SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1906.06307 · 2019

A Signal Propagation Perspective for Pruning Neural Networks at Initialization

arXiv · PDF · Open in the Atlas

Code that ran

10 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 10 have no contract verdict; 0 failures of our instrument; 3 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/1906.06307.json · badge: /api/ran/1906.06307.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
namhoonlee/spp-public canonical (classifier label)
classifier basis: census
10 ran · 3 no recorded run
FunctionStatusWhere it lives
cache_json Ran
ran on a synthesized input
namhoonlee/spp-public/spp/helpers.py
code served (permissive licence) · get_code("a5829b43dc01b183")
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. 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 json
import os
import tempfile

# Create a temporary directory for testing
temp_dir = tempfile.mkdtemp()

# Define a simple zero-argument function that returns JSON-serializable data
def sample_func():
    return {"key": "value", "number": 42, "list": [1, 2, 3]}

# Construct the arguments
filename = os.path.join(temp_dir, "test_cache.json")
func = sample_func
makedir = True
allow_load = True

ARGS = {
    "filename": filename,
    "func": func,
    "makedir": makedir,
    "allow_load": allow_load,
}
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 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
dense_to_one_hot Ran
ran on a synthesized input
namhoonlee/spp-public/spp/mnist.py
code served (permissive licence) · get_code("b757a2b89740842b")
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 (5, 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

# Create valid arguments for dense_to_one_hot function
labels_dense = np.array([0, 1, 2, 1, 0], dtype=np.int64)
num_classes = 3

ARGS = {
    "labels_dense": labels_dense,
    "num_classes": num_classes,
}
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 (5, 3) — shape (5, 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
extract_images Ran
ran on a synthesized input
namhoonlee/spp-public/spp/mnist.py
code served (permissive licence) · get_code("dc6a20f23af74deb")
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, 4, 3, 1) 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 gzip
import numpy as np
import io
import struct

# Create a minimal valid MNIST-format gzipped file
def create_mock_mnist_file():
    """Create a minimal valid MNIST image file in gzip format."""
    # MNIST format:
    # [offset] [type]          [value]
    # 0000     32 bit integer  0x00000803 (2051) - magic number
    # 0004     32 bit integer  num_images
    # 0008     32 bit integer  num_rows
    # 0012     32 bit integer  num_cols
    # Rest     unsigned bytes  pixel data
    
    num_images = 2
    rows = 4
    cols = 3
    
    buffer = io.BytesIO()
    
    # Write magic number (2051 in big-endian)
    buffer.write(struct.pack('>I', 2051))
    # Write num_images
    buffer.write(struct.pack('>I', num_images))
    # Write rows
    buffer.write(struct.pack('>I', rows))
    # Write cols
    buffer.write(struct.pack('>I', cols))
    
    # Write pixel data (num_images * rows * cols bytes)
    pixel_data = np.random.randint(0, 256, size=(num_images, rows, cols), dtype=np.uint8)
    buffer.write(pixel_data.tobytes())
    
    # Compress with gzip
    buffer.seek(0)
    compressed = io.BytesIO()
    with gzip.GzipFile(fileobj=compressed, mode='wb') as gz:
        gz.write(buffer.read())
    
    # Write to temporary file path
    import tempfile
    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.gz')
    compressed.seek(0)
    temp_file.write(compressed.read())
    temp_file.close()
    
    return temp_file.name

# Create the mock file
mock_filename = create_mock_mnist_file()

ARGS = {
    'filename': mock_filename
}
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 4, dtype family int, shape (2, 4, 3, 1) — shape (2, 4, 3, 1) 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
extract_labels Ran
ran on a synthesized input
namhoonlee/spp-public/spp/mnist.py
code served (permissive licence) · get_code("57f1bd325f3272fb")
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 int, shape (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 gzip
import numpy as np
import io
import struct

# Create a minimal valid MNIST label file in gzip format
def create_valid_mnist_label_file():
    """Create a minimal valid gzip-compressed MNIST label file."""
    buffer = io.BytesIO()
    
    with gzip.GzipFile(fileobj=buffer, mode='wb') as gz:
        # Write magic number (2049 in big-endian)
        gz.write(struct.pack('>I', 2049))
        # Write number of items (5 labels)
        gz.write(struct.pack('>I', 5))
        # Write 5 label bytes (values 0-4)
        gz.write(bytes([0, 1, 2, 3, 4]))
    
    buffer.seek(0)
    return buffer

# Create the gzip file as bytes
gzip_buffer = create_valid_mnist_label_file()
gzip_bytes = gzip_buffer.getvalue()

# Write to a temporary file-like object that can be opened
import tempfile
import os

temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.gz')
temp_file.write(gzip_bytes)
temp_file.close()

ARGS = {
    'filename': temp_file.name,
    'one_hot': False,
}
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 1, dtype family int, shape (5,) — shape (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
for_loop Ran
ran on a synthesized input
namhoonlee/spp-public/spp/model.py
code served (permissive licence) · get_code("78dc5962d5d666ac")
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 EagerTensor, ndim 3, dtype family float, shape (3, 1, 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 tensorflow as tf
import numpy as np

# Simple loop function that returns a tensor
def simple_loop_fn(i):
    """Returns a tensor based on iteration number"""
    return tf.constant([[1.0, 2.0, 3.0, 4.0]], dtype=tf.float32) * tf.cast(i, tf.float32)

# Nested loop function that returns multiple tensors
def nested_loop_fn(i):
    """Returns a nested structure (dict) of tensors"""
    val = tf.cast(i, tf.float32)
    return {
        'a': tf.constant([[1.0, 2.0, 3.0, 4.0]], dtype=tf.float32) * val,
        'b': tf.constant([[5.0, 6.0]], dtype=tf.float32) * val
    }

# Define dtypes for simple case
simple_dtypes = tf.float32

# Define dtypes for nested case
nested_dtypes = {
    'a': tf.float32,
    'b': tf.float32
}

ARGS = {
    'loop_fn': simple_loop_fn,
    'loop_fn_dtypes': simple_dtypes,
    'iters': 3,
    'parallel_iterations': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 3, dtype family float, shape (3, 1, 4) — shape (3, 1, 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
get_activation Ran
ran on a synthesized input
namhoonlee/spp-public/spp/network.py
code served (permissive licence) · get_code("66d9542e7fc1cb71")
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 tensorflow as tf

ARGS = {
    'activation': 'relu'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — 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
read_data_sets Ran
ran on a synthesized input
namhoonlee/spp-public/spp/tiny_imagenet.py
code served (permissive licence) · get_code("4793e5a1bc03daaf")
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. 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 tempfile
import shutil
import numpy as np
import cv2

# Create a temporary directory structure mimicking the Tiny ImageNet dataset format
temp_dir = tempfile.mkdtemp()

# Create wnids.txt with class labels
wnids_path = os.path.join(temp_dir, 'wnids.txt')
with open(wnids_path, 'w') as f:
    f.write('n00000001\n')
    f.write('n00000002\n')

# Create train directory structure
train_dir = os.path.join(temp_dir, 'train')
os.makedirs(train_dir)

# Create training images for class 1
train_class1_dir = os.path.join(train_dir, 'n00000001', 'images')
os.makedirs(train_class1_dir)
img1 = np.random.randint(0, 256, (64, 64, 3), dtype=np.uint8)
cv2.imwrite(os.path.join(train_class1_dir, 'img_1.JPEG'), img1)
img2 = np.random.randint(0, 256, (64, 64, 3), dtype=np.uint8)
cv2.imwrite(os.path.join(train_class1_dir, 'img_2.JPEG'), img2)

# Create training images for class 2
train_class2_dir = os.path.join(train_dir, 'n00000002', 'images')
os.makedirs(train_class2_dir)
img3 = np.random.randint(0, 256, (64, 64, 3), dtype=np.uint8)
cv2.imwrite(os.path.join(train_class2_dir, 'img_3.JPEG'), img3)

# Create val directory structure
val_dir = os.path.join(temp_dir, 'val')
val_images_dir = os.path.join(val_dir, 'images')
os.makedirs(val_images_dir)

# Create validation images
img4 = np.random.randint(0, 256, (64, 64, 3), dtype=np.uint8)
cv2.imwrite(os.path.join(val_images_dir, 'val_img_1.JPEG'), img4)
img5 = np.random.randint(0, 256, (64, 64, 3), dtype=np.uint8)
cv2.imwrite(os.path.join(val_images_dir, 'val_img_2.JPEG'), img5)

# Create val_annotations.txt
val_annotations_path = os.path.join(val_dir, 'val_annotations.txt')
with open(val_annotations_path, 'w') as f:
    f.write('val_img_1.JPEG\tn00000001\n')
    f.write('val_img_2.JPEG\tn00000002\n')

ARGS = {
    'path_dataset': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — 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
read_data_sets Ran
ran on a synthesized input
namhoonlee/spp-public/spp/cifar.py
code served (permissive licence) · get_code("183b71f2551de460")
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. 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 pickle
import numpy as np
import tempfile
import shutil

# Create a temporary directory structure with mock CIFAR-10 data
temp_dir = tempfile.mkdtemp()

# Create mock data batches
for batch_num in range(1, 6):
    batch_data = {
        b'data': np.random.randint(0, 256, (2, 3072), dtype=np.uint8),
        b'labels': np.random.randint(0, 10, (2,), dtype=np.int32)
    }
    filename = os.path.join(temp_dir, f'data_batch_{batch_num}')
    with open(filename, 'wb') as f:
        pickle.dump(batch_data, f)

# Create mock test batch
test_batch_data = {
    b'data': np.random.randint(0, 256, (2, 3072), dtype=np.uint8),
    b'labels': np.random.randint(0, 10, (2,), dtype=np.int32)
}
test_filename = os.path.join(temp_dir, 'test_batch')
with open(test_filename, 'wb') as f:
    pickle.dump(test_batch_data, f)

ARGS = {
    'path_dataset': temp_dir
}
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 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
read_data_sets Ran
ran on a synthesized input
namhoonlee/spp-public/spp/fashion_mnist.py
code served (permissive licence) · get_code("44bd5219b1a3f018")
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. 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
import gzip
import tempfile

# Create temporary directory and mock MNIST files
temp_dir = tempfile.mkdtemp()

# Helper function to create mock gzip files
def create_mock_mnist_file(path, num_samples, is_labels=False):
    """Create a mock MNIST gzip file with proper headers."""
    with gzip.open(path, 'wb') as f:
        if is_labels:
            # Label file header: magic number (2 bytes) + version (2 bytes) + num_items (4 bytes)
            header = np.array([0, 0, 0, 1, 0, 0, 0, num_samples], dtype=np.uint8)
            data = np.random.randint(0, 10, num_samples, dtype=np.uint8)
        else:
            # Image file header: magic number (2 bytes) + version (2 bytes) + num_items (4 bytes) + rows (4 bytes) + cols (4 bytes)
            header = np.array([0, 0, 0, 3, 0, 0, 0, num_samples, 0, 0, 0, 28, 0, 0, 0, 28], dtype=np.uint8)
            data = np.random.randint(0, 256, num_samples * 28 * 28, dtype=np.uint8)
        f.write(header.tobytes())
        f.write(data.tobytes())

# Create mock training files
num_train = 100
num_test = 50

create_mock_mnist_file(os.path.join(temp_dir, 'train-labels-idx1-ubyte.gz'), num_train, is_labels=True)
create_mock_mnist_file(os.path.join(temp_dir, 'train-images-idx3-ubyte.gz'), num_train, is_labels=False)
create_mock_mnist_file(os.path.join(temp_dir, 't10k-labels-idx1-ubyte.gz'), num_test, is_labels=True)
create_mock_mnist_file(os.path.join(temp_dir, 't10k-images-idx3-ubyte.gz'), num_test, is_labels=False)

ARGS = {
    'path_dataset': temp_dir,
    'one_hot': False
}
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 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
static_size Ran
ran on a synthesized input
namhoonlee/spp-public/spp/helpers.py
code served (permissive licence) · get_code("75427fd93ad6afdd")
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 int, ndim 0, dtype family int, 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
from unittest.mock import Mock

torch.manual_seed(0)

# The function expects an object x with a .shape attribute
# that has an .as_list() method (TensorFlow-style shape)
# We need to mock this since we're in PyTorch

class MockTFShape:
    def __init__(self, shape_list):
        self.shape_list = shape_list
    
    def as_list(self):
        return self.shape_list

class MockTFTensor:
    def __init__(self, shape_list):
        self.shape = MockTFShape(shape_list)

# Test case 1: valid shape with no None values
x = MockTFTensor([2, 4, 3])

ARGS = {
    "x": x
}
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 int, ndim 0, dtype family int, 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
batch_jacobian Not yet run
unverified (harvested, no recorded run)
namhoonlee/spp-public/spp/model.py
code served (permissive licence) · get_code("c621a9782d26e315")
How it ran

no invocation recorded.

get_initializer Not yet run
unverified (harvested, no recorded run)
namhoonlee/spp-public/spp/network.py
code served (permissive licence) · get_code("b532f60f30000832")
How it ran

no invocation recorded.

reinitialize Not yet run
unverified (harvested, no recorded run)
namhoonlee/spp-public/spp/model.py
code served (permissive licence) · get_code("03d261d081e3c32d")
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, 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 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.

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

Network pruning is a promising avenue for compressing deep neural networks. A typical approach to pruning starts by training a model and then removing redundant parameters while minimizing the impact on what is learned. Alternatively, a recent approach shows that pruning can be done at initialization prior to training, based on a saliency criterion called connection sensitivity. However, it remains unclear exactly why pruning an untrained, randomly initialized neural network is effective. In this work, by noting connection sensitivity as a form of gradient, we formally characterize initialization conditions to ensure reliable connection sensitivity measurements, which in turn yields effective pruning results. Moreover, we analyze the signal propagation properties of the resulting pruned networks and introduce a simple, data-free method to improve their trainability. Our modifications to the existing pruning at initialization method lead to improved results on all tested network models for image classification tasks. Furthermore, we empirically study the effect of supervision for pruning and demonstrate that our signal propagation perspective, combined with unsupervised pruning, can be useful in various scenarios where pruning is applied to non-standard arbitrarily-designed architectures.

For agents

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

get_harvested_code_for_paper("1906.06307")
get_code_for_paper("1906.06307")
have("1906.06307")

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

curl https://syntology.ai/api/ran/1906.06307.json

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

[![Syntology run record](https://syntology.ai/api/ran/1906.06307.svg)](https://syntology.ai/paper/1906.06307)

Connect an agent — have() is free.