To celebrate the 10 year anniversary of the first gravitational wave detection, I made a tutorial for high school students that explains how to use python to visualize the original gravitational wave data of GW150914. (Tested and approved by a very motivated German 10th grade student 😄)
#GW10Years

You can also add parmigiano, but I wish I could determine the amount myself (maybe in the next version?)

Have you wondered how to make plotting more fun?
Astrophysicists are there to help you out: Use pastamarker!

How long is "recently" in #ML? 2 month?

6/ 📊 Results:
1️⃣ FMPE performs as well as nested sampling across different noise levels.
2️⃣ FMPE is faster, more scalable, and achieves higher IS efficiencies.
3️⃣ IS not only corrects inaccuracies but also builds confidence in ML-based retrievals.

5/ 🧪 Key innovations:
1️⃣ FMPE allows greater flexibility in neural architectures compared to NPE and trains ~3x faster! 🚀
2️⃣ IS spots model failures, corrects inaccuracies, and facilitates model comparison via evidence ratios.
3️⃣ Noise-level conditioning enables models to adapt to varying error bars.

4/ 🔍 Our approach:
We propose a ML-based framework that tackles these challenges:
✨ Flow Matching Posterior Estimation (FMPE): Fast, flexible, and scalable.
✨ Importance Sampling (IS): Verifies ML results, corrects deviations, and computes the Bayesian evidence.

3/ 🤖 Enter machine learning (ML)!
ML methods like Neural Posterior Estimation (NPE) have shown promise for atmospheric retrieval. But they come with challenges:
1️⃣ Ensuring reliability and accuracy of the estimated posteriors.
2️⃣ Adapting to different noise levels and models on the fly.

2/ 🪐 Context:
To understand how exoplanets form, evolve, and whether they might harbor life, we need to infer their atmospheric properties from observed emission spectra. 🌈
Traditionally, this is done with Bayesian methods like nested sampling, but these are computationally expensive. 💻💰

1/ 🌌 New Paper Alert: How can we decode the atmospheres of exoplanets efficiently and reliably?
The latest work by my amazing collaborator @timothygebhard.bsky.social introduces Flow Matching Posterior Estimation for atmospheric retrieval. 🚀🧵👇
#AI #MachineLearning #Physics #Astronomy #AcademicSky

Take-away: Don't give up 🙃

How would you like to be listed in my academic family tree? 😄 phdcomics.com/comics/archi...

6/ 📊 Key Results
In our experiments:
1️⃣ FAB outperformed traditional methods in high dimensions.
2️⃣ Achieved higher sampling efficiency with fewer evaluations of the target density.
3️⃣ No need for costly pre-generated training data—saving time and compute!

3/ 💡 Surrogate Models
To accelerate sampling from matrix elements, surrogate models (like normalizing flows) are used to approximate these distributions. But how can we train normalizing flows in the best way? We compare different training approaches, including FAB. 🌟

2/ 🎯 The Challenge
The analysis of high-energy physics (HEP) data relies on large amounts of simulated samples drawn from analytically tractable distributions ('matrix elements'). Keeping up with the increasing demand of simulated data is impossible with current methods.

1/ 🚀 New Paper Alert: Spotlight at NeurIPS ML and the Physical Sciences Workshop!
We explore the intersection of high-energy physics and machine learning. What's the challenge we’re targeting, and why does it matter? Let's dive in! 🧵👇
🚀 #AI #MachineLearning #Physics #ML4PS #NeurIPS #AcademicSky

Let's make this a dream-come-true list for #SBI people!
#aiforscience #machinelearning

If you're interested in #MLforGWs, here's a poster on #DINGO.