Deep learning

Deep learning

Deep learning, sometimes referred to as deep structured learning or hierarchical learning, is a subset of artificial intelligence and machine learning whereby computers are trained to perform human-like tasks (such as making predictions, understanding and identifying images or recognising speech). 

Deep learning involves building and training neural networks using large data sets. By performing the set task repeatedly, the machine finds patters and learns from experience.  

Our researchers have made major contributions to advancing the mathematical tools that underpin deep learning theory. We can and do use this world class expertise to help organisations better understand their data.

 

Projects

  • Learning the deep structure of images

    This project seeks to develop technologies that will help computer vision interpret the whole visible scene, rather than just some of the objects therein. Existing automated methods for understanding images perform well at recognising specific objects in canonical poses, but the problem of whole image interpretation is far more challenging. Convolutional neural networks (CNN) have underpinned recent progress in object recognition, but whole-image understanding cannot be tackled similarly because the number of possible combinations of objects is too large. The project thus proposes a graph-based generalisation of the CNN approach which allows scene structure to be learned explicitly. This would represent an important step towards providing computers with robust vision, allowing them to interact with their environment.

    Professor Anton van den Hengel; Dr Anthony Dick; Dr Lingqiao Liu

  • Semantic change detection through large-scale learning

    Identifying whether there has been a significant change in a scene from a set of images is an important practical task, and has received much attention. The problem has been, however, that although existing statistical techniques perform reasonably well, it has been impossible to achieve the high levels of accuracy demanded by most real applications. This is due to the fact that changes in pixel intensity are not a particularly good indicator of significant change in a scene. We propose a semantic change detection approach which aims to classify the content of an image before attempting to identify change. This technology builds upon recent developments in large-scale classification which have dramatically improved both accuracy and speed.

    Professor Anton van den Hengel; Professor Chunhua Shen; Dr Anders Eriksson; Dr Qinfeng Shi; BAE Systems

  • Scalable classification for massive datasets: randomised algorithms

    Classification is a fundamental data analysis technology and is applied every day in fields from astronomy to zoology. It is used to identify causes of disease, forms of tax evasion, and sources of oil, but is even more critical to developing data-bound sciences such as genomics, semantic document analysis and precision agriculture. This project will develop classification technologies capable of distinguishing between tens of thousands of classes, which are trained and applied to massive datasets. These technologies will deliver a significant increase in the scale of problem which may be tackled, and to the scale of benefits which may be achieved.

    Professor Anton van den Hengel; Professor Chunhua Shen; Dr Qinfeng Shi; LBT Innovations

  • Combined shape and appearance descriptors for visual object recognition

    The quantity of video generated each year is expanding rapidly. This increasing volume of visual information means that it is more likely that any particular event will be recorded, but that the footage will be harder to find. This applies to a collection of home videos as much as to television and movie footage. The object-recognition method to be developed has the potential to alleviate this situation, in which vast amounts of video data are available but have little value. Such an outcome would be a boon for Australian industry and offer a valuable export opportunity.

    Professor Anton van den Hengel; Associate Professor Anthony Dick

  • Image search for simulator content creation

    3D content creation represents one of the most labour intensive stages of the process of constructing virtual environments such as simulators and games. In many cases it is possible to capture images or video of the environment to be simulated which may be used to assist the modelling process. This project aims to develop technologies based on search by which such imagery may provide both shape and semantic information to assist in the modelling process. The project builds upon recent developments in bag-of-words methods for image search. Particularly, we propose a novel method by which information latent in the image database may be identified and used to improve generative model underpinning this type of image search.

    Professor Anton van den Hengel; Associate Professor Anthony Dick; Sydac

  • Computational infrastructure for machine learning in computer vision

    Machine learning is responsible for many recent advances in image-based information analysis, from finding minerals in satellite images, to image-based guidance of autonomous vehicles. This progress is due to new methods for learning from the vast volumes of image-based data that are now available. These images present a great opportunity that is only just beginning to be exploited, as automated image analysis methods still lag far behind the human ability to interpret image information. This project will develop the specific infrastructure required to tackle this problem, allowing Australia researchers to carry out the large-scale image-based machine learning required to achieve automated understanding of the world through images.

    A. van den Hengel; I.D. Reid; S. Venkatesh; B. Vo; D.Suter; S. Gould; S.M. Lucey; A.R. Dick; C.Shen; D.Q. Phung