Unifiying Machine Learning (ML) Frameworks: Source-to-Source Transpilation of ML Code for Complete Interoperability of Frameworks, Versions and Hardware to Streamline Business ML Implementations for Revenue Increases and Operational Costs/Waste Reductions

统一机器学习 (ML) 框架：ML 代码的源到源转换，实现框架、版本和硬件的完全互操作性，以简化业务 ML 实施，从而增加收入并减少运营成本/浪费

• 批准号: 10061942
• 金额: $ 41.02万
• 依托单位: DEEP IVY LTD.
• 依托单位国家: 英国
• 项目类别: Collaborative R&D
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=10061942
• 关键词: Unifiying; Machine; Learning; ML; Frameworks

In an increasingly data-driven world and against a backdrop of increasing business costs, AI and ML are of vital importance in helping businesses gain commercial advantage (increasing sales, higher returns on investment and reducing operating costs).The implementation of ML is predominantly achieved with "frameworks" (software interfaces with tools/libraries for building/deploying/scaling models). Such frameworks are "locked" into their providers and are incompatible with others, preventing easy use of libraries/models written in different frameworks. This radically reduces usability and access to the wealth of ML developments constantly released, or requires time-consuming/costly/error-prone manual transcription of code.Our innovative source-to-source transpiler ("Ivy-SST") overcomes this lack of interoperability by unifying ML frameworks to enable _any ML code_ to be run on _any version_ of _any framework_ on _any hardware_, and _permitting further manipulation/development/elaboration_ in the framework of choice by outputting _human-readable code_.In doing so, we will help businesses achieve ML implementations faster/cheaper/more efficiently, for measurable business benefits: increased revenues (+11.5%), reduced operational costs (-30%) and waste reduction (-20%) for sustainability/efficiency improvements.

在数据日益驱动的世界中，在业务成本不断增加的背景下，人工智能和机器学习对于帮助企业获得商业优势（增加销售额、更高的投资回报和降低运营成本）至关重要。机器学习的实施主要通过“框架”（与用于构建/部署/扩展模型的工具/库的软件接口）来实现。此类框架被“锁定”到其提供者中并且与其他框架不兼容，从而妨碍了在不同框架中编写的库/模型的轻松使用。这从根本上降低了可用性和对不断发布的大量 ML 开发的访问，或者需要耗时/昂贵/容易出错的手动转录代码。我们创新的源到源转译器（“Ivy-SST”）通过统一 ML 框架来克服互操作性的缺乏，使_任何 ML 代码_能够在_任何硬件_上的_任何框架_的_任何版本_上运行， 通过输出_人类可读的代码_，在选择的框架内_允许进一步操作/开发/精加工_。这样做，我们将帮助企业更快/更便宜/更高效地实现机器学习实施，以获得可衡量的商业利益：增加收入（+11.5％），降低运营成本（-30％）和减少浪费（-20％），以提高可持续性/效率。