SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2503.09707 · NeurIPS · 2025

Revisiting Semi-Supervised Learning in the Era of Foundation Models

Wei-Lun Chao, Zheda Mai, Ping Zhang, Quang-Huy Nguyen

arXiv · PDF · Open in the Atlas

Code that ran

2 of 4 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; 2 failures of our instrument; 2 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/2503.09707.json · badge: /api/ran/2503.09707.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
osu-mlb/ssl-foundation-models canonical (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
OSU-MLB/SSL-Foundation-Models marked official in Papers with Code
Papers with Code archive
0 ran · 2 no recorded run
FunctionStatusWhere it lives
expand_fields Ran
ran; our contract draft was wrong, not the code
osu-mlb/ssl-foundation-models/gen_config_pet.py
code served (permissive licence) · get_code("9041c26dbe464d45")
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) 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 itertools

def expand_fields(d):
    keys = list(d.keys())
    values = list(d.values())
    combinations = itertools.product(*values)
    result = [{key: value for key, value in zip(keys, combination)} for combination in combinations]
    return result

# Construct valid arguments for expand_fields
# The function takes a dictionary where values are iterables
# It creates all combinations of values across keys

ARGS = {
    "d": {
        "color": ["red", "blue", "green"],
        "size": ["small", "large"],
        "quantity": [1, 2]
    }
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — 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
get_pet_sources Ran
ran; our contract draft was wrong, not the code
osu-mlb/ssl-foundation-models/gen_config_pet.py
code served (permissive licence) · get_code("64340821a2d9ddd8")
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) and returned tuple. 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 os
import numpy as np
from collections import defaultdict
import tempfile
import shutil

# Create a temporary directory structure for config files
temp_dir = tempfile.mkdtemp()

def create_config_file(peft, dset, shot, net, hyperparams):
    """Helper to create dummy config files"""
    path = os.path.join(temp_dir, f"config/{peft}/supervised/{dset}/{shot}/{net}/{'/'.join(hyperparams)}")
    os.makedirs(path, exist_ok=True)
    config_file = os.path.join(path, "config.yaml")
    with open(config_file, 'w') as f:
        f.write("# dummy config\n")
    return config_file

# Create sample rows with structure: [peft, alg, dset, shot, net, hyperparams, metric1-7]
rows = [
    # Self-training candidate rows
    ['lora', 'supervised', 'cifar10', '5', 'resnet18', "['lr_0.001', 'bs_32']", 0.85, 0.87, 0.86, 0.88, 0.84, 0.89, 0.86],
    ['lora', 'supervised', 'cifar10', '5', 'resnet18', "['lr_0.0005', 'bs_16']", 0.82, 0.84, 0.83, 0.85, 0.81, 0.86, 0.83],
    ['lora', 'supervised', 'cifar10', '5', 'resnet18', "['lr_0.002', 'bs_64']", 0.80, 0.82, 0.81, 0.83, 0.79, 0.84, 0.81],
    
    # Ensembled candidate rows (note: typo in code checks 'supervisped')
    ['lora', 'supervisped', 'cifar10', '5', 'resnet18', "['lr_0.001', 'bs_32']", 0.85, 0.87, 0.86, 0.88, 0.84, 0.89, 0.86],
    ['lora', 'supervisped', 'cifar10', '5', 'resnet18', "['lr_0.0005', 'bs_16']", 0.82, 0.84, 0.83, 0.85, 0.81, 0.86, 0.83],
    
    # Ensembled-across-nets candidate rows
    ['lora', 'supervised', 'cifar10', '5', 'resnet18', "['lr_0.001', 'bs_32']", 0.85, 0.87, 0.86, 0.88, 0.84, 0.89, 0.86],
    ['lora', 'supervised', 'cifar10', '5', 'resnet50', "['lr_0.0005', 'bs_16']", 0.82, 0.84, 0.83, 0.85, 0.81, 0.86, 0.83],
    ['lora', 'supervised', 'cifar10', '5', 'vit', "['lr_0.002', 'bs_64']", 0.80, 0.82, 0.81, 0.83, 0.79, 0.84, 0.81],
]

# Create the config files
for row in rows:
    peft, alg, dset, shot, net, hyperparams_str, *metrics = row
    hyperparams = eval(hyperparams_str)
    create_config_file(peft, dset, shot, net, hyperparams)

# Update the config path in os module to use our temp directory
original_exists = os.path.exists
def mock_exists(path):
    if path.startswith("config/"):
        return original_exists(os.path.join(temp_dir, path))
    return original_exists(path)

os.path.exists = mock_exists

ARGS = {
    'strategy': 'self-training',
    'dset': 'cifar10',
    'shot': '5',
    'peft': 'lora',
    'net': 'resnet18',
    'rows': rows,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple 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
Hook Not yet run
unverified (harvested, no recorded run)
OSU-MLB/SSL-Foundation-Models/semilearn/algorithms/hooks/pseudo_label.py
code served (permissive licence) · get_code("6b2071ea8351df61")
How it ran

no invocation recorded.

PseudoLabelingHook Not yet run
unverified (harvested, no recorded run)
OSU-MLB/SSL-Foundation-Models/semilearn/algorithms/hooks/pseudo_label.py
code served (permissive licence) · get_code("dd08e159932d775a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 3 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
Hook
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2407.15383 (2024-07)
Is user feedback always informative? Retrieval Latent Defending for Semi-Supervised Domain Adaptation without Source Data

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

Semi-supervised learning (SSL) enhances model performance by leveraging abundant unlabeled data alongside limited labeled data. As vision foundation models (VFMs) become central to modern vision applications, this paper revisits SSL in the context of these powerful pre-trained models. We conduct a systematic study on tasks where frozen VFMs underperform and reveal several key insights when fine-tuning them. First, parameter-efficient fine-tuning (PEFT) using only labeled data often surpasses traditional SSL methods-even without access to unlabeled data. Second, pseudo-labels generated by PEFT models offer valuable supervisory signals for unlabeled data, and different PEFT techniques yield complementary pseudo-labels. These findings motivate a simple yet effective SSL baseline for the VFM era: ensemble pseudo-labeling across diverse PEFT methods and VFM backbones. Extensive experiments validate the effectiveness of this approach, offering actionable insights into SSL with VFMs and paving the way for more scalable and robust semi-supervised learning in the foundation model era. Recently, pre-training on external labeled or unlabeled data has become the de facto standard across many machine learning application domains [29,46]. For example, in computer vision, many recent algorithms are built upon vision foundation models (VFMs) such as CLIP [55] and DINOv2 [53]. These models were pre-trained on massive datasets-millions, if not billions, of data points. As a result, they have demonstrated remarkable generalizability across a wide range of tasks, often requiring only slight fine-tuning or, in some cases, functioning effectively as a frozen backbone [79]. Given the promising advancements in both fields, we explore their interaction in this paper. Specifically, we seek to address the following questions: Are existing SSL algorithms still effective when using VFMs as the backbone? What adjustments, if any, are needed to improve their performance? Finally, can we leverage the power of VFMs to develop more effective, yet simpler, SSL algorithms?

For agents

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

get_harvested_code_for_paper("2503.09707")
get_code_for_paper("2503.09707")
have("2503.09707")

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

curl https://syntology.ai/api/ran/2503.09707.json

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

[![Syntology run record](https://syntology.ai/api/ran/2503.09707.svg)](https://syntology.ai/paper/2503.09707)

Connect an agent — have() is free.