Peter Anderson Seminar
Date: Monday, 4th June 2018, 11.00am - 12noon
Venue: The Australian Hearing Hub, Level 3, Room 3.610, Macquarie University
Speaker: Peter Anderson, Department of Computing, Macquarie University
Host: Professor Mark Johnson
Topic: Language and Vision using Deep Neural Nets
Abstract
In recent years the machine learning community has made tremendous strides towards systems that combine linguistic and visual understanding. This talk will focus on models that seek to describe the content of images and answer questions about images - the tasks known as image captioning and visual question answering (VQA), respectively. As well as giving an overview of the current state of the art, we will discuss the difficulty of evaluating the outputs of these models, and recent advances that equip deep neural networks with attention mechanisms. Finally, I will introduce recent work that 'embodies' these models within simulation environments, to tackle problems such as navigating from natural language instructions.
Bio
Peter Anderson recently submitted his PhD in Computer Science at the Australian National University, supervised by Prof. Stephen Gould. He is currently a Research Fellow at Macquarie University, working with Prof. Mark Johnson, and a researcher within the Australian Centre for Robotic Vision. His PhD focused on deep learning combining visual understanding with natural language. His team won first place in the visual question answering (VQA) challenge at CVPR 2017. During his PhD he visited numerous universities and research labs including Adelaide University, Macquarie University, Queensland University of Technology and Microsoft Research in Seattle.
Venue: The Australian Hearing Hub, Level 3, Room 3.610, Macquarie University
Speaker: Peter Anderson, Department of Computing, Macquarie University
Host: Professor Mark Johnson
Topic: Language and Vision using Deep Neural Nets
Abstract
In recent years the machine learning community has made tremendous strides towards systems that combine linguistic and visual understanding. This talk will focus on models that seek to describe the content of images and answer questions about images - the tasks known as image captioning and visual question answering (VQA), respectively. As well as giving an overview of the current state of the art, we will discuss the difficulty of evaluating the outputs of these models, and recent advances that equip deep neural networks with attention mechanisms. Finally, I will introduce recent work that 'embodies' these models within simulation environments, to tackle problems such as navigating from natural language instructions.
Bio
Peter Anderson recently submitted his PhD in Computer Science at the Australian National University, supervised by Prof. Stephen Gould. He is currently a Research Fellow at Macquarie University, working with Prof. Mark Johnson, and a researcher within the Australian Centre for Robotic Vision. His PhD focused on deep learning combining visual understanding with natural language. His team won first place in the visual question answering (VQA) challenge at CVPR 2017. During his PhD he visited numerous universities and research labs including Adelaide University, Macquarie University, Queensland University of Technology and Microsoft Research in Seattle.