Deep learning with STRUCTURE

Deep learning with STRUCTURE

Charlie Tang is a PhD student in the machine learning group at the University of Toronto, working with Geoffrey Hinton and Ruslan Salakhutdinov, whose the interests include machine learning, computer vision and cognitive science. More specifically, he had developed various higher-order extensions to generative models in deep learning for vision.

At the deep learning Summit in Boston next month, Charlie'll present ' deeplearning with Structure'. Supervised neural networks trained on massive datasets has recently achieved impressive performance in computer vision, s Peech recognition, and many other tasks. While extremely flexible, neural nets is often criticized because their internal representations is distributed codes an D lack interpretability; During his presentation at the summit, Charlie would reveal how we can address some of these concerns.

We had a quick q&a with Charlie ahead of the "deep learning Summit" to "hear more" on thoughts and CH Allenges in deep learning.

What is the key factors that has the enabled recent advancements in deep learning?
The three key factors are:
-The steadfast belief and knowledge that supervised neural networks trained with enough labelled data can achieve great T EST set generalization.
-The availability of high performance hardware and software, in particular, Nvidia's CUDA architecture and SDK. This allowed more experimentation and the learning from large-scale data.
-The development of superior models:switching to rectified linear hidden units from the sigmoid or hyperbolic tangent un Its and the invention of regularization techniques, specifically "dropout".

What is the main types of problems now being addressed in the deep learning space?
Almost all problems in statistical machine learning is currently being investigated using deep learning techniques. They include visual and speech recognition, reinforcement learning, natural language processing, medical and health applic Ations, financial engineering and many others.

What is the practical applications of your work and what sectors is most likely to be affected?
The deep learning revolution allows models trained in big data to drastically improve accuracy. This means, many artificial intelligence recognition tasks can is now automated, which previously necessitated a human In-the-loop.

What developments can we expect to see in deep learning in the next 5 years?
Deep learning algorithms would be gradually adopted for more tasks and would "solve" more problems. For example, 5 years ago, algorithmic face recognition accuracy was still somewhat worse than human performance. However, currently, super-human performances is reported on the main face recognition Datasets (LFW) and the standard imag E classification DataSet (Imagenet). In the next 5 years, harder and harder problems such as video recognition, medical imaging or text processing would be SUCC Essfully tackled by deep learning algorithms. We can also expect deep learning algorithms to being ported to commercial products, much what's the face detector is Incor porated to consumer cameras in the past years.

What advancements excite is the most in the field?
I feel like the most exciting advance are the availability of Low-energy mobile hardware that supports deep learning Algori THMs. This would inevitably leads to many real-time systems and mobile products which would be a part of our daily lives.

The deep learning Summit are taking place in Boston on 26-27. For more information and to registers, please visit the event website here.

Join the conversation with the event hashtag #reworkDL

Deep learning with STRUCTURE