iiinova — where your digital self keeps evolving

Paul Brunzema

@pbrunzema.bsky.social

37 followers 57 following 16 posts

PhD student on learning-based control; interested in everything around Bayesian optimization and robotics

Posts Replies Media

Paul Brunzema @pbrunzema.bsky.social · 28/09/2026

Excellent blog post worth reading IMO! :) The title is also the TLDR blog.colinbreck.com/i-dont-want-...

blog.colinbreck.com

I Don’t Want to Read What You Didn’t Write

Intentional writing will likely become more valuable.

000

Paul Brunzema @pbrunzema.bsky.social · 22/08/2026

Two weeks ago, we uploaded our preprint on agentic Bayesian optimization from my internship in NYC. TLDR is that we place an agent at the center of the BO loop to leverage both, fine-grained numerical search with the vast prior knowledge embedded in LLMs. arxiv.org/abs/2608.00316

arxiv.org

Agentic Bayesian Optimization through Surrogate-Augmented Autoresearch

Bayesian optimization (BO) has become the standard tool for sample-efficient optimization and owes its efficiency to uncertainty-aware search driven by generic statistical priors. Richer domain priors...

020

Paul Brunzema @pbrunzema.bsky.social · 20/08/2026

Played around with local LLMs for the first time today—there is something magical about knowing the the model is running on your GPU as opposed to getting responses from the magic API to frontier models

000

Paul Brunzema @pbrunzema.bsky.social · 13/08/2026

On one of the five papers I reviewed for NeurIPS there was one thread of a author-reviewer discussion where both sides clearly copy pasted LLM outputs—this got to a point where after multiple back-and-forth I was no longer able to understand what they were even discussing anymore…

110

Paul Brunzema @pbrunzema.bsky.social · 26/04/2025

If you’re attending AISTATS this year, be sure to check out Mikkel’s paper on insights into deep ensembles!

010

Paul Brunzema @pbrunzema.bsky.social · 22/04/2025

📣 Excited to present our ICLR spotlight paper at #ICLR2025! "Bayesian Optimization via Continual Variational Last Layer Training" 📍 Poster #368 (Hall 3 + Hall 2B) ⏰ Friday during Poster Session 3 If you're in Singapore and interested in BO or Bayesian methods, definitely stop by! More infos👇

111

Paul Brunzema @pbrunzema.bsky.social · 25/01/2025

Our paper got accepted to ICLR 2025! 🎉 Looking forward to meeting everyone in Singapur!

011

Reposted by Paul Brunzema

BÜNDNIS 90/DIE GRÜNEN @gruene.de · 20/01/2025

Wenn Donald Trump mit „America First“ zu Nationalismus aufruft, rufen wir mit 450 Millionen Europäer*innen zurück: „Europe United“ – Zusammenhalt, Freiheit, Demokratie in einem vereinten, starken Europa. Dafür braucht es uns alle. 👉🏻 gruene.de/mitglied-werden

Blaue Kappe mit der Aufschrift „EUROPE UNITED“, umgeben von Sternen; darüber Text: „Unsere Antwort auf America First?“

331030207

Paul Brunzema @pbrunzema.bsky.social · 17/12/2024

Beautiful Vancouver—enjoying some days off after an inspiring yet exhausting NeurIPS 2024. With 16,500 attendees, countless technical discussions, and a few nice beers on the side, it’s been a great experience!

An image of a tall building in Vancouver with the sun in the background.

Image of an ice hockey game taken from the upper stands. Vancouver is leading 2:0.

010

Paul Brunzema @pbrunzema.bsky.social · 14/12/2024

📚 New Paper with collaborators from DTU, Vector, & Google DeepMind A neural net-based approach to BO that performs well in both classic, small-scale problems, and can efficiently scale far beyond GP surrogate models. Visit our poster @ NeurIPS Bayesian decision-making workshop today! More info👇

BO with Variational Last Layers as Surrogate

163