How ASTROPHYSICISTS use AI

Dr. Becky
Dr. Becky
15,356
0
Published 2024-08-01
AD - Thanks to Skillshare for sponsoring this video! The first 500 people to use my link will get a 1-month free trial of Skillshare: skl.sh/drbecky07241 | AI, or artificial intelligence, feels like a buzz word you just can’t escape from these days. It is everywhere. And astrophysics is no exception with a steady increase in the number of research papers published either using or mentioning AI tools such as machine learning and deep learning. So in this video let's chat about what actually is AI - what can it do, what can it not do, because there’s a lot of misinformation out there, before we then chat about 4 ways that us astrophysicists use AI: i) classification of data, ii) finding weird things (anomaly detection), iii) inference of data, and iv) emulation of simulations. This is by no means an exhaustive list, I’m sure many of my colleagues are also using AI in other ways for the research, not to mention using well known tools like GitHub’s copilot to help them write code, or chatGPT to help them with inspiration when trying to summarise research papers or grant proposals into abstracts. AI has perhaps changed the way we work as astrophysicists, but has not yet changed astrophysics itself. The key word there though being *yet*, because with the pace of advancement of AI that we’re seeing, who knows what the next decade could bring!

Classify the shapes of galaxies in images from the brand new Euclid telescope to help train an AI deep learning algorithm - galaxyzoo.org/

Check out Katie Bouman's TED talk explaining how the black hole image is made by the Event Horizon Telescope team:    • How to take a picture of a black hole...  

Smith & Geach (2023) - arxiv.org/pdf/2211.03796
Kembhavi & Pattnaik (2022) - link.springer.com/article/10.1007/s12036-022-09871…
Fotopoulou (2024) - arxiv.org/pdf/2406.17316
Huppenkothen et al. (2023) - arxiv.org/pdf/2310.12528
Walmsley et al. (2020) - arxiv.org/pdf/1905.07424
Walmsley et al. (2023) - joss.theoj.org/papers/10.21105/joss.05312
Bowles et al. (2021) - arxiv.org/pdf/2012.01248
Huertas-Company et al. (2023) - arxiv.org/pdf/2305.02478
Robertson et al. (2023) - arxiv.org/pdf/2208.11456
Yu et al. (2019) - arxiv.org/pdf/1904.02726
Osborn et al. (2020) - arxiv.org/pdf/1902.08544
Muthukrishna et al. (2019) - arxiv.org/pdf/1902.08544
Sooknunan et al. (2021) - arxiv.org/pdf/1811.08446
Cheng et al. (2021) - arxiv.org/pdf/2009.11932
Tohill et al. (2024) - arxiv.org/pdf/2306.17225
Muthukrishna et al. (2022) - arxiv.org/pdf/2111.00036
Perez-Carrasco et al. (2023) - iopscience.iop.org/article/10.3847/1538-3881/ace0c…
Mohale & Lochner (2024) -arxiv.org/pdf/2311.14157
Angeloudi et al. (2024) - arxiv.org/pdf/2407.00166
Rose et al. (2024) - arxiv.org/pdf/2405.00766
Conceição et al. (2024) - arxiv.org/pdf/2304.06099

00:00 - Introduction
01:33 - What is AI? Machine learning vs deep learning
05:46 - AD - Skillshare
07:17 - (i) Classification of data
11:21 - (ii) Anomaly detection
13:29 - (iii) Inference of data
15:07 - (iv) Emulation of simulations
17:40 - The impact of AI on astrophysics
19:19 - Bloopers

Video filmed on a Sony ⍺7 IV

---

📚 My new book, "A Brief History of Black Holes", out NOW in hardback, paperback, e-book and audiobook (which I narrated myself!): lnk.to/DrBecky

---

👕 My new merch, including JWST designs, are available here (with worldwide shipping!): dr-becky.teemill.com/

---

🎧 Royal Astronomical Society Podcast that I co-host: podfollow.com/supermassive

---

🔔 Don't forget to subscribe and click the little bell icon to be notified when I post a new video!

---

👩🏽‍💻 I'm Dr. Becky Smethurst, an astrophysicist at the University of Oxford (Christ Church). I love making videos about science with an unnatural level of enthusiasm. I like to focus on how we know things, not just what we know. And especially, the things we still don't know. If you've ever wondered about something in space and couldn't find an answer online - you can ask me! My day job is to do research into how supermassive black holes can affect the galaxies that they live in. In particular, I look at whether the energy output from the disk of material orbiting around a growing supermassive black hole can stop a galaxy from forming stars.


drbecky.uk.com/
rebeccasmethurst.co.uk/
All Comments (21)
  • @charleslaurice
    Dr. Becky 70-year-old man from the Philippines here I am no astrophysicist but I do love your conversations and teachings on the matter. Thank you very very much for finally bringing this conversation to the Internet for us to listen to. I’m gonna listen to it again and again and bring more as soon as you get it thank you for all your effort that you share with us.
  • @dgrossman3
    Galaxy Zoo says "Thanks to all your hard work we've finished classifying everything for now. Please check back soon!"😥
  • @Fortunes.Fool.
    I would laugh if an AI responded to astronomer looking for intelligent life on other planets with, “Bruh, I’m trying to find intelligent life on this damn planet.”
  • @SlimGreen
    i. Classification: Convolutional neural networks based on the visual cortex mechanism in neurobiology in case of image data, k-means in case of clustering ii Anomaly detection: random cut forest iii. Inference: GAN iv: Simulation recurrent neural networks
  • @philurbaniak1811
    👍👍 a-i is everywhere, except Liverpool where they only have a-a-a (and my beloved Manchester, where we only have e-r 😅)
  • @kayinoue2497
    I'm likely going to apply AI/Machine Learning tools for my Ph.D. research to not only classify observed candidates, but also identify the observations with a high signal-to-noise ratio. Honestly, it's such a godsend in so many ways. I'm very against generative AI, but I love deep learning tools and machine learning. It's giving us so many opportunities to apply pattern recognition to enormous data sets like large sky surveys. I'm glad you mentioned the Euclid mission and multidimensional data---as spectra are very key to what I'd like to do. I heard a great interview recently with someone who was using these tools for astrometric mapping, which is definitely underappreciated and giving us a better idea of how our own milky way galaxy is evolving. Obligatory mention though, "Why is my code busted?!"
  • @MarcKoetse
    Extremely well explained! Very useful outside of astrophysics as well! I am going to use this video in the future! Thanks a lot!
  • @smedleyx
    yes I read an article a few years ago where a survey would have been impossible without neural net programming, and made discoveries never foreseen. That's what I always thought AI was for. It's like we use electronic calculation to find Pi to a million digits because we're darn well not going to work that out by hand
  • @matthewrowell8518
    I saw that little snippet of Cleo Abrams in there. 2 of my most absolute favourite YouTube science promoters. Both making the world a great and exciting place
  • @hpmmiggie
    Using a disco ball as an analogy to describe how radio VLBI works is so excellent. Thanks. Will be using that in the future, hope you dont mind.
  • @infinitumneo840
    The limiting factor for AI that we are currently facing is electrical power supply. Expansion of existing data centers has maxed out the existing electrical grid. New data centers are being built in other areas of the grid, but electrical resources are spread very thin. In conjunction with more EVs and the growing digital coin, our current electrical grid is unable to handle all these new pressures. Infrastructure is key, and that cost a lot.
  • @thejll
    Great video! Good list of papers also. Do you know of any astro papers that use regression (not classification) of images?
  • @jorgevicent9808
    In Earth Observation and Earth sciences, AI is also broadly used with similar applications. Some future prospects are explainable (or physics-aware) AI, and causality
  • @lyledal
    Just watched an interview on Bad Boy of Science about AI in particle physics. This stuff is everywhere.
  • @robertvazquez5607
    It's so cool to see your bloopers. Most scientist hate to have there faults exposed.🤓!