MLOps World | Gen AI Summit 2025
Toronto Machine Learning Summit (TMLS) 2025
Scaling Down, Powering Up: Can Efficient Training Beat Scaling Laws?
Speaker
Challenging the assumption that scaling parameters and data is the only path to better LLMs, this talk surveys state-of-the-art data-centric strategies (data mixing, filtering, deduplication) and model-centric approaches (pruning, distillation, quantization, model merging, parameter-efficient fine-tuning, and test-time scaling) that achieve strong performance at a fraction of the cost — with DeepSeek as a compelling real-world case study.