SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.15567 · 2024

Do not trust what you trust: Miscalibration in Semi-supervised Learning

arXiv · PDF · Open in the Atlas

Code that ran

12 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; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 10 failures of our instrument; 1 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/2403.15567.json · badge: /api/ran/2403.15567.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
shambhavicodes/miscalibration-ssl canonical (classifier label)
classifier basis: census
12 ran · 1 no recorded run
FunctionStatusWhere it lives
array_to_bytes Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/preprocess/preprocess_superb.py
pointer only (licence: MIT); the stored copy get_code serves, from aah94/JEPAMatch, is pointer only (licence: NOASSERTION) · get_code("dc61eede543c56fc")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from aah94/JEPAMatch, and GitHub could not classify the licence of that repository (NOASSERTION). 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.

average_calibration_error Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/metrics.py
code served (permissive licence) · get_code("518ef932a7896b3d")
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 math

COUNT = 'count'
CONF = 'conf'
ACC = 'acc'
BIN_ACC = 'bin_acc'
BIN_CONF = 'bin_conf'

# Create sample data for the function
# confs: list of confidence scores (floats between 0 and 1)
# preds: list of predicted labels (integers)
# labels: list of ground truth labels (integers)
# num_bins: number of bins for calibration (default 10)

confs = [0.9, 0.8, 0.7, 0.6, 0.95, 0.5, 0.85, 0.75]
preds = [0, 1, 0, 1, 0, 1, 1, 0]
labels = [0, 1, 0, 1, 0, 0, 1, 0]
num_bins = 10

ARGS = {
    'confs': confs,
    'preds': preds,
    'labels': labels,
    'num_bins': num_bins
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
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 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
bytes_to_array Ran
ran, honoured the contract we drafted
shambhavicodes/miscalibration-ssl/preprocess/preprocess_superb.py
pointer only (licence: MIT); the stored copy get_code serves, from aah94/JEPAMatch, is pointer only (licence: NOASSERTION) · get_code("b19e0120c50d4fd0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from aah94/JEPAMatch, and GitHub could not classify the licence of that repository (NOASSERTION). 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.

comatch_contrastive_loss Ran
ran, honoured the contract we drafted
shambhavicodes/miscalibration-ssl/semilearn/algorithms/comatch/comatch.py
code served (permissive licence) · get_code("3346c66c873fff6a")
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 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

torch.manual_seed(0)

# Analyze the function:
# - feats_x_ulb_s_0: tensor used in torch.mm(feats_x_ulb_s_0, feats_x_ulb_s_1.t())
#   Shape should be (batch_size, feature_dim)
# - feats_x_ulb_s_1: tensor used in torch.mm(..., feats_x_ulb_s_1.t())
#   Shape should be (batch_size, feature_dim)
# - Q: tensor used in element-wise multiplication with sim_probs
#   Shape should match sim_probs which is (batch_size, batch_size)
# - T: scalar temperature parameter (default 0.2)

batch_size = 2
feature_dim = 4

# Create feature tensors
feats_x_ulb_s_0 = torch.randn(batch_size, feature_dim)
feats_x_ulb_s_1 = torch.randn(batch_size, feature_dim)

# Create Q tensor - should be (batch_size, batch_size) for the multiplication
# Q represents pseudo-labels or target probabilities
Q = torch.randn(batch_size, batch_size)
Q = torch.softmax(Q, dim=1)  # Make it a valid probability distribution

T = 0.2

ARGS = {
    "feats_x_ulb_s_0": feats_x_ulb_s_0,
    "feats_x_ulb_s_1": feats_x_ulb_s_1,
    "Q": Q,
    "T": T,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
expected_calibration_error Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/metrics.py
code served (permissive licence) · get_code("2249c8712edfa8ce")
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 numpy as np
import math

# Set seed for reproducibility
torch.manual_seed(0)

# Analyze the function expected_calibration_error:
# Parameters:
# - confs: list/array of confidence scores (float values between 0 and 1)
# - preds: list/array of predictions (class indices)
# - labels: list/array of ground truth labels (class indices)
# - num_bins: integer number of bins (default 10)

# The function iterates through these arrays with len() and indexing [i]
# It treats confidence values as floats for math operations
# It compares predictions with labels for equality (==)

# Create small test data
num_samples = 8
num_classes = 3

# Confidence scores: values between 0 and 1
confs = [0.9, 0.85, 0.72, 0.65, 0.55, 0.48, 0.35, 0.92]

# Predictions: class indices
preds = [0, 1, 2, 0, 1, 2, 0, 1]

# Labels: ground truth class indices
labels = [0, 1, 2, 0, 2, 2, 0, 1]

# Number of bins for calibration error calculation
num_bins = 10

ARGS = {
    'confs': confs,
    'preds': preds,
    'labels': labels,
    'num_bins': num_bins
}
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 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
maximum_calibration_error Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/metrics.py
code served (permissive licence) · get_code("887e7bb16dc8c6b5")
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 math

COUNT = 'count'
CONF = 'conf'
ACC = 'acc'
BIN_ACC = 'bin_acc'
BIN_CONF = 'bin_conf'

def _bin_initializer(bin_dict, num_bins=10):
    for i in range(num_bins):
        bin_dict[i][COUNT] = 0
        bin_dict[i][CONF] = 0
        bin_dict[i][ACC] = 0
        bin_dict[i][BIN_ACC] = 0
        bin_dict[i][BIN_CONF] = 0

def _populate_bins(confs, preds, labels, num_bins=10):
    bin_dict = {}
    for i in range(num_bins):
        bin_dict[i] = {}
    _bin_initializer(bin_dict, num_bins)
    num_test_samples = len(confs)
    for i in range(0, num_test_samples):
        confidence = confs[i]
        prediction = preds[i]
        label = labels[i]
        binn = int(math.ceil(num_bins * confidence - 1))
        bin_dict[binn][COUNT] = bin_dict[binn][COUNT] + 1
        bin_dict[binn][CONF] = bin_dict[binn][CONF] + confidence
        bin_dict[binn][ACC] = bin_dict[binn][ACC] + (1 if label == prediction else 0)
    for binn in range(0, num_bins):
        if bin_dict[binn][COUNT] == 0:
            bin_dict[binn][BIN_ACC] = 0
            bin_dict[binn][BIN_CONF] = 0
        else:
            bin_dict[binn][BIN_ACC] = float(bin_dict[binn][ACC]) / bin_dict[binn][COUNT]
            bin_dict[binn][BIN_CONF] = bin_dict[binn][CONF] / float(bin_dict[binn][COUNT])
    return bin_dict

def maximum_calibration_error(confs, preds, labels, num_bins=10):
    bin_dict = _populate_bins(confs, preds, labels, num_bins)
    ce = []
    for i in range(num_bins):
        bin_accuracy = bin_dict[i][BIN_ACC]
        bin_confidence = bin_dict[i][BIN_CONF]
        ce.append(abs(bin_accuracy - bin_confidence))
    return max(ce)

# Setup code for valid arguments
confs = [0.9, 0.8, 0.7, 0.95, 0.85, 0.75, 0.92, 0.88]
preds = [0, 1, 0, 1, 0, 1, 0, 1]
labels = [0, 1, 0, 1, 1, 1, 0, 1]
num_bins = 10

ARGS = {
    'confs': confs,
    'preds': preds,
    'labels': labels,
    'num_bins': num_bins
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
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 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
param_groups_weight_decay Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/semilearn/nets/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from yewen1486/emtal, is pointer only (licence: NONE) · get_code("88f4962784e09b88")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from yewen1486/emtal, 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.

read_TextClassificationDatasets Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/preprocess/preprocess_aclimdb.py
code served (permissive licence) · get_code("d954c448e1eee8c4")
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, ndim 2, 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 numpy as np
import os
import tempfile
import shutil

# Create a temporary directory structure with mock CSV files
temp_dir = tempfile.mkdtemp()
dataset_dir = os.path.join(temp_dir, 'TextClassificationDatasets', 'ag_news_csv')
os.makedirs(dataset_dir, exist_ok=True)

# Create a mock train.csv file for ag_news_csv format
# Format: "label","title","description"
train_csv_path = os.path.join(dataset_dir, 'train.csv')
with open(train_csv_path, 'w') as f:
    f.write('"1","World News","This is a world news article"\n')
    f.write('"2","Sports","This is a sports article"\n')
    f.write('"3","Business","This is a business article"\n')

ARGS = {
    'src_path': os.path.join(temp_dir, 'TextClassificationDatasets/'),
    'data_set': 'ag_news_csv',
    'data_type': 'train'
}
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, ndim 2, 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
read_aclImdb Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/preprocess/preprocess_aclimdb.py
code served (permissive licence) · get_code("eed03e5ff5639e09")
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, ndim 2, 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 numpy as np
import os
import random
import tempfile
import shutil

# Create a temporary directory structure that mimics aclImdb
temp_dir = tempfile.mkdtemp()
aclimdb_path = os.path.join(temp_dir, 'aclImdb')
os.makedirs(os.path.join(aclimdb_path, 'train', 'pos'), exist_ok=True)
os.makedirs(os.path.join(aclimdb_path, 'train', 'neg'), exist_ok=True)
os.makedirs(os.path.join(aclimdb_path, 'train', 'unsup'), exist_ok=True)
os.makedirs(os.path.join(aclimdb_path, 'test', 'pos'), exist_ok=True)
os.makedirs(os.path.join(aclimdb_path, 'test', 'neg'), exist_ok=True)

# Create sample files for train/pos
for i in range(2):
    with open(os.path.join(aclimdb_path, 'train', 'pos', f'file_{i}.txt'), 'w') as f:
        f.write(f'This is a positive review<br />\nwith some text {i}')

# Create sample files for train/neg
for i in range(2):
    with open(os.path.join(aclimdb_path, 'train', 'neg', f'file_{i}.txt'), 'w') as f:
        f.write(f'This is a negative review<br />\nwith some text {i}')

# Create sample files for train/unsup
for i in range(2):
    with open(os.path.join(aclimdb_path, 'train', 'unsup', f'file_{i}.txt'), 'w') as f:
        f.write(f'This is an unlabeled review<br />\nwith some text {i}')

# Create sample files for test/pos
for i in range(2):
    with open(os.path.join(aclimdb_path, 'test', 'pos', f'file_{i}.txt'), 'w') as f:
        f.write(f'This is a test positive review<br />\nwith some text {i}')

# Create sample files for test/neg
for i in range(2):
    with open(os.path.join(aclimdb_path, 'test', 'neg', f'file_{i}.txt'), 'w') as f:
        f.write(f'This is a test negative review<br />\nwith some text {i}')

ARGS = {
    'src_path': aclimdb_path + '/',
    'data_type': 'train'
}
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, ndim 2, 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
resize_pos_embed_vit Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
shambhavicodes/miscalibration-ssl/semilearn/nets/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from aah94/JEPAMatch, is pointer only (licence: NOASSERTION) · get_code("4c0668013863ad3f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from aah94/JEPAMatch, and GitHub could not classify the licence of that repository (NOASSERTION). 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.

rotate_img Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
shambhavicodes/miscalibration-ssl/semilearn/algorithms/crmatch/crmatch.py
code served (permissive licence) · get_code("4331f5778c460265")
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, 4, 3, 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

torch.manual_seed(0)

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

# Test with rotation of 90 degrees
rot = 90

ARGS = {
    "img": img,
    "rot": rot
}
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, 4, 3, 5) — shape (2, 4, 3, 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
select_data Ran
ran; our contract draft was wrong, not the code
shambhavicodes/miscalibration-ssl/preprocess/preprocess_aclimdb.py
code served (permissive licence) · get_code("05dda2f18675eb59")
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 torch
import numpy as np

torch.manual_seed(0)

# Construct arguments for select_data function
dataset = 'aclImdb'

# Create sample sentences (ori_sen)
ori_sen = [
    "This is a great movie!",
    "I really enjoyed this film.",
    "Terrible waste of time.",
    "Not worth watching.",
    "Amazing performance by the actors.",
    "Boring and predictable.",
    "One of the best films ever made.",
    "Absolutely horrible.",
    "Loved every minute of it.",
    "Could not finish watching it."
]

# Create sample labels - using 1-indexed labels to test the conversion logic
label = ['1', '1', '2', '2', '1', '2', '1', '2', '1', '2']

# data_type can be 'test', 'train', or 'dev'
data_type = 'train'

# Default train and dev numbers per class
train_num_per_class = 5000
dev_num_per_class = 2000

ARGS = {
    'dataset': dataset,
    'ori_sen': ori_sen,
    'label': label,
    'data_type': data_type,
    'train_num_per_class': train_num_per_class,
    'dev_num_per_class': dev_num_per_class
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 — 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
load_checkpoint Not yet run
unverified (harvested, no recorded run)
shambhavicodes/miscalibration-ssl/semilearn/nets/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from aah94/JEPAMatch, is pointer only (licence: NOASSERTION) · get_code("b158ea331635a6c9")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from aah94/JEPAMatch, and GitHub could not classify the licence of that repository (NOASSERTION). 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. 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, 10 are also present in other papers whose code we hold (10 with at least one paper dated earlier, 0 the same month, 8 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
param_groups_weight_decay 15 other papers
9 dated earlier, 6 later
2210.08823 (2022-10)
Scaling &amp; Shifting Your Features: A New Baseline for Efficient Model Tuning
2304.14065 (2023-04)
Lightweight, Pre-trained Transformers for Remote Sensing Timeseries
2305.13002 (2023-05)
Rethinking Semi-supervised Learning with Language Models
and 12 more not named here
resize_pos_embed_vit 6 other papers
4 dated earlier, 2 later
2305.13002 (2023-05)
Rethinking Semi-supervised Learning with Language Models
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
2310.18910 (2023-10)
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
and 3 more not named here
load_checkpoint 5 other papers
4 dated earlier, 1 later
2305.13002 (2023-05)
Rethinking Semi-supervised Learning with Language Models
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
2310.18910 (2023-10)
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
and 2 more not named here
read_TextClassificationDatasets 5 other papers
3 dated earlier, 2 later
2208.07204 (2022-08)
USB: A Unified Semi-supervised Learning Benchmark for Classification
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
2310.18910 (2023-10)
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
and 2 more not named here
read_aclImdb 5 other papers
3 dated earlier, 2 later
2208.07204 (2022-08)
USB: A Unified Semi-supervised Learning Benchmark for Classification
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
2310.18910 (2023-10)
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
and 2 more not named here
select_data 5 other papers
3 dated earlier, 2 later
2208.07204 (2022-08)
USB: A Unified Semi-supervised Learning Benchmark for Classification
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
2310.18910 (2023-10)
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
and 2 more not named here
array_to_bytes 4 other papers
2 dated earlier, 2 later
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
2310.18910 (2023-10)
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
2409.19245 (2024-09)
Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning
and 1 more not named here
bytes_to_array 4 other papers
2 dated earlier, 2 later
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
2310.18910 (2023-10)
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
2409.19245 (2024-09)
Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning
and 1 more not named here
comatch_contrastive_loss 3 other papers
3 dated earlier
2208.07204 (2022-08)
USB: A Unified Semi-supervised Learning Benchmark for Classification
2305.13002 (2023-05)
Rethinking Semi-supervised Learning with Language Models
2306.04621 (2023-06)
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
rotate_img 1 other paper
1 dated earlier
2305.13002 (2023-05)
Rethinking Semi-supervised Learning with Language Models

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

State-of-the-art semi-supervised learning (SSL) approaches rely on highly confident predictions to serve as pseudo-labels that guide the training on unlabeled samples. An inherent drawback of this strategy stems from the quality of the uncertainty estimates, as pseudo-labels are filtered only based on their degree of uncertainty, regardless of the correctness of their predictions. Thus, assessing and enhancing the uncertainty of network predictions is of paramount importance in the pseudo-labeling process. In this work, we empirically demonstrate that SSL methods based on pseudo-labels are significantly miscalibrated, and formally demonstrate the minimization of the min-entropy, a lower bound of the Shannon entropy, as a potential cause for miscalibration. To alleviate this issue, we integrate a simple penalty term, which enforces the logit distances of the predictions on unlabeled samples to remain low, preventing the network predictions to become overconfident. Comprehensive experiments on a variety of SSL image classification benchmarks demonstrate that the proposed solution systematically improves the calibration performance of relevant SSL models, while also enhancing their discriminative power, being an appealing addition to tackle SSL tasks.

For agents

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

get_harvested_code_for_paper("2403.15567")
get_code_for_paper("2403.15567")
have("2403.15567")

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

curl https://syntology.ai/api/ran/2403.15567.json

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

[![Syntology run record](https://syntology.ai/api/ran/2403.15567.svg)](https://syntology.ai/paper/2403.15567)

Connect an agent — have() is free.