Visual Question Answering Problems: Reasoning With Deep Learning

Ilija Ilievski is a PhD student at the National University of Singapore, studying interdisciplinary research in the intersection of vision and language. He believes question answering over multimodal data is the next frontier of deep learning, focusing his research on 'Visual Question Answering'. As a side project, he created deeplearningtutorials.com, a place to share his experience developing deep learning methods for real-world problems, with the hope of clearing up the "dark magic" surrounding the development and application of deep learning models for novel problems. At the Deep Learning Summit in Singapore (20-21 October), Ilija will introduce the Visual Question Answering (VQA) problem, its application and significance, as well as presenting a deep learning model able to associate question words to specific image objects. I spoke with him ahead of the summit and asked a few questions to learn more of his thoughts on deep learning.What has driven you to work in the area of deep learning?Deep learning methods have attracted a lot of attention recently by achieving state of the art results in problems like computer vision and speech recognition. But, deep learning is not just another better-performing machine learning method. What's fascinating to me is that deep learning methods are able to outperform other methods but without using human engineered features, and this makes them the best candidate for achieving artificial general intelligence.   Which industries do you think will be disrupted by deep learning in the future, and how? I expect all industries to be disrupted to some degree. I think deep learning, and machine learning in general, will change the way we work the same way computers did in the past. Every business generates data that deep learning methods can use to help business owners make better decisions, increase their efficiency, develop new products and so on.     What do you feel are the most valuable applications of deep learning? One of the most valuable applications of deep learning is in the sciences. I expect as deep learning models have increasingly more reasoning power, they will help scientists by for example pruning unpromising experiments or even proposing possible solutions to existing problems.  This will advance the field much faster, which in turn will bring even more advanced machine learning models.     What advice would you give someone who would like to work in this field?Don't be dissuaded by the steep learning curve. Deep learning may seem as daunting but in fact, the theory behind it is rather simple. Further, there are many excellent resources available online, from books and courses to web portals and open-source libraries.    What developments can we expect to see in deep learning in the next 5 years? I hope to see the development of deep learning methods applied to natural language processing problems that will transform the field. There is also an increasing interest in developing deep learning models for unsupervised learning and reinforcement learning, so we can expect significant advances in these fields as well.Ilija Ilievski will be speaking at the Deep Learning Summit in Singapore on 20-21 October. Other speakers include Brian Cheung, Google Brain; Modar Alaoui, Eyeris; Pradeep Kumar, Lenovo; and Vassilios Vonikakis, Advanced Digital Sciences Center.The summit will showcase the opportunities of advancing trends in deep learning and the impact on business and society. Explore the latest advances in deep learning technologies like pattern recognition, NLP, neural networks and reinforcement learning, and learn how they will impact will impact communications, manufacturing, healthcare and transportation. Tickets are limited for this event, so book early to avoid disappointment! For more information and to register, please visit the website here.