The Calgary Machine Learning Lab is a research group led by Yani Ioannou within the Schulich School of Engineering at the University of Calgary. Our research is driven by the overarching goal of advancing efficient, trustworthy, and accessible Artificial Intelligence.
Lab photo from May 2026 outside the ICT building at the University of Calgary.
Central to our work is the concept of sparse neural network training and inference, which we pursue with four key motivations:
- Democratize AI: removing redundant computation, making state-of-the-art models accessible to all
- Sustainable and Trustworthy AI: by fundamentally reducing the carbon footprint of these models while rigorously auditing how compression impacts algorithmic bias, we are working to ensure that real-world deployment of next-generation AI is both sustainable and safer
- Learning Structure in Neural Networks: automatically learn neural network topologies tailored for novel data domains
- Understanding the Mechanics of Neural Network Training: sparse neural network training provides a unique theoretical lens to better understand the underlying principles of neural network training, and its remarkable effectiveness.
news
| Jul 01, 2026 | Yani Ioannou was promoted to the position of Associate Professor, with Tenure, at the University of Calgary. |
|---|---|
| Apr 30, 2026 | Adnan Mohammed’s paper, “SparseOpt: Addressing Normalization-induced Gradient Skew in Sparse Training” (Adnan et al., 2026), has been accepted at the International Conference on Machine Learning (ICML), 2026. This work introduces SparseOpt, a novel optimization algorithm designed to mitigate the issue of gradient skew caused by normalization in sparse training. SparseOpt effectively balances the gradient updates across all parameters, leading to improved convergence and performance in sparse neural networks. |
| Jan 26, 2026 | Mike Lasby’s collaborative work with Cerebras, “REAP the experts: Why pruning prevails for one-shot moe compression” (Lasby et al., 2026), has been accepted at the International Conference on Learning Representations (ICLR), 2026. This work explores the compression of Sparse Mixture of Experts (SMoE) models through expert compression techniques, demonstrating that REAP (Router-weighted Expert Activation Pruning) outperforms existing expert merging and pruning methods in terms of compressed model quality retention. |
