The main stairwell and office wing on both second and third floors of the Physics building will be closed until necessary repairs to the main stairwell are completed.
Administrative offices have been relocated to PHY 345.
Please do not cross any caution tapes whilst in the building.

In order to properly clean rooms and buildings due to fire damage, the following classes and midterms (listed by subject and number) being held up to June 15 have been temporarily relocated. To see if your course/midterm has been impacted please visit the Registrar's Temporary Relocations page.

You are here

Roger Melko’s Group is Pioneering Machine Learning Techniques to Solve Fundamental Problems in Condensed Matter

Monday, December 4, 2017

t-sne IsingIn 2012, a group of University of Toronto researchers led by Geoffrey Hinton sparked a revolution in artificial intelligence that has dramatically changed the world. Their breakthrough came after decades of research designing machine learning algorithms for use in computer vision tasks, such as feature extraction from raw image data. Since that flagship work, scientists and engineers have come to recognize that the paradigm of machine learning — or designing algorithms that respond and adapt to new data — provides an exceptionally powerful platform for tackling many difficult tasks in computer science and information technology. This new breed of machine learning algorithm has conquered benchmarks previously thought to be decades away, such as natural language comprehension and game play, known to be highly complex in a mathematical sense.

In condensed matter, the fundamental goal pursued by physicists, including Waterloo’s Roger Melko,  is to understand how macroscopic phenomena observed in nature, such as superconductivity, can arise from the myriad different interactions present between atoms as viewed at the microscopic level. This problem can again be mathematically complex, rendering opaque the relationship between microscopic interactions and real-world phenomena, and leaving the detailed mechanisms underlying many observed phenomena in some matter and materials shrouded in mystery, even to advanced simulation algorithms running on massive supercomputers.

Now, researchers are adopting a wide array of machine learning tools to perform fundamental research into the many-body physics problem.  Industry standard algorithms have been repurposed in striking new ways, ranging from the application of supervised learning for the classification of topological phases, unsupervised learning to perform quantum state tomography, to reinforcement learning of the many-body quantum wavefunction.  In physics, the stakes are high.  The ability to predict phases of matter from the microscopic configurations of atoms would imply the ability to engineer quantum materials, like room temperature superconductors, exotic metals, or novel substrates for quantum computing.  Any improvement made by machine learning to the physicist’s algorithmic arsenal, no matter how small, could have vast consequences for science and technology. This promising approach is being pursued by teams of multi-disciplinary researchers at the University of Waterloo, the Perimeter Institute for Theoretical Physics, and the new Vector Institute for Artificial Intelligence, where Hinton is the Chief Scientific Advisor.  Along with the staggering potential for scientific disruption in our computer simulation technology for condensed matter and materials, physicists are transferring technology back to the tech sector, including the potential use of near-term quantum computers to accelerate certain machine learning tasks.

  1. 2018 (12)
    1. June (2)
    2. May (2)
    3. April (1)
    4. March (4)
    5. February (1)
    6. January (2)
  2. 2017 (19)
    1. December (2)
    2. November (3)
    3. October (1)
    4. September (2)
    5. May (1)
    6. April (4)
    7. March (1)
    8. February (3)
    9. January (2)
  3. 2016 (22)
    1. December (2)
    2. November (1)
    3. October (2)
    4. September (4)
    5. July (2)
    6. June (1)
    7. May (1)
    8. April (2)
    9. March (1)
    10. February (5)
    11. January (1)
  4. 2015 (30)
  5. 2014 (12)
  6. 2013 (7)
  7. 2012 (9)
  8. 2011 (3)
  9. 2010 (7)
  10. 2009 (1)