Timothy Gebhard

Use the 4S (Signal-Safe Speckle Subtraction): Explainable Machine Learning reveals the Giant Exoplanet AF Lep b in High-Contrast Imaging Data from 2011

Tue, 01 Apr 2025 00:00:00 +0000

Flow matching for atmospheric retrieval of exoplanets: Where reliability meets adaptive noise levels

Wed, 01 Jan 2025 00:00:00 +0000

Parameterizing pressure-temperature profiles of exoplanet atmospheres with neural networks

Wed, 20 Dec 2023 00:00:00 +0000

Inferring Atmospheric Properties of Exoplanets with Flow Matching and Neural Importance Sampling

Wed, 13 Dec 2023 00:00:00 +0000

What Giordano Bruno did (not) say

Thu, 05 Oct 2023 00:00:00 +0000

I have recently come across the following quote, which is attributed to Giordano Bruno (1548–1600):

In space there are countless constellations, suns and planets; we see only the suns because they give light; the planets remain invisible, for they are small and dark. There are also numberless earths circling around their suns, no worse and no less than this globe of ours.

Isn’t it fantastic? In a time when the prevailing opinion was still that the Earth was at the center of the Universe, and where people—including Giordano Bruno himself—were burned at the stake for daring to challenge this view, Bruno predicted the existence of exoplanets.

CROCODILE: Incorporating medium-resolution spectroscopy of close-in directly imaged exoplanets into atmospheric retrievals via cross-correlation

Wed, 20 Sep 2023 00:00:00 +0000

Chasing rainbows and ocean glints: Inner working angle constraints for the Habitable Worlds Observatory

Thu, 03 Aug 2023 00:00:00 +0000

Comparing Apples with Apples: Robust Detection Limits for Exoplanet High-Contrast Imaging in the Presence of non-Gaussian Noise

Wed, 19 Jul 2023 00:00:00 +0000

Using git-latexdiff for paper rebuttals

Tue, 06 Jun 2023 00:00:00 +0000

As the title already suggests, the following post assumes that you write your research papers with LaTeX (or XeTeX / LuaTeX / …). If you are in one of the unfortunate fields of science that insist on using Word to write papers, your best bet is probably to use the built-in “track changes” feature?

When you write a rebuttal for a research paper, you usually have to attach a revised version of your manuscript where you have marked all the changes that you have made. Talking to some of my colleagues at work recently, I was surprised to learn that many people seem to do this manually: For example, when they change a sentence, they delete the old one and write the new one in bold, using \textbf{}. Obviously, this is a strategy that works, but it is not very efficient: If you go through several rounds of changes or revisions, it quickly becomes difficult to keep track of what is “new” (and relative to what), and of course, once your manuscript is accepted, you have to go through it again to remove all the commands you added to mark up your changes.¹

LaTeX and typography 101

Wed, 03 May 2023 00:00:00 +0000

Over the past 15-or-so years that I’ve been using LaTeX, I have picked up a couple of opinions on how to best prepare a manuscript, both technically and from a typographic perspective. I have decided to collect them here (in no particular order) to make it easier to share them with others.

1. One sentence = one line

In theory, there is nothing that prevents you from putting your entire LaTeX document into just a single line. In practice, obviously, that is a bad idea. A simple trick not to lose your sanity complete when trying to debug a LaTeX document that won’t compile anymore is to place every single sentence on it’s on separate line. That is, one line per sentence. Not one line per paragraph, one sentence. That way, when things do break, and the LaTeX compiler gives you the number of the line that it chokes on, you at least know in which sentence you messed up. It also makes it easier to remove single sentences during editing by simply commenting out the line that contains it.

Finding the smallest separation of Venus and Jupiter

Thu, 02 Mar 2023 00:00:00 +0000

If you have looked up at the (cloud-free) sky around dawn or dusk these days, you may have noticed an unusual celestial sight: Venus and Jupiter in very close conjunction.

The picture is from Michael Khan's Flickr (published under a CC BY-ND 2.0 license) because I did not have much luck with the cloud situation... Check out the full size original, you can even see some of Jupiter's moons!

The reason I bring it up here is that a good friend of mine used it as a sort of astronomical ammunition to nerd-snipe me. He complained that all the popular science articles in mainstream news outlets just mentioned that Venus and Jupiter were going to get really close, but never said when exactly that would happen or how close exactly they would get. He then went on to say:

Keeping track of your top $k$ objects using a heap

Fri, 06 Jan 2023 00:00:00 +0000

When I am writing code for my research, it sometimes happens that I need to keep track of the top $k$ objects, like the top 10 inputs that provided the best accuracy for my machine learning model, or the lowest loss. Of course, if you can fit all your objects in memory, it is easy to use min() or max() to achieve this, but in some cases—like when you process data in batches—it’s more useful to maintain a “running tally.”

Atmospheric retrievals of exoplanets using learned parameterizations of pressure-temperature profiles

Sat, 03 Dec 2022 00:00:00 +0000

Inferring molecular complexity from mass spectrometry data using machine learning

Sat, 03 Dec 2022 00:00:00 +0000

Half-sibling regression meets exoplanet imaging: PSF modeling and subtraction using a flexible, domain knowledge-driven, causal framework

Sat, 01 Oct 2022 00:00:00 +0000

Cartographing our galactic neighborhood

Fri, 09 Sep 2022 00:00:00 +0000

The other day, somebody on one of the Slack workspaces related to my day job asked if anybody had a map of all the stars in our local galactic neighborhood (say, within 20 parsec ≈ 65 lightyears), ideally with spectral types and the number of detected exoplanets. I didn’t, but naive as I was, I thought to myself: “That shouldn’t be too hard to come up with, right? Just query a database and plot the results!” Well, I was not wrong, not technically, but once again, I had not reckoned with the ever-impressive ability of the astronomy community to make their own lives harder by insisting that nothing ever has a unique identifier, but rather 50 or more different names.

HTCondor tips and tricks

Fri, 09 Sep 2022 00:00:00 +0000

Based on the Google results to some of my problems, I sometimes feel like I might be the last person to use the HTCondor batch queue system. I guess the cool kids these days are all using Slurm? Anyway, just in case you are like me (or a future version of me!) and still use HTCondor, here are some useful tricks and commands that I have collected over the years.

See all your running (or held) jobs

To get a list of all your current jobs (plus which machines they are running on), simply use:

Named groups for regex in Python

Sat, 23 Jul 2022 00:00:00 +0000

Regular expressions (regex) are still a bit of my Achilles’ heel in programming. I think they are super cool and useful, but I always have to rely heavily on cheatsheets, StackOverflow, and websites like Pythex to really get them to do what I want. Today was at least the third time in the past couple of weeks that I found myself searching for “named groups” and how exactly they work in Python. Of course, I ended up on the same StackOverflow question as always, and as all the times before, I was briefly confused because the code I wanted to copy is in the question, not in the accepted answer.¹

Of languages and ligatures

Tue, 07 Jun 2022 00:00:00 +0000

People who know me better will probably know that I am a bit of a typography nerd, which is a slightly less cynical way of saying that I sometimes like to spend time getting some details right that other people might not even notice. This is the story of one of these adventures.

The curious case of the missing ligatures

A few weeks ago, I was putting together a little private website using a font that I really like¹, EB Garamond, an modern open-source implementation of a classic Antiqua font dating back to 1592. Because it is such a classic font, I was a little surprised when I noticed that the standard ligatures for letter combinations such as fi, ffi and fl were not working.² Just in case you have never heard about ligatures, this is what I am talking about:

Descriptive placeholder figures with TikZ

Fri, 03 Jun 2022 00:00:00 +0000

When I am just starting to write the first draft of a new paper, I like to use dummy figures as placeholders. This allows me to structure everything and think about the overall “story” without immediately getting myself into the rabbit hole of making nice figures.

There are a number of options out there for placeholder figures in LaTeX, and this StackOverflow thread should give you a nice overview.¹ However, none of them really seems to give me what I would want, namely, the option to specify the size of the figure and add a description of what this figure will contain once I get around to actually making it. So I just made my own macro using TikZ²:

Custom cookie cutters from a 3D printer

Wed, 13 Apr 2022 00:00:00 +0000

Obligatory disclaimer: 3D printing your own custom cookie cutter might be a bad idea. Most filaments (i.e., the plastic that the printer uses) are not food-safe, and even if you do find one that claims to be okay for this purpose, most printer nozzles contain lead, which is probably not something you want to end up in your food. I do not know how much of a problem this really is in practice, but don’t say I didn’t warn you.

Adjusting the spacing of references in $\LaTeX$

Fri, 18 Mar 2022 00:00:00 +0000

Putting the final polishing touches on the paper that I am currently working on, I noticed there was something weird about the spacing of the entries in the bibliography: For some reason, the space between the lines of one entry was larger than the space between different entries!

Here is an example of what I am talking about:

The references with default spacing.

Ugly, right? Now, I do admit that this might not be the most important thing, and that some people perhaps would not even notice this, but to me it stuck out like a sore thumb, and frankly, it bothered me. I have put a lot of work into this paper, and I want it to look good. I mean, wanting your work to look good is how the whole $\TeX$ thing got started in the first place. So I did the inevitable and went on a journey through the depths of StackExchange…

Richer progress bars for _rich_

Wed, 16 Mar 2022 00:00:00 +0000

I recently came across rich, “a Python library for rich text and beautiful formatting in the terminal.” It has about a gazillion features, but the one that I have used the most so far is the progress bar provided in rich.progress.track(), which works very similar to tqdm, but looks much nicer. The only thing that has stopped me from ditching tqdm and replacing it with rich is that the latter, at least by default, does not show the elapsed time or the number of tasks completed.

Can we see JWST using Hubble?

Thu, 06 Jan 2022 00:00:00 +0000

Ever since its Christmas present-like launch about two weeks ago, I have closely followed the progress of the James Webb Space Telescope (JWST, or Webb, for short). During one of the live streams, I was a bit surprised to learn that there are actually no cameras at all on JWST that can be used to visually monitor the progress of the deployments. There are a number of good reasons for this (including, e.g., the fact that they would require artificial light, add weight and complexity, and would have to work in pretty extreme conditions), but for me, that just made the whole thing even more impressive. Talking to a friend about this, we jokingly came up with the idea that, in case something goes wrong and JWST needs debugging, we could try to use the Hubble Space Telescope (Hubble, or HST) to take a closer look!

A simple coin toss game?

Thu, 16 Dec 2021 00:00:00 +0000

Let’s imagine we are playing the following simple game:

In the beginning, we both have $N$ coins. Each round, we flip a coin, and if the result is “heads”, you give me one of your coins, while if the result is “tails”, I give you one of my coins. To make things a bit more interesting, let us assume that the coin is not necessarily fair, and that the probability for “heads” is $p \in (0, 1)$. The game ends as soon as one of us runs out of coins (that person loses).

Physically constrained causal noise models for high-contrast imaging of exoplanets

Mon, 12 Oct 2020 00:00:00 +0000

Enhancing Gravitational-Wave Science with Machine Learning

Thu, 07 May 2020 00:00:00 +0000

Convolutional neural networks: A magic bullet for gravitational-wave detection?

Thu, 26 Sep 2019 00:00:00 +0000

ConvWave: Searching for Gravitational Waves with Fully Convolutional Neural Nets

Fri, 01 Dec 2017 00:00:00 +0000

Software Quality Control at Belle II

Sun, 01 Oct 2017 00:00:00 +0000

Sample Size Estimation for Outlier Detection

Wed, 18 Nov 2015 00:00:00 +0000

Mon, 01 Jan 0001 00:00:00 +0000

About Me

Mon, 01 Jan 0001 00:00:00 +0000

Hi, my name is Timothy!

I am a Ph.D. student at the Max Planck ETH Center for Learning Systems (CLS), which means my studies are shared between ETH Zürich and the Max Planck Institute for Intelligent Systems in Tübingen. My supervisors are Bernhard Schölkopf (in Tübingen) and Sascha Quanz (in Zürich).

Before I began my Ph.D. in September 2018, I studied physics at the Karlsruhe Institute of Technology, earning both a Bachelor's and a Master's degree with distinction. During my studies, I was generously supported by the Studienstiftung des deutschen Volkes.

Imprint and Privacy Information

Mon, 01 Jan 0001 00:00:00 +0000

Contact Information

Address:
Timothy Gebhard
℅ Max-Planck-Institut für Intelligente Systeme
Max-Planck-Ring 4
DE-72076 Tübingen

E-Mail: contact@timothygebhard.de
Website: https://timothygebhard.de

What personal data do I collect and why do I collect it?

No personal data is collected on this website. If you disagree or if I made a mistake, please contact me immediately (see above).

Links to other websites

Other websites referenced here may collect information about you, use cookies, embed additional tracking services from third parties, and record your interaction with that embedded content, including your interaction with the embedded content if you have an account and are signed in to that website.