Julian Rodemann

🎉 Honored to receive the IJAR Young Researcher Award (🥈)

The award is given to researchers who demonstrate excellence at an early stage of their scientific careers, sponsored by the International Journal of Approximate Reasoning (IJAR).

isipta25.sipta.org/awards

Explaining Bayesian Optimization by Shapley Values Facilitates Human-AI Collaboration Bayesian optimization (BO) with Gaussian processes (GP) has become an indispensable algorithm for black box optimization problems. Not without a dash of irony, BO is often considered a black box itsel...

Read the paper here:
📄 arxiv.org/abs/2403.04629
#MachineLearning #BayesianOptimization #XAI #HumanInTheLoop #Exosuits #ShapleyValues

🗣️ Reviewer feedback on our paper:

“Practical relevance with a well-motivated human-in-the-loop use case.”

“A pleasure to read... good figures, insightful explanations.”

“Novel integration of Shapley values into Bayesian Optimization for interpretability.”

🦿 Exosuits support the lower back during physical labor — like pallet building.
📽️ See them in action:

💡 Why does this matter?

Because it helps personalize wearable robotic exosuits more efficiently — by enabling human users to understand and intervene in the process.

🧠 We open up the black box using Shapley values — to explain why Bayesian Optimization chooses specific parameters.

Bayesian Optimization is great for black box optimization — but often a black box itself.

🔍 Why does it pick the parameters it does?

We wanted to find out.👇

Glad to share this paper was accepted at @ecmlpkdd.org !

We show interpreting BayesOpt helps personalize soft exosuits (see picture)

Great collaboation w/ Fede Croppi, @giuseppe88.bsky.social et al.
@lmumuenchen.bsky.social @munichcenterml.bsky.social
@harvard.edu @wyssinstitute.bsky.social

Guaranteed confidence-band enclosures for PDE surrogates We propose a method for obtaining statistically guaranteed confidence bands for functional machine learning techniques: surrogate models which map between function spaces, motivated by the need build ...

Getting coverage guarantees over functional surrogate models? This is what A. Gray and V. Gopakumar (they did the heavy lifting) have done using conformal predictions and zonotopes over reduced dimensions.

It will be present at UAI 2025 (@auai.org), but a preview is here: arxiv.org/abs/2501.18426.

YouTube video by Imprecise Probabilities Channel of SIPTA SIPTA Seminar by Krikamol Muandet: Imprecise generalisation

In case you are interested in very recent advances of learning under imprecision with probability sets, you can check out the recent and fantastic SIPTA virtual talk by @krikamol.bsky.social, now on youtube: www.youtube.com/watch?v=gGPF...

Helen Alber's DAGStat 2025 talk yesterday showed how LLMs streamline sentence classification and how to fix their mistakes.

💡 The solution? A two-step approach: LLMs pre-classify, experts refine, and Sim-SIMEX corrects errors, boosting efficiency and handling imbalanced data as well as MC-SIMEX.

If you're attending #NeurIPS2024, don't miss our spotlight poster on multicriteria benchmarking TODAY 11am in West Ballroom A-D

Talk: neurips.cc/virtual/2024...

Paper: openreview.net/pdf?id=jXxvS...

Where? West Ballroom A-D #6501

When? Thu 12 Dec 11 a.m. - 2 p.m. local time

#neurips #neurips24

#MachineLearning is all about learning parameters from data, right? Well, not quite…

🤸‍♂️Actually, it sometimes works the other way around.

🤔Curious how?

👉Check out our poster on reciprocal learning: neurips.cc/virtual/2024...

@neuripsconf.bsky.social #NeurIPS2024 #NeurIPS