CVPR Daily - Wednesday

Matthew’s picks of the day (sessions 4, 5, 6): Matthew Vowels is a researcher in the Cog Vis at the Centre for Vision, Speech, and Signal Processing (CVSSP) at the University of Surrey and is working on ways to integrate structure and causality into modern architectures. “ Having been an associate lecturer in transducer engineering for nearly 10 years, I got interested in machine learning for its breadth and scope of application. But before I went back to study it, my wife persuaded me to do a Masters in Psychology with her, and through this I recognized a great need for machine learning models which are amenable to structural/causal constraint according to our theories as scientists. Hence the focus of my work: incorporating and discovering structure and causality with machine learning ” (7) CausalVAE: Disentangled Representation Learning via Neural Structural Causal Models (7) Causal Attention for Vision-Language Tasks (8) ACRE: Abstract Causal REasoning Beyond Covariation (9) Trajectory Prediction With Latent Belief Energy-Based Model (9) Causal Hidden Markov Model for Time Series Disease Forecasting For today, Wednesday 23 @matt_vowels 2 Matthew’s Picks DAILY CVPR Wednesday Likes include: recording covers of childhood anthems, weightlifting, singing at weddings. Dislikes: singing at weddings, hyperparameter search. Learn more about Matthew’s work! “ Come check out our work VDSM: Unsupervised Video Disentanglement with State- Space Modeling and Deep Mixtures of Experts, and our workshop paper Shadow- Mapping for Unsupervised Neural Causal Discovery! ”