@grrddm

Bayesian methods and machine learning.

 London, England
grdm.io
Joined May 2016
  • Tweets189
  • Following124
  • Followers622
62Media
Gerardo Duran-Martin retweeted
Karl Deisseroth
@KarlDeisserothOct 5
Thanks to everyone sharing in this moment! I’m so grateful to my trainees & collaborators… dlab.stanford.edu/group-memb… (pic: optogenetics conference chairs, Tuscany) Reading for all those asking: for technical details: dlab.stanford.edu for everyone: deisseroth.org/projections
173
392
37
4,415
133,312
Gerardo Duran-Martin retweeted
Stat.ML Papers@StatMLPapersApr 13
A Predictive View on Streaming Hidden Markov Models ift.tt/rvSYRI5

A Predictive View on Streaming Hidden Markov Models

We develop a predictive-first optimisation framework for streaming hidden Markov models. Unlike classical approaches that prioritise full posterior recovery under a fully specified generative...

arxiv.org
5
23
1,602
Notes on recursive learning and regime detection using hidden Markov models (#HMMs) grdm.io/posts/hidden-markov-…

Streaming hidden Markov models

A one-dimensional introduction on streaming hidden Markov models (HMMs) for regime detection and prediction.

grdm.io
1
1
158
Implementing an two-hidden layer MLP, training it with adam, and plotting an animated loss curve, using only javascript + HTML is now just a prompt...
Claude
@claudeaiMar 12
Claude can now build interactive charts and diagrams, directly in the chat. Available today in beta on all plans, including free. Try it out: claude.ai
3
7
143
26,528
Our paper “Martingale Posterior Neural Networks for Fast Sequential Decision Making” has been accepted at #neurips2025! Joint work with @l_sbetancourt, @AlvaroCartea and @sirbayes Blog: grdm.io/posts/bnn-without-ba… Paper: arxiv.org/abs/2506.11898 Code: github.com/gerdm/martingale-…
5
34
11
319
116,291
more replies
2. In sequential decision making, we find that sampling from the posterior predictive is 10-100x faster than sampling from the posterior while delivering comparable or superior performance in sequential decision making tasks.
1
2
10
1,746
3. It enables us to separate Bayesian uncertainty quantification from model parameter updates. In particular, we propose a new frequentist method for sequentially training neural networks that retains Bayesian-style uncertainty while being faster and more data efficient.
1
12
1,454
Had a great time presenting our BONE paper at the Sensorimotor AI Journal Club! If you’re interested in the foundations of learning, these sessions are definitely worth checking out. Excited for the upcoming RL debate series!
Hadi Vafaii@hadivafaii5 Sep 2025
In the 2nd meeting of 'Sensorimotor AI Journal Club' we covered: 📜(B)ayesian (O)nline learning in (N)on-stationary (E)nvironments (BONE; Duran-Martin et al. 2025) Many thanks to the first author, @grrddm, for an amazing presentation! 📽️ watch it here: youtube.com/watch?v=49PPmv9I…

Bayesian online learning in non-stationary environments (BONE) by...

In our second meeting, we discussed BONE: https://openreview.net/fo...

youtube.com
1
5
913
Gerardo Duran-Martin retweeted
Hadi Vafaii@hadivafaii9 Jul 2025
Announcing the new "Sensorimotor AI" Journal Club — please share/repost! w/ Kaylene Stocking, @TommSalvatori, and @EliSennesh Details + sign up link in the first reply 👇
1
10
37
2,112
Some thoughts on Gaussian Processes, uncertainty quantification, and whether some methods are inherently "Bayesian". grdm.io/posts/bayes-free-gau…

Do Gaussian processes really need Bayes?

A frequentist view of Gaussian processes for regression as best linear unbiased predictors.

grdm.io
1
11
585
My PhD thesis is online! I explore how a classical framework—Bayesian filtering—can tackle modern challenges in online continual learning, bandits, and one-step-ahead forecasting. Enables adaptivity, robustness, and scalability. arxiv.org/abs/2505.07267 grdm.io/posts/bayesian-onlin…
1
5
29
1,687
Gerardo Duran-Martin retweeted
UCL CSML@uclcsml4 Jun 2025
The next seminar is this Friday (June 6th) and starts at 12pm midday UK time! Gerardo Duran Martin from Oxford is going to talk about “A unifying framework for generalized Bayesian online learning in non-stationary environment”! ucl.zoom.us/j/99748820264 This seminar is hybrid. More info 👉ucl-ellis.github.io/jt_csml_…

Join our Cloud HD Video Meeting

Zoom is the leader in modern enterprise cloud communications.

ucl.zoom.us
4
9
1,260
A Bayesian’s take on filtering without Bayes. Part III: The Kalman filter. In this post, we walk through the derivation of the Kalman filter without priors or posteriors and explore its application to time-series forecasting and online learning. grdm.io/posts/filtering-note…
6
28
1,322
Happy to announce that our paper on (generalised) Bayesian online learning in non-stationary environments has been accepted at @TmlrOrg! 🦴 This is joint work with @l_sbetancourt, A. Shestopaloff, and @sirbayes. arxiv.org/abs/2411.10153
5
1
30
11,274
Part II of a Bayesian’s take on (Kalman) filtering—without Bayes. All about the basics of state-space models and their measures of uncertainty. grdm.io/posts/filtering-note…

Filtering notes (II): state-space models

Introduction to linear state-space models: basic properties, measures of uncertainty, and the ARMA model.

grdm.io
2
3
39
2,511
Gerardo Duran-Martin retweeted
Jeremias Knoblauch@LauchLab26 Feb 2025
We are excited to announce that registration for the inaugural post-Bayes workshop on May 15./16. at UCL is now open! Website: postbayes.github.io/workshop… Registration link: tinyurl.com/postBayesWorksho…
7
3
20
5,369
Part I of a Bayesian’s take on (Kalman) filtering—without Bayes. I will cover filtering and applications—from classical tracking problems to time-series forecasting, sequential learning for neural networks, and fully-online reinforcement learning gerdm.github.io/posts/filter…

Filtering notes (I): signal plus noise models

Part I of the filtering notes series: Part I: signal plus noise models Part II: state-space-models Part III: state-space-models Introduction From tracking epidemics and predicting financial markets...

grdm.io
1
8
41
2,726
Gerardo Duran-Martin retweeted
Kevin Patrick Murphy
@sirbayes18 Nov 2024
I'm happy to announce our new review paper on methods for online prediction (and segmentation) in the presence of non-stationarity, covering both gradual and sudden changes. arxiv.org/abs/2411.10153 Led by @grrddm with Leandro Sánchez-Betancourt and Alex Shestopaloff.
3
43
341
33,626
Load more