Amir Khoshaman

Photo of Amir Khoshaman


PhD Electrical & Computer Engineering, University of British Columbia

I am currently developing ideas on interlacing graphical models into the fabric of deep learning to make them more interpretable with immediate applications from Natural Language Processing. I am also fascinated by the theory of learning and am working on understanding the high-dimensional parameter space of neural nets, and come up with models that are more parsimonious in their parameters, while not sacrificing their expressibility.

Before joining Borealis AI, I developed generative models with discrete latent variables that achieved state-of-art performance and resulted in top-tier publications. With my previous team, I came up with “Quantum Variational Autoencoder,” the first model that uses quantum mechanics to perform machine learning on large-scale datasets.

During my PhD studies, I contributed to 30+ publications in the field of non-equilibrium statistical mechanics.

I’m an avid reader with interests in philosophy, classics, history, psychology, science and fiction. I’ve been reading an average of two books per week since 2011. My two favourite books of all the time are “The fabric of reality” and “The beginning of infinity,” both by David Deutsch. These books have deeply influenced my take on epistemology.

Research Areas

Natural Language Generation

Bayesian Inference

Deep Learning

Optimization in Machine Learning

Picture of Amir Khoshaman