paper("2511.22305")
have("2511.22305")

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

curl https://syntology.ai/api/ran/2511.22305.json

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

[![Syntology run record](https://syntology.ai/api/ran/2511.22305.svg)](https://syntology.ai/paper/2511.22305)

Connect an agent — have() is free.