Purpose

“The more you know, the less you know”. Still, to this day, I am not able to express this (il)logical statement in a way that is both mathematically abstract and sound (and that I’d be happy with). Theorem: the statement is equivalent up to isomorphism to a weighted sum of all my fervent excitement as I burn new cognizance into my cerebral SSD, and determination to shovel harder and faster into this endless mountain of knowledge. This theorem strongly applies in particular to knowledge relating to “consciousness”, “human intelligence”, “mind”, abstract intuitive notions that we, as Homo sapiens, have yet to provide them with a unifying theory. I am Viet. I have but one purpose: to understand human intelligence and consciousness by creating and studying artificial existences that think and act intelligently.

I got spaghettified into the AI wormhole in my pre-university years where for the first time I conveniently let a machine learn and label anomalies in a dataset for me. With the initiation ritual done, I eagerly proceeded to study introductory machine learning in my first year of college, gaining enough understanding of its applied aspects to research in Prof. Le’s lab at the INRS, Summer 2019, on enhancing Wi-Fi human activity and localization tracking with machine learning algorithms. In Winter 2020, I found my calling in reinforcement learning, a domain I feel to be closer to human intelligence than current statistical methods and deep heuristics. Summer 2020, with Prof. Precup at Mila, I passionately studied provably efficient exploration, a topic I intend to pursue in my graduate career. Provable exploration methods are of primordial importance, as exploration relates to the acquisition of more knowledge and information, provable methods offer fundamental insight into the limitations of our learning process. I recently started my independent research in mathematically formalizing consciousness and relating it to agent behavior in a reinforcement learning setting.

I was lucky to realize the importance of a solid mathematical background. I challenged myself to a wide array of graduate courses such as real and functional analysis, probability, statistical learning theory among others, as well as specialized topics courses such as concentration phenomena (prof. Lin) and mathematical foundations for machine learning (Prof. Panangaden), enabling me to thoroughly digest the high-level abstractions and proofs in the frontier research I was constantly exposed to. Giving abstract neural network propaganda at SUMM 2020 (and to curious peers in the Math lounge) as well as working as teaching assistant for IFT 3395 and 6390 allowed me to reinforce the subset of my mathematical skills applied to machine learning and its subfields.

My journey to further human knowledge and grasp human intelligence is a never-ending one, and my immediate goals up ahead of me are to study and develop the current framework of exploration in reinforcement learning. As a corollary to the theorem and this statement of purpose, years from now I may still be pondering about how we know less when we know more (among other research), vibing in a research institute or in a CS department office. The theorem may follow immediately, to curious readers.