CV

Antonios Valkanas

Contact Information

• Lab Address: 3380 Blvd Robert-Bourassa, Montreal, QC H2X 2G6
• Email: antonios.valkanas@mail{dot}mcgill{dot}ca
• Website: antonvalk.github.io
• GitHub: github.com/AntonValk
• Status: PhD Student - McGill University / MILA / ILLS

Education

PhD - Electrical & Computer Engineering | McGill University - MILA - ILLS, Montreal, Canada (Sept 2022 - Present)
Advisor: Prof. Mark Coates

• Research areas: Bayesian learning, online learning, probabilistic reasoning with large language models.
• Publications in NeurIPS, ICML, AISTATS, AAAI, TMLR, IEEE Journals & Conferences.

MSc - Electrical & Computer Engineering | McGill University, Montreal, Canada (Sept 2019 - May 2022)
Advisor: Prof. Mark Coates

• Thesis: Graph Modelling of Bag Relations in Multiple Instance Learning.
• Publications in AAAI, IEEE Conferences.

BEng - Electrical Engineering (Honors) | McGill University, Montreal, Canada (Sept 2015 - May 2019)

• Honors thesis advisor: Prof. Dennis Giannacopoulos.

Research Experience

Applied Scientist Intern - Amazon (Seattle, WA, USA) (Jun 2024 - Sept 2024 & May 2025 - Aug 2025)
Managers: Dr. Vijay Nirmalananda (2024), Dr. Shervin Malmasi (2025)

• Developed pricing research models for 100M+ Amazon store items by integrating regression models with M5 CLIP multi-modal text-vision features.
• Built methods for epistemic uncertainty quantification and third-party item price identification using Multi-SWAG.
• Delivered code using AWS, EC2, S3, Python, PyTorch, Git, Docker, and Jupyter; model adopted by the team and deployed to production.

Associate Researcher (Intern) - Quantum Technologies (Montreal, Canada) (Feb 2022 - Sept 2023)
Manager: Yingxue Zhang

• Led literature review, experiment design, and algorithm development for negative item sampling in GNN recommender systems.
• Improved recall@k by 10-20% on large-scale benchmarks (Netflix, Taobao 14, Yelp, Gowalla) and used incremental learning to reduce training time by 80%.
• Tech stack: Python, TensorFlow, Git, Jupyter Notebook, R (ggplot).

Research Scientist Intern - Unity Technologies (Labs, Montreal, Canada) (May 2021 - Sept 2021)
Manager: Dr. Boris Oreshkin

• Tackled high motion-capture costs by building a transformer model for automatic high-quality human motion animation.
• Integrated the model into Unity Editor with engineering partners; responsibilities included data gathering/cleaning, baselines, algorithm design, and experiments.
• Tech stack: Python, PyTorch, Git, Docker, Jupyter Notebook.

Undergraduate Research Assistant - McGill University (Dept. Electrical & Computer Engineering) (2017 - 2019)
Supervisor: Prof. Dennis Giannacopoulos

• Conducted computational electromagnetics research, performing electric field simulations of conductors, designing experiments, and parallelizing code (Java, Python, Git).

Undergraduate Research Assistant - McGill University (Dept. Electrical & Computer Engineering) (2019)
Supervisor: Prof. Harry Leib

• Focused on engineering education; designed a low-cost satellite signal reception system for communication teaching labs.
• Built an antenna with household materials to receive and decode NASA NOAA weather satellite images using SDR.
• Resulting work published in an education journal; tools included SDR# and WXtoIMG.

Professional Experience

• 2025 & 2024: Applied Scientist Intern - Amazon
• 2022: Associate Researcher Intern - Quantum Technologies
• 2021: Research Scientist Intern - Unity Technologies (Labs)
• 2020: Data Scientist Intern - CAE
• 2018-2025: Faculty Lecturer / Teaching Assistant / Course Staff - Dept. Electrical & Computer Engineering, McGill University

Teaching Experience

ECSE 316 - Signals and Networks (McGill University)

• TA & Lab Instructor across 3 semesters (~90 students each) since Fall 2020; contributed new lab material and tooling.
• Faculty lecturer for Fall 2025.

ECSE 416 - Intro to Telecommunications Systems (McGill University)

• TA & Lab Instructor for 3 iterations (Winter 2020-2023, ~40 students/term).
• Received the departmental Teaching Award (best-ranked among 500 courses in the Faculty of Engineering).

ECSE 211 - Design Principles & Methods (McGill University)

• TA & Lab Instructor plus grading across 5 offerings between 2019 and 2022 (~120 students per term).

MAIS 202 - Intro to Machine Learning (McGill AI Society)

• Course lecturer for a zero-credit seminar recognized by the Faculty of Engineering; designed and taught the curriculum for ~30 students over 2 semesters in Fall 2019.

Selected Awards, Fellowships, and Grant Funding

• 2024 - PGSD (NSERC) Doctoral Award ($140,000 over 4 years)
• 2022 - Vadasz Scholar, McGill Engineering Doctoral Award (MEDA) ($128,000 over 4 years)
• 2022 - Niarchos Fellowship for Excellence in Graduate Education ($74,500 over 2 years)
• 2020 - NSERC CGSM Alexander Graham Bell Award ($17,000)
• 2012 - Silver Medal, Youth National Math Olympiad (Greece), awarded by the Hellenic Mathematical Society

Selected Publications

(First author denoted by (*).)

Journal Articles

• () A. Valkanas, Y. Wang, Y. Zhang, M. Coates - *Personalized Negative Reservoir for Incremental Learning in Recommender Systems, Transactions on Machine Learning Research (2025).
• S. Pal, A. Valkanas, M. Coates - Population Monte Carlo with Normalizing Flow, IEEE Signal Processing Letters 31, 16-20 (2023).
• B. N. Oreshkin, () A. Valkanas, F. G. Harvey, L. S. Menard, F. Bocquelet, M. Coates - *Motion In-Betweening via Deep Interpolator, IEEE Transactions on Visualization and Computer Graphics (2024).
• () A. Valkanas, D. Pandey, H. Leib - *Surfing the Radio Spectrum using RTL-SDR, IETE Journal of Education (2019).

Conference Papers

• () A. Valkanas, S. Pal, P. Rumiantsev, Y. Zhang, M. Coates - *C3PO: Optimized Large Language Model Cascades with Probabilistic Cost Constraints for Reasoning, NeurIPS 2025.
• M. A. Alomrani, Y. Zhang, D. Li, Q. Sun, S. Pal, Z. Zhang, Y. Hu, R. D. Ajwani, … A. Valkanas - Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs, arXiv:2507.02076 (2025).
• Y. Zhang, L. Ma, A. Valkanas, B. N. Oreshkin, M. Coates - SKOLR: Structured Koopman Operator Linear RNN for Time-Series Forecasting, ICML 2025.
• () A. Valkanas, B. N. Oreshkin, M. Coates - *MODL: Multilearner Online Deep Learning, AISTATS 2025.
• T. Glavas, J. Chataoui, F. Regol, W. Jabbour, A. Valkanas, B. N. Oreshkin, … - Dynamic Layer Selection in Decoder-Only Transformers, NeurIPS Efficient Natural Language and Speech Processing Workshop (ENLSP-IV) 2024.
• B. Thapaliya, A. Nguyen, Y. Lu, T. Xie, I. Grudetskyi, F. Lin, A. Valkanas, J. Liu, … - ECGN: A Cluster-Aware Approach to Graph Neural Networks for Imbalanced Classification, arXiv:2410.11765 (2024).
• Y. Wang, Y. Zhang, A. Valkanas, R. Tang, C. Ma, J. Hao, M. Coates - Structure Aware Incremental Learning with Personalized Imitation Weights for Recommender Systems, AAAI 2023.
• () A. Valkanas, A.-W. Panzini, M. Coates - *Towards Bayesian Learning of the Architecture, Graph and Parameters for Graph Neural Networks, 56th Asilomar Conference on Signals, Systems, and Computers (2022).
• Y. Zhang, F. Regol, A. Valkanas, M. Coates - Contrastive Learning for Time Series on Dynamic Graphs, EUSIPCO 2022.
• S. Pal, A. Valkanas, F. Regol, M. Coates - Bag Graph: Multiple Instance Learning Using Bayesian Graph Neural Networks, AAAI 2022.
• () A. Valkanas, F. Regol, M. Coates - *Learning from Networks of Distributions, Asilomar Conference on Signals, Systems, and Computers (2020).
• () A. Valkanas, D. Giannacopoulos - *A Neural Network-Based Electromagnetic Simulator, COMPUMAG 2019.

Thesis

• A. Valkanas (2022) - Graph Modelling of Bag Relations in Multiple Instance Learning.

Presentations

Invited Talks

• Winter 2024: Efficient Inference using Large Language Models - Bellairs ML Workshop, Barbados.
• Winter 2023: Online Learning with Multiple Models - Bellairs ML Workshop, Barbados.
• Winter 2022: Human Motion Animation with Transformers - Bellairs ML Workshop, Barbados.
• Winter 2021: Graph-based Multiple Instance Learning - Bellairs Graph Learning Workshop, Barbados.

Contributed Presentations

• () *Deep Learning for Human Motion - Staracom Annual Meeting, Montreal (2023), Best Poster Award ($1000).
• Bayesian Learning of Architecture Graph & Parameters - SURE Fair, McGill (2023); served as co-presenter and mentored an undergraduate student.

Mentoring

• 2022: A.-W. Panzini - Undergraduate Research Intern (McGill SURE Program); role: mentor.

Service & Outreach

Volunteer Work

• 2018: Volunteer Math & Science Tutor - Toronto District School Board (500+ hours).

Peer Review

• Reviewer for NeurIPS, ICML, ICLR, IEEE Journals, AISTATS, Artificial Intelligence Review (Springer), and TMLR.

Professional Memberships

• IEEE - Student Member
• ISBA - Student Member