ReLMI Lab Industry Showcase

Representation Learning on Large Scale Administrative Data

 

Image: Stocksnap (front) and Gerd Altmann (back) / Pixabay. Originals edited.
 

Research Lead
Dr Guodong Long

AAII Lab
ReLMI Lab

Collaborators
Australian Commonwealth Department of Health

The key to effective policy development is implementing the right policy for the right group of people at the right time. To support this ethos, the public service is keen to develop predictive modelling methods to leverage the information hidden within its administrative data about different population cohorts.

In a large-scale service system, such as the Australian healthcare system, user data exists in complex structures reflective of the complexity of human interactions. Appropriate ways to model this complex structured data is a challenge common to many organisations, from sector-scale commercial services to national-scale public services.

A generalized representation of complex data is the most desirable data format for processing, particularly for large organisations and government agencies that typically deal with data on a massive scale. In particular, utilising powerful deep learning tools could be one of the implementation framework for representation learning on large-scale administrative dataset. The framework will preserve the complex relationships in an unsupervised or self-supervised manner, and to protect privacy in a decentralised training scheme without direct access of the sensitive data.

The DSKD Lab has collaborated with Australian Commonwealth Department of Health a series of projects that apply the most advanced machine learning techniques to model and analyse healthcare administrative data to promote effective policy development.

 

Dialogue-to-Action: A new way to reshape Business Intelligence

Research Lead
Dr Guodong Long

AAII Lab
ReLMI Lab

Business intelligence (BI) comprise the strategies and technologies used by enterprises for the data analysis of business information. “Enterprise-based, natural language processing (NLP)-powered self-service solutions, also known as ‘Intelligent Assistants (IAs), have gained traction and proven their value over the past ten years”. Their usefulness is evident in many enterprise application scenarios, e.g., customer care, marketing, internal workflow automation, and business analysis”. Existing Enterprise Intelligent Assistant (EIA) techniques derive from Personal Intelligent Assistants (PIAs) and heavily rely on pre-defined rules for well-specified scenarios. However, rapidly changing business and user scenarios urgently need the self-evolving capability of EIAs.

This project aims to develop a novel Self-evolving EIA system by using sequence-to-sequence modelling based Dialogue-to-Action (D2A) framework. The system will integrate Chatbot-based dialogue technology to acquire information, infer user intentions, understand languages, and determine the next actions to take. Possible actions include collecting information, providing information, conducting a transaction, or initialling a series of business operations to fulfil a user’s request in a complex enterprise management environment.

The game-changing Dialogue-to-Action research will reshape business intelligence by empowering the cutting-edge NLP with deep learning technique. Specifically, it will fundamentally transform current enterprise intelligence assistance from a massive rule-based process to neural-based seq2seq modelling. The developed self-evolving EIA will immediately enhance the broad service sector, and will enable high-quality service to be maintained at lower costs, as well as ensure fast adaptation to business changes by deploying and maintaining EIA systems. The project’s outcome will not only strengthen Australia’s world leadership in business and the public service sector but will also pave the way to Artificial Generalised Intelligence by advancing the theoretical foundations for self-evolving artificial intelligence research.