Patrick Cannon (@pw_cannon) 's Twitter Profile
Patrick Cannon

@pw_cannon

ML research scientist

ID: 2776352494

linkhttp://patrickcannon.cc calendar_today19-09-2014 17:10:31

107 Tweet

246 Followers

625 Following

Patrick Cannon (@pw_cannon) 's Twitter Profile Photo

Really wondeful ambient synths from Haruomi Hosono - apparently written to be played in Muji's Tokyo shop in 1984, though only one track was used youtube.com/watch?v=34UutD…

Improbable (@improbableio) 's Twitter Profile Photo

Thanks to WIRED UK for talking to us about the vital role real-world simulation can play in supporting international efforts to respond to the world's most pressing challenges.

Patrick Cannon (@pw_cannon) 's Twitter Profile Photo

Really looking forward to this collaboration! Also my first PhD supervision, which I'm very happy to be sharing with Matteo Fasiolo and Mark Beaumont

Jeremias Knoblauch (@lauchlab) 's Twitter Profile Photo

Saturday morning snack: a 5-minute video by Patrick Cannon summarizing some of our ideas with @SebastianSchmon on connecting generalized Bayesian inference within Approximate Bayesian Inference. youtube.com/watch?v=VWakGn…

Patrick Cannon (@pw_cannon) 's Twitter Profile Photo

ABC is an exact algorithm, targeting a generalised Bayesian posterior arxiv.org/abs/2011.08644 Our new work, accepted at AABI, explores how this view can help us with robustness to misspecification. Joint work with the brilliant @SebastianSchmon & Jeremias Knoblauch.

Sebastian Schmon (@sebayesian) 's Twitter Profile Photo

Approximate Bayesian computation with time series models doesn't have to be difficult! Path signatures can help to automatically design expressive summary statistics. New preprint lead by fantastic PhD student Joel Dyer, joint work with Patrick Cannon: arxiv.org/abs/2106.12555

INET Complexity (@inet_complexity) 's Twitter Profile Photo

Black-box Bayesian inference for economic agent-based models - our new working paper by Joel Dyer, Patrick Cannon, Doyne Farmer and Sebastian Schmon is now online inet.ox.ac.uk/publications/n…

Patrick Cannon (@pw_cannon) 's Twitter Profile Photo

If you’re interested in the intersection of AI and agent-based models, make sure to follow AI for Agent-Based Modelling Community (AI4ABM) — a much needed hub for a really exciting area of research! There’s even an upcoming workshop at ICML Conference

Sebastian Schmon (@sebayesian) 's Twitter Profile Photo

Multi-agent reinforcement learning (MARL) with more than 10000 agents? We (Jordan, Patrick Cannon) are exploring ways to scale MARL to (tens of) thousands of agents using high-performance ABM simulators. Will be presented AI for Agent-Based Modelling Community (AI4ABM) ICML Conference arxiv.org/abs/2207.03945

Joel Dyer (@joelnmdyer) 's Twitter Profile Photo

Thanks to Christian Schroeder de Witt & the organising committee for a great workshop! Proud to have won Best Proposal Paper w/ Sebastian Schmon Patrick Cannon & #DoyneFarmer for our work on fitting #ABMs to microdata w/ #GNNs. Very nice talks from my INET Oxford colleagues Blas Kolic & Aymeric Vié too!

Sebastian Schmon (@sebayesian) 's Twitter Profile Photo

Simulation-based inference (SBI) methods can perform poorly if the model is misspecified. In our #neurips paper we show that model criticism can help robustifying SBI (with RNPE) by learning what aspects a simulator can explain and where it falls short: arxiv.org/abs/2210.06564

Simulation-based inference (SBI) methods can perform poorly if the model is misspecified.

In our #neurips paper we show that model criticism can help robustifying SBI (with RNPE) by learning what aspects a simulator can explain and where it falls short:
arxiv.org/abs/2210.06564
Joel Dyer (@joelnmdyer) 's Twitter Profile Photo

Next week I’ll be UAI 2025 #UAI2024 & AABI #AABI2024 in Vienna to present my Spotlight paper w/ Patrick Cannon & Sebastian Schmon on approx. Bayesian inference for time series simulators using path signatures: openreview.net/forum?id=MsU6s…. Come chat about simulators & #Bayes!

Patrick Cannon (@pw_cannon) 's Twitter Profile Photo

@MendeleySupport My fault - used the wrong email address. Confusingly given "Oops" message rather than "this email address not registered".