Dr. John Hawkins is the Chief Scientist with GumGum, director of Artificial Intelligence research with the non-profit Pingla Institute and a machine learning researcher affiliated with the Transitional AI Group out of UNSW.

He works across multiple industries applying machine learning to business problems and publishing academic research in the application and analysis of machine learning systems. His current research interests are in advanced text prcocessing, model interpretation, algorithmic reasoning and the quantification of uncertainty. He continues to develop multiple open source libraries for these projects in a commitment to open and reproducible data science.

Projects

Uncertainty Quantification and Media Impact in Eye Tracking Studies

Multiple streams of research aimed at understanding and improving processes for measuring eye tracking based consumer attention on digital media.

• Estimating Attention Thresholds for Brand Impact - (Submitted)
• Estimating Gaze Duration Error from Eye Tracking Data
• Evaluating Ad Creative and Web Context Alignment with Attention Measurement
• Brands, Verticals and Contexts: Coherence Patterns in Consumer Attention

Natural Language Processing, Reasoning and Explainability

Investigations into applications and interpretations of machine learning methods for pure text based systems: including classification, generation and reasoning from text. Projects including classification of internet content and filtering of research papers extracted from biomedical database searches.

• Literature Filtering for Systematic Reviews with Transformers
• Enhanced Review Detection and Recognition: A Platform-Agnostic Approach with Application to Online Commerce

Modern advances in interpretable machine learning are generally less applicable to text data sources. We are exploring methods to render the contributions of unstructured text data more transparent regardless of how it is being processed or the machine learning model that is used.

• What’s in a Domain? Analysis of URL Features
• Texturizer Library available on PyPI or Github
• Textplainer Library in development. Github

Minimum Viable Model Estimation

To what extent can we quantify in advance how good a machine learning system needs to be in order to solve a problem? In this work we are building tools to estimate the required performance levels of a model in order to meet business requirements specified in terms of the minimal ROI and the expected returns for predictions of differing levels of accuracy.

• MinViME/Minimum Viable Model Estimator
• Minimum Viable Model Estimates for Machine Learning Projects
• MinViME Library available on PyPI or Github

Data Science Project Management & Integration

Data science projects require rapid experimentation and exploration of problem spaces. This is typically enabled by using loosely coupled tools and processes. In this work we are developing simple CLI tooling for connecting data and experimental resources so that results can easily be compared and managed without undue coupling and dependency creation.

• Projit library and command line tool: PyPI or Github

Contact

John Hawkins