This project investigates novel architectures & processes to efficiently label, train and test machine learning systems for business applications. This includes the use of AutoML and new collaborative “code-free” technologies to simplify AI system design/production within a large enterprise.
When it comes to transitioning AI systems from experimental environments into business settings, there is a huge gap between theory and practice. A solution must bring value to the organisation and fit within the organisation’s IT strategy and its established processes concerned with software development, maintenance, data governance, evolution, etc. These solutions must also leverage existing IT and data assets, comply with the relevant regulations, and satisfy ethical considerations.
This project will need a rethink of many traditional software engineering practices for example:
Software architecture: efforts so far have mostly been focused on data engineering of lakes and data warehouses for reporting and statistical modelling. AI/ML requires new modelling approaches that allow quick development and maintainability (e.g., low-code approaches) as well as the ability to efficiently access and process production data. The right architecture should not only satisfy requirements but also leverage AI, cloud and digital transformation technologies.
Development processes: today’s SDLC and agile methods used within the industry need to evolve to fully cater to the experimental data-driven nature of AI/ML projects. This project will consider using emerging SoftwareEngineering4AI (SE4AI) techniques.
Requirements engineering: problems should not be expressed in terms of an ML task but as a set of business objectives with associated measures such as competitiveness, success, and financial benefits. Capturing requirements such as how to achieve transparency (explanations), usability (UI Design, UI aids), trust and performance (information quality) all at the same time represents a difficult challenge.
These issues are all interlinked e.g., adding business objectives may reduce usability and decrease performance, adding more transparency may obscure and decrease trust, and adding more usability may decrease performance. In some cases, ethical and compliance with regulations are other important considerations that need to be taken into account when developing the system.
This project will focus on development practices that provide the ability to “personalise” AI/ML in different contexts using new approaches such as AutoML and collaborative “code-free” technologies. As a case study, this project will investigate how to design infrastructures that allow the use of ML techniques for analysing IoT timeseries data (e.g., indoor/outdoor air temperature data, indoor/outdoor air quality data, relative humidity data) for the purpose of monitoring compliance with building regulations (e.g. WELL standard).
Project objectives:
- Investigate novel requirements and design approaches for building and deploying machine learning systems within the enterprise.
- Exploiting the opportunities offered by AutoML tools.
- Applying new techniques in the area of using machine learning in IoT data analysis for compliance. (e.g., building compliance)
Research team:
Fethi Rabhi (Lead - UNSW CSE), Alan Ng (PhD student), Andy Zeng (BrewAI), Armin Chitizadeh (UNSW CSE), Yuchao Jiang (UNSW CSE), Madhushi Bandara (UTS)
Industry partners:
BrewAI, Capsifi