Seminar: Integrating Explainability into Machine Learning Models

Date/Time: 10-11am, Friday 3/3/2023
Topic: Integrating Explainability into Machine Learning Models
Speaker: Dr. Zahid Islam, CSIRO
Venue: Online via MS Teams. A link will be emailed prior to the seminar.

Abstract: Machine learning is everywhere. One of the fundamental goals of developing machine learning models is to reduce loss (measured by some function) and increase accuracy (measure by comparing predictions and ground truth). Often high performing machine learning models are complex Deep Neural Network which performs numerical computation based on inputs and many parameters. Loss and accuracy metrices drive the computations in the network. However, the metrices are not indicative of the actual reasoning performed internally by the models. High performing model can erroneously give good predictions by incorrect reasoning. In this talk I will discuss how explainability can help build reliable machine learning models in practice.

Bio: I am a post-doctoral fellow in the Conservation Decisions Team of CSIRO's Environment business unit. I am in the Machine Learning and Artificial Intelligence (MLAI) FSP. My interest is in the pragmatic applications of MLAI. My research is focused on Explainable Artificial Intelligence (XAI). I have been trying to explain complex time series models developed using deep neural networks and graph neural networks to humans. In doing so, I have been exploring any latent relationships between features and targets.

I obtained a PhD in Information Sciences from the University of South Australia. Within my PhD, I developed ensemble methods for text classification incorporating XAI. I have also completed an MSc in Computer Science from St Francis Xavier University, Canada. As part of my MSc, I developed a framework for healthcare workflow verification using Computation Tree Logic (CTL). I have obtained a BSc in Computer Science and Engineering from Khulna University, Bangladesh. In between studies, I have worked as part of the teaching faculty at Khulna University, Bangladesh and as a software engineer at different organizations.