Dr. Isaac Tamblyn
Dr. Isaac Tamblyn
Research Scientist, National Research Council
Isaac Tamblyn is a research scientist at the National Research Council of Canada. He obtained a PhD in Physics from Dalhousie (2009) as a Killam Scholar. Subsequently, he carried out postdoctoral work at Lawrence Berkeley Laboratory and Lawrence Livermore National Laboratory. He currently leads a team focused on application of artificial intelligence and deep learning to understanding and controlling the properties of materials at the nanoscale.
Untangling nets – understanding deep learning and modern AI | Tuesday, October 2 at 11:15 AM
Deep learning is an approach to machine learning which is built upon mathematical models of large networks of interconnected and interacting neurons. Depending on the connectivity and training method used in the construction of such nets, many different classes of problems can be solved. These include superficially unrelated tasks such as natural language processing and translation, computer vision, image recognition, text-to-speech (and speech-to-text), synthetic data generation and augmentation, dimensionality reduction, and clustering. It is rare to see a new algorithmic paradigm which can be applied so effectively and so broadly.
Deep learning has proven to be particularly powerful when coupled with complementary techniques such as reinforcement learning. Seemingly challenging games such as Atari, Chess, DOTA 2, and most famously Go, have all been mastered by software agents powered by deep neural networks. When trained on modern hardware accelerators, these agents rapidly achieve superhuman skill levels.
The definition of difficult is changing.
We are now in an era where autonomous learning algorithms are widely available and accessible to researchers working within many scientific disciplines. Given the demonstrated successes of deep learning, a growing number of researchers are applying these techniques to longstanding challenges in the physical sciences. Dr. Tamblyn will discuss some of the network architectures which currently exist, how they work, and give examples of what scientists are using them for.