Fast and Accurate Probabilistic Time Series Classification

Explore cutting-edge techniques for time series classification in this 47-minute talk by Daniel Schmidt from the Finnish Center for Artificial Intelligence FCAI. Delve into the challenges of assigning labels to time series data for applications like gesture recognition and land use classification. Learn about MINIROCKET, a novel approach that achieves near-state-of-the-art performance while being significantly faster than existing methods like HIVE-COTE 2.0. Discover how MINIROCKET utilizes linear combinations of pooled convolutional filters to process datasets in minutes rather than weeks. Examine recent advancements in producing probabilistic predictions using L2-regularized multinomial logistic regression models, enabling well-calibrated probabilistic outputs without substantial computational overhead. Gain insights from Schmidt's expertise in statistical genomics, Bayesian inference, and machine learning education as he presents this innovative research in time series classification and forecasting at scale.

Daniel Schmidt: Fast and Accurate Probabilistic Time Series Classification