dionhaefner

6 ms·

1.
Tesseract Core: Universal components for differentiable scientific computing(github.com)
2 points by dionhaefner 2mo ago|0 comments
2.
by dionhaefner 2mo ago
No idea lol. I assume it’s possible since both Enzyme and GPU programming are pervasive in Julia. Let us know if you end up trying.
3.
by dionhaefner 2mo ago
Happy to take the blame for the lack of persuasian and engagement then :) Thanks for the feedback, although I’d like to believe there’s a way to have that cake and eat it too.
4.
by dionhaefner 2mo ago
Indeed, just like I let my compiler write (at least) chunks of my AD logic. Not great when the tool becomes a leaky abstraction, but overall net positive don't you think?
5.
by dionhaefner 2mo ago
Not naive! The answer is yes and no. Array layout: yes, for the simple cases this post relies on. In the demo the work arrays are literally calloc'd in a C wrapper and handed straight to the Fortran routine, and it just works. (Column-
6.
Differentiable Fortran with LFortran and Enzyme(docs.pasteurlabs.ai)
57 points by dionhaefner 2mo ago|17 comments
7.
by dionhaefner 2mo ago
Author here — I work on Tesseract at Pasteur Labs, and I wrote this up because the "what if this was possible" was bugging me for way too long :) I was surprised by how well this worked, the LFortran + Enzyme stack seems to be a v
8.
Parametric Shape Optimization of Rocket Fins with Ansys SpaceClaim and PyAnsys(si-tesseract.discourse.group)
3 points by dionhaefner 10mo ago|1 comments
9.
by dionhaefner 10mo ago
This is a case study in making gradient-based optimization (a la gradient descent) work with tools that weren’t designed for it. The goal is optimizing rocket grid fin stiffness (8 bars, 16 angular parameters), with a pipeline that includes
10.
Parametric shape optimization with differentiable FEM simulation(docs.pasteurlabs.ai)
18 points by dionhaefner 1y ago|6 comments
11.
Show HN: Tesseract Core – Autodiff-native, self-documenting software components(github.com)
15 points by dionhaefner 2y ago|0 comments
12.
by dionhaefner 5y ago
Climate models are vastly complex, and you need to bring together many experts from many disciplines to write and maintain one, and analyze the output. This seems to lead to the simplest methods coming out on top. Perhaps it could be solved
13.
by dionhaefner 5y ago
I agree with your point regarding non-composability and ”algorithm lock in” (which may or may not be solvable woth better abstractions), but explicit time stepping schemes are still the main workhorse of global ocean modelling, so I’m not s
14.
by dionhaefner 5y ago
The thing with reduced precision is that things may look fine at first, but then you eventually notice unphysical features in your solution (like additional wave modes after very long simulation times, or energy conservation issues). So we
15.
by dionhaefner 5y ago
True, but unfortunately Pytorch is not quite there yet when it comes to more complex benchmarks: https://github.com/dionhaefner/pyhpc-benchmarks#example-resu... JAX really is the only library that comes close to low-le
16.
by dionhaefner 5y ago
I was mostly referring to the millions (billions?) of dollars getting poured into Python library development by tech companies. With the effect that Python stays relevant and has a thriving library ecosystem. Maybe I'm wrong and Julia
17.
by dionhaefner 5y ago
A bit late to the party, but here are some reasons: - When we started Veros (~4 years ago) Julia was very new on our radar and we didn't know whether it would stick. And to be frank, I'm still not convinced whether it will stick.
18.
Supercharged high-resolution ocean simulation with Jax(dionhaefner.github.io)
90 points by dionhaefner 5y ago|48 comments
19.
by dionhaefner 5y ago
Yes that would probably work! GAMs are a bit of a black box to me so I find it hard to reason about them. I think my method is a bit simpler and it’s more straightforward to get uncertainties, but certainly less elegant than a smooth soluti
20.
by dionhaefner 5y ago
I basically determine p via Bayesian inference within every bin (via a conjugate beta prior for p which gives a beta posterior). If that’s not Bayesian then I don’t know what is :) Yes the pruning can be done with a frequentist method too.
21.
Bayesian histograms for rare event classification(dionhaefner.github.io)
74 points by dionhaefner 5y ago|6 comments
22.
by dionhaefner 5y ago
Yes, it’s possible. There are some proofs of concept for that IIRC. If you prefer to write everything in Javascript you can do that and do all heavy lifting in the client (plus some extra work to provide indexing and metadata like Terracott
23.
by dionhaefner 5y ago
Only if the client has access to the full bit depth. The rendered PNGs are uint8, but the underlying raw data is often stored in 16 or even 32 bit precision.
24.
by dionhaefner 5y ago
I’m not sure which standards you are referring to (I guess that means no). If you have a link or open an issue we could scope this out.
25.
by dionhaefner 5y ago
Rasterio wheels ship with GDAL binaries. No C compiler needed.
26.
by dionhaefner 5y ago
That is true to some degree, but with Terracotta we allow the frontend to do manipulations on the fly. Like changing contrast, applying colormaps, or band math. So the number of possible tiles is infinite. If you don’t need that, you can ju
27.
by dionhaefner 5y ago
I think the point was that Terracotta might be better off using GDAL directly instead of via rasterio. Maybe, but I think mapbox is doing a great job with rasterio, and so far we didn’t run into any unsolvable problems (although we got clos
28.
by dionhaefner 5y ago
Author here. Not really, since we rely on cloud-optimized GeoTiffs for performance, in particular their overviews. This makes it so we can serve tiles from huge datasets (10s of GB) in less than 100ms. I don’t think that would be possible w
29.
by dionhaefner 5y ago
Yes. Fortran is used extensively in climate modelling (even newer models).
30.
Learning to Play Yahtzee with Advantage Actor-Critic (A2C)(dionhaefner.github.io)
2 points by dionhaefner 5y ago|0 comments

More ›