TinyGen: Portable and Compact Code Generation for Tiny Machine Learning

Gaeun Ko; Seonyeong Heo

doi:10.1145/3771775.3786278

TinyGen: Portable and Compact Code Generation for Tiny Machine Learning

Gaeun Ko
, Seonyeong Heo

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Tiny machine learning (TinyML) enables low-power microcontrollers to leverage the power of artificial intelligence without relying on remote computing resources. Typically, developing a TinyML application relies primarily on existing TinyML frameworks, which provide runtime APIs for loading and executing machine learning models. However, such framework-based TinyML development has limitations in terms of portability, programmability, and resource efficiency, motivating the need for a new approach to the TinyML development and deployment process. To address these challenges, this work introduces TinyGen, a new code generation framework for TinyML. TinyGen generates portable high-level code directly from a target model without depending on external runtime APIs. It statically analyzes the tensor and operator usage of the target model to enable compact code generation. This work demonstrates that TinyGen reduces binary code size by 31.2% on average compared to an existing TinyML framework while requiring fewer lines of code to write TinyML applications.

Original language	English
Title of host publication	CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction
Editors	Martin Kong, Uday Bondhugula, Tobias Grosser
Publisher	Association for Computing Machinery, Inc
Pages	131-142
Number of pages	12
ISBN (Electronic)	9798400722745
DOIs	https://doi.org/10.1145/3771775.3786278
Publication status	Published - 28 Jan 2026
Event	35th ACM SIGPLAN International Conference on Compiler Construction, CC 2026 - Sydney, Australia Duration: 31 Jan 2026 → 1 Feb 2026

Publication series

Name	CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction

Conference

Conference	35th ACM SIGPLAN International Conference on Compiler Construction, CC 2026
Country/Territory	Australia
City	Sydney
Period	31/01/26 → 1/02/26

Bibliographical note

Publisher Copyright:
© 2026 Owner/Author.

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 8 Decent Work and Economic Growth
SDG 12 Responsible Consumption and Production

Keywords

automatic code generation
multi-level intermediate representation
systems for machine learning
Tiny machine learning

Access to Document

10.1145/3771775.3786278

Cite this

Ko, G., & Heo, S. (2026). TinyGen: Portable and Compact Code Generation for Tiny Machine Learning. In M. Kong, U. Bondhugula, & T. Grosser (Eds.), CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction (pp. 131-142). (CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction). Association for Computing Machinery, Inc. https://doi.org/10.1145/3771775.3786278

Ko, Gaeun ; Heo, Seonyeong. / TinyGen : Portable and Compact Code Generation for Tiny Machine Learning. CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction. editor / Martin Kong ; Uday Bondhugula ; Tobias Grosser. Association for Computing Machinery, Inc, 2026. pp. 131-142 (CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction).

@inproceedings{f90d951c162843b8b8dd0a54ec735ecd,

title = "TinyGen: Portable and Compact Code Generation for Tiny Machine Learning",

abstract = "Tiny machine learning (TinyML) enables low-power microcontrollers to leverage the power of artificial intelligence without relying on remote computing resources. Typically, developing a TinyML application relies primarily on existing TinyML frameworks, which provide runtime APIs for loading and executing machine learning models. However, such framework-based TinyML development has limitations in terms of portability, programmability, and resource efficiency, motivating the need for a new approach to the TinyML development and deployment process. To address these challenges, this work introduces TinyGen, a new code generation framework for TinyML. TinyGen generates portable high-level code directly from a target model without depending on external runtime APIs. It statically analyzes the tensor and operator usage of the target model to enable compact code generation. This work demonstrates that TinyGen reduces binary code size by 31.2\% on average compared to an existing TinyML framework while requiring fewer lines of code to write TinyML applications.",

keywords = "automatic code generation, multi-level intermediate representation, systems for machine learning, Tiny machine learning",

author = "Gaeun Ko and Seonyeong Heo",

note = "Publisher Copyright: {\textcopyright} 2026 Owner/Author.; 35th ACM SIGPLAN International Conference on Compiler Construction, CC 2026 ; Conference date: 31-01-2026 Through 01-02-2026",

year = "2026",

month = jan,

day = "28",

doi = "10.1145/3771775.3786278",

language = "English",

series = "CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction",

publisher = "Association for Computing Machinery, Inc",

pages = "131--142",

editor = "Martin Kong and Uday Bondhugula and Tobias Grosser",

booktitle = "CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction",

}

Ko, G & Heo, S 2026, TinyGen: Portable and Compact Code Generation for Tiny Machine Learning. in M Kong, U Bondhugula & T Grosser (eds), CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction. CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction, Association for Computing Machinery, Inc, pp. 131-142, 35th ACM SIGPLAN International Conference on Compiler Construction, CC 2026, Sydney, Australia, 31/01/26. https://doi.org/10.1145/3771775.3786278

TinyGen: Portable and Compact Code Generation for Tiny Machine Learning. / Ko, Gaeun; Heo, Seonyeong.
CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction. ed. / Martin Kong; Uday Bondhugula; Tobias Grosser. Association for Computing Machinery, Inc, 2026. p. 131-142 (CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - TinyGen

T2 - 35th ACM SIGPLAN International Conference on Compiler Construction, CC 2026

AU - Ko, Gaeun

AU - Heo, Seonyeong

PY - 2026/1/28

Y1 - 2026/1/28

N2 - Tiny machine learning (TinyML) enables low-power microcontrollers to leverage the power of artificial intelligence without relying on remote computing resources. Typically, developing a TinyML application relies primarily on existing TinyML frameworks, which provide runtime APIs for loading and executing machine learning models. However, such framework-based TinyML development has limitations in terms of portability, programmability, and resource efficiency, motivating the need for a new approach to the TinyML development and deployment process. To address these challenges, this work introduces TinyGen, a new code generation framework for TinyML. TinyGen generates portable high-level code directly from a target model without depending on external runtime APIs. It statically analyzes the tensor and operator usage of the target model to enable compact code generation. This work demonstrates that TinyGen reduces binary code size by 31.2% on average compared to an existing TinyML framework while requiring fewer lines of code to write TinyML applications.

AB - Tiny machine learning (TinyML) enables low-power microcontrollers to leverage the power of artificial intelligence without relying on remote computing resources. Typically, developing a TinyML application relies primarily on existing TinyML frameworks, which provide runtime APIs for loading and executing machine learning models. However, such framework-based TinyML development has limitations in terms of portability, programmability, and resource efficiency, motivating the need for a new approach to the TinyML development and deployment process. To address these challenges, this work introduces TinyGen, a new code generation framework for TinyML. TinyGen generates portable high-level code directly from a target model without depending on external runtime APIs. It statically analyzes the tensor and operator usage of the target model to enable compact code generation. This work demonstrates that TinyGen reduces binary code size by 31.2% on average compared to an existing TinyML framework while requiring fewer lines of code to write TinyML applications.

KW - automatic code generation

KW - multi-level intermediate representation

KW - systems for machine learning

KW - Tiny machine learning

UR - https://www.scopus.com/pages/publications/105029841261

U2 - 10.1145/3771775.3786278

DO - 10.1145/3771775.3786278

M3 - Conference contribution

AN - SCOPUS:105029841261

T3 - CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction

SP - 131

EP - 142

BT - CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction

A2 - Kong, Martin

A2 - Bondhugula, Uday

A2 - Grosser, Tobias

PB - Association for Computing Machinery, Inc

Y2 - 31 January 2026 through 1 February 2026

ER -

Ko G, Heo S. TinyGen: Portable and Compact Code Generation for Tiny Machine Learning. In Kong M, Bondhugula U, Grosser T, editors, CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction. Association for Computing Machinery, Inc. 2026. p. 131-142. (CC 2026 - Proceedings of the 35th ACM SIGPLAN International Conference on Compiler Construction). doi: 10.1145/3771775.3786278

TinyGen: Portable and Compact Code Generation for Tiny Machine Learning

Abstract

Publication series

Conference

Bibliographical note

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this