SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1910.05110 · 2019

Neural networks on microcontrollers: saving memory at inference via operator reordering

arXiv · PDF · Open in the Atlas

Code that ran

12 of 14 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 12 have no contract verdict; 0 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/1910.05110.json · badge: /api/ran/1910.05110.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
oxmlsys/tflite-tools canonical (classifier label)
classifier basis: census
12 ran · 2 no recorded run
FunctionStatusWhere it lives
AbsOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/AbsOptions.py
code served (permissive licence) · get_code("aa810b70bad0e948")
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

# Mock FlatBuffers builder class
class MockFlatBuffersBuilder:
    def __init__(self):
        self.objects = []
        self.current_object = {}
    
    def EndObject(self):
        """Simulates ending a FlatBuffers object and returning a reference"""
        obj_ref = len(self.objects)
        self.objects.append(self.current_object)
        self.current_object = {}
        return obj_ref

# Create a builder instance
builder = MockFlatBuffersBuilder()

ARGS = {
    "builder": builder
}
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
AddNOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/AddNOptions.py
code served (permissive licence) · get_code("0ddef8025e31d92a")
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 torch.nn as nn

# This function appears to be part of a FlatBuffers builder pattern
# It takes a builder object and calls EndObject() on it
# We need to create a minimal builder-like object that has an EndObject method

class MockFlatBuffersBuilder:
    """Minimal mock of a FlatBuffers Builder object"""
    def __init__(self):
        self.finished = False
        self.objects_stack = []
    
    def EndObject(self):
        """Simulates ending a FlatBuffers object"""
        self.finished = True
        return 42  # Return an arbitrary offset/handle

# Create the builder instance
builder = MockFlatBuffersBuilder()

# Define arguments dict
ARGS = {
    "builder": builder
}
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 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
AddOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/AddOptions.py
code served (permissive licence) · get_code("1ffb75597fc7bd2b")
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

# Mock FlatBuffers-like builder object
class MockFlatBuffersBuilder:
    def __init__(self):
        self.objects = []
        self.finished = False
    
    def EndObject(self):
        """Simulates ending a FlatBuffers object"""
        self.finished = True
        return len(self.objects)

# Create a builder instance
builder = MockFlatBuffersBuilder()

# The function takes a builder and calls EndObject() on it
# We need to provide a valid builder that has this method
ARGS = {
    "builder": builder
}
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
ArgMaxOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/ArgMaxOptions.py
code served (permissive licence) · get_code("b81bbd359f30bbf0")
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

# The function ArgMaxOptionsEnd takes a builder parameter and calls builder.EndObject()
# We need to create a minimal builder object that has an EndObject method

class MockFlatBufferBuilder:
    """Minimal FlatBuffers builder stand-in"""
    def __init__(self):
        self.finished = False
    
    def EndObject(self):
        """Return the end object result"""
        self.finished = True
        return 42  # Return a dummy value

# Create the builder instance
builder = MockFlatBufferBuilder()

# Define ARGS mapping all parameters
ARGS = {
    "builder": builder
}
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
ArgMinOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/ArgMinOptions.py
code served (permissive licence) · get_code("6de21b13beace7bc")
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

# Mock builder class that mimics FlatBuffers builder behavior
class MockBuilder:
    def __init__(self):
        self.objects = []
        self.current_object = None
    
    def EndObject(self):
        """End the current object being built and return a reference to it"""
        if self.current_object is not None:
            obj_ref = len(self.objects)
            self.objects.append(self.current_object)
            self.current_object = None
            return obj_ref
        return None

# Create builder instance
builder = MockBuilder()

# Initialize a mock object being built
builder.current_object = {"type": "ArgMinOptions"}

ARGS = {
    "builder": builder
}
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
BatchToSpaceNDOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/BatchToSpaceNDOptions.py
code served (permissive licence) · get_code("2466f15207cc39b0")
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 torch.nn as nn

# Mock builder class that mimics FlatBuffers builder interface
class MockFlatBuffersBuilder:
    def __init__(self):
        self.objects = []
    
    def EndObject(self):
        """End current object and return an object reference"""
        return len(self.objects)

# Construct the builder argument
builder = MockFlatBuffersBuilder()

# Define ARGS mapping
ARGS = {
    "builder": builder
}
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 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
BidirectionalSequenceLSTMOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/BidirectionalSequenceLSTMOptions.py
code served (permissive licence) · get_code("0b1c80069a231389")
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 torch.nn as nn

# Looking at the function signature:
# def BidirectionalSequenceLSTMOptionsEnd(builder):
#     return builder.EndObject()
#
# This function takes a builder object and calls EndObject() on it.
# The builder appears to be a FlatBuffers builder object.
# We need to provide a mock builder that has an EndObject() method.

class MockFlatBuffersBuilder:
    """Mock FlatBuffers builder for testing"""
    def __init__(self):
        self.objects = []
    
    def EndObject(self):
        """Simulate ending an object in FlatBuffers"""
        return len(self.objects)

# Create a builder instance
builder = MockFlatBuffersBuilder()

ARGS = {
    "builder": builder
}
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 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
BidirectionalSequenceRNNOptionsEnd Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/BidirectionalSequenceRNNOptions.py
code served (permissive licence) · get_code("6c7b30b439c8aba3")
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 torch
import numpy as np

# Mock builder class that simulates a FlatBuffers builder
class MockBuilder:
    def __init__(self):
        self.objects = []
    
    def EndObject(self):
        """Simulates ending a FlatBuffers object construction"""
        obj = {"type": "BidirectionalSequenceRNNOptions"}
        self.objects.append(obj)
        return obj

# Create a builder instance
builder = MockBuilder()

# Define ARGS dict with the builder parameter
ARGS = {
    "builder": builder
}
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 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
ModelStartBuffersVector Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/Model.py
code served (permissive licence) · get_code("cbcbf947a3c73834")
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

# Mock FlatBuffers Builder class
class MockFlatBuffersBuilder:
    def __init__(self):
        self.buffers = []
    
    def StartVector(self, element_size, num_elements, alignment):
        """
        Mock implementation of StartVector.
        Returns a vector start offset/handle.
        """
        vector_handle = len(self.buffers)
        self.buffers.append({
            'element_size': element_size,
            'num_elements': num_elements,
            'alignment': alignment
        })
        return vector_handle

# Create a mock builder instance
builder = MockFlatBuffersBuilder()

# Set numElems to a small positive integer
numElems = 5

ARGS = {
    'builder': builder,
    'numElems': numElems
}
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
ModelStartOperatorCodesVector Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/Model.py
code served (permissive licence) · get_code("8edfabcc3bbb4d87")
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

# Create a minimal FlatBuffers builder mock
class MockFlatBuffersBuilder:
    def __init__(self):
        self.call_history = []
    
    def StartVector(self, element_size, num_elements, alignment):
        """
        Mock implementation of FlatBuffers StartVector.
        element_size: size of each element in bytes
        num_elements: number of elements
        alignment: alignment requirement in bytes
        """
        self.call_history.append({
            'method': 'StartVector',
            'element_size': element_size,
            'num_elements': num_elements,
            'alignment': alignment
        })
        return len(self.call_history) - 1

# Create builder instance
builder = MockFlatBuffersBuilder()

# The function calls builder.StartVector(4, numElems, 4)
# So we need:
# - builder: an object with a StartVector method
# - numElems: an integer representing the number of elements
numElems = 5

ARGS = {
    'builder': builder,
    'numElems': numElems
}
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
ModelStartSubgraphsVector Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite/Model.py
code served (permissive licence) · get_code("695cc78898a40eac")
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

# Mock builder class that implements FlatBuffers-like interface
class MockBuilder:
    def StartVector(self, elementSize, numElements, alignment):
        """
        Mock implementation of FlatBuffers StartVector.
        elementSize: size of each element in bytes
        numElements: number of elements in the vector
        alignment: alignment requirement in bytes
        Returns an offset/handle for the vector
        """
        return numElements * elementSize

# Construct arguments
builder = MockBuilder()
numElems = 5

ARGS = {
    "builder": builder,
    "numElems": numElems
}
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
cluster_weights Ran
ran on a synthesized input
oxmlsys/tflite-tools/tflite_tools/tflite_model.py
code served (permissive licence) · get_code("4a98be502c8d5284")
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 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 numpy as np

# Create a small numpy array of weights to cluster
weights = np.array([1.2, 2.1, 1.5, 8.9, 9.1, 8.7, 15.2, 14.9], dtype=np.float32)

# Number of clusters
n_clusters = 3

ARGS = {
    "weights": weights,
    "n_clusters": n_clusters,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — 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
get_buffer_as_numpy Not yet run
unverified (harvested, no recorded run)
oxmlsys/tflite-tools/tflite_tools/tflite_model.py
code served (permissive licence) · get_code("a04c650fa9cc9971")
How it ran

no invocation recorded.

get_buffer_element_size Not yet run
unverified (harvested, no recorded run)
oxmlsys/tflite-tools/tflite_tools/tflite_model.py
code served (permissive licence) · get_code("ee1162f08109ee09")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 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 14 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

Designing deep learning models for highly-constrained hardware would allow imbuing many edge devices with intelligence. Microcontrollers (MCUs) are an attractive platform for building smart devices due to their low cost, wide availability, and modest power usage. However, they lack the computational resources to run neural networks as straightforwardly as mobile or server platforms, which necessitates changes to the network architecture and the inference software. In this work, we discuss the deployment and memory concerns of neural networks on MCUs and present a way of saving memory by changing the execution order of the network's operators, which is orthogonal to other compression methods. We publish a tool for reordering operators of TensorFlow Lite models and demonstrate its utility by sufficiently reducing the memory footprint of a CNN to deploy it on an MCU with 512KB SRAM.

For agents

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

get_harvested_code_for_paper("1910.05110")
get_code_for_paper("1910.05110")
have("1910.05110")

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

curl https://syntology.ai/api/ran/1910.05110.json

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

[![Syntology run record](https://syntology.ai/api/ran/1910.05110.svg)](https://syntology.ai/paper/1910.05110)

Connect an agent — have() is free.