Alexander Detkov

Hello, my name is Alex, and I’m currently in the final year of my undergraduate degree in Engineering Physics and Mathematics at the University of Alberta. I am interested in making deep learning more trustworthy and robust by incorporating symbolic reasoning, interpretability, and inductive biases. I am also very excited about the opportunity to create innovative applications of artificial intelligence (AI) that advance our understanding of the natural sciences. Ultimately, I wish to work and contribute towards a cognitive “System 2” AI.

Currently, I am working with Prof. Di Niu on enhancing Graph Neural Network (GNN) performance and explainability with symbolic and reinforcement learning. Previously I interned at Huawei Research for a year working on hardware-aware Neural Architecture Search (NAS) and Convolutional Neural Networks (CNN).

Outside of research, I enjoy reading and playing the piano. My all-time favorite novel is Roadside Picnic by Arkady and Boris Strugatsky. The classic paper by Eugene Wigner “The Unreasonable Effectiveness of Mathematics in the Natural Sciences” is also a favorite of mine. Currently, I am reading “A New Kind of Science” by Stephen Wolfram which takes a novel computational perspective on fundamental physics. On the piano I am learning “Fantaisie-Impromptu in C♯ Minor” by Chopin.

Research Interests

Artificial Intelligence: AI Reasoning, Neuro-symbolic AI, Neural Architecture Search, Trustworthy and Explainable AI (XAI).
Deep Learning: Foundation Models, Convolutional and Graph Neural Networks, Transformers, Geometric Deep Learning.
Natural Sciences: AI-Driven Scientific Discovery, Molecular Dynamics, Computational Physics.

News

[May 2023] Joined Prof. Di Niu to work on explainable Graph Neural Networks.
[Jan. 2023] Our paper on the learning dynamics of Convolutional Neural Networks under reparameterization has been accepted to ICLR 2023.
[June 2022] Presented our work on Molecular Dynamics of polymer imbibition at CSME 2022.
[Jan. 2022] Joined Huawei AI Research to work on Neural Architecture Search.
[May 2022] Joined Computational BioSystems Lab under Prof. Stroberg.

Publications & Talks

ICLR

Reparameterization through Spatial Gradient Scaling

Alexander Detkov, Mohammad Salameh, Muhammad Fetrat Qharabagh, Jialin Zhang, Wei Lui, Shangling Jui, Di Niu

International Conference on Learning Representations (ICLR), 2023.

PDF Poster

CSME

Imbibition of a Single Polymer into a Nanocapillary

Alexander Detkov, Wylie Stroberg

Canadian Society for Mechanical Engineering (CSME), 2022.

PDF Oral Presentation