CPAL: Conference on Parsimony and Learning

CPAL 2026: the Conference on Parsimony and Learning is dedicated to the principle that the best models are often the simplest ones that still explain the data — a principle with deep roots in statistics, signal processing, and cognitive science that is gaining renewed urgency in the era of overparameterized foundation models. Topics include sparse coding and compressed sensing, model compression and pruning, structured sparsity, parsimonious representation learning, optimization for sparse models, theoretical understanding of implicit regularization, and the application of parsimony principles to large-scale deep learning.

CPAL brings together researchers from machine learning, statistics, signal processing, neuroscience, and mathematics who share an interest in efficiency, interpretability, and principled model selection. Practitioners working on deploying models under memory or compute constraints will find the compression and efficiency research directly applicable. Graduate students interested in the foundations of learning will find CPAL's theoretical depth and distinctive perspective on efficiency a valuable complement to the dominant paradigm of ever-larger models.

sparse coding model compression representation learning optimization deep learning signal processing foundation models structured sparsity