Research areas
- switching-state-estimation
- Probabilistic graphical models, dynamic Bayesian networks
- Bayesian machine learning, approximate inference
- State-space models, hidden Markov models, switching Kalman filters
- Author of Machine Learning: A Probabilistic Perspective (MIT Press, 2012) and Probabilistic Machine Learning: An Introduction / Advanced Topics (MIT Press, 2022 / 2023) — three of the field’s standard graduate textbooks
Key papers
- murphy-1998-switching-kalman-filters — unifies GPB1, GPB2, IMM, variational SKF, and MHT under a single moment-matching collapse + EM framework; the reference technical report on SKF inference and learning
Recent work
- Probabilistic Machine Learning textbook series (MIT Press, 2022–2023)
- Bayesian deep learning, sequential Monte Carlo, JAX-based probabilistic programming at Google DeepMind
Collaborators
- Stuart Russell (PhD advisor, UC Berkeley)
- Zoubin Ghahramani (collaborator on switching state-space models)
- Geoffrey Hinton (Toronto / Google)
My notes
Murphy’s 1998 SKF tech report is the canonical “what is a switching Kalman
filter and why is exact inference intractable” reference. Its taxonomy
(collapsing / selection / sampling / variational) maps cleanly onto every
modern switching-filter algorithm — including the Rao-Blackwellised particle
filter used in SimMdlPrices/src/rbpf.jl (a selection-class member that
keeps regime histories and runs an exact Kalman filter conditional on each
path).