Nikan Doosti

PhD Candidate @AaltoUniversity |
Machine Learning, Digital Fabrication |
Ex-Founder | Mentor |

Aligning with RNGs

[Nikan Doosti @NIkronic]

An awesome event given my passion for Graphics4Science/Eng by pushing explicit representation, here geometry!

[Lennart Bastian @lennart_bastian]

Inspired by #CVPR2026?

📣 Announcing: Call For Papers on Geometric Intelligence @ ECCV26!

https://t.co/NNDb2aw7pa

Submission deadline: 13th of July

👀 Computer vision and AI are increasingly playing a role in scientific discovery. 1/n

[Nikan Doosti @NIkronic]

My entire timeline is about @CVPR talks/papers/highlights/etc, and as a person with FOMO, I'm in pain while browsing the articles. :)
Hoping for the future...

[Nikan Doosti @NIkronic]

@thaoshibe I think what @f_dangel did was a novel, fun and sustainable idea!
Multipurpose functionality in a cool way :)

Ref: A poster towel https://t.co/BU5qzaNOgP

[Nikan Doosti @NIkronic]

@keenanisalive Wow great!
So great that one doesn't need to enable AI-hype-resistance features to learn from your talk! 😍

[Nikan Doosti @NIkronic]

I've managed to DOUBLED the number of attendees in 'Info Workshops' at Aalto by changing the style of events!

I gave an honest and "brutal" feedback (very rare here) to the managers of these events. I expected punishment as usual of being honest, but was invited to collaborate!

So I'd say this is very few events that my honesty has been appreciated rather than punished.
So the design of the material now follows more down to earth content, even the audience might feel offended, uncomfortable, etc.

[Nikan Doosti @NIkronic]

During my master's thesis, for a period I had only access to my $200 student laptop, hence couldn't run any big simulation or training.
The result was that I independently introduced a Fourier feature upscaling for topology optimization which was included in the paper I published

[kache @yacineMTB]

if you're doing AI research at all; I recommend doing the "ETH zurich" route

Train models that use a single GPU. Make sure that it takes less than a minute to train models. Pufferlib is a great example.

The more models you train the more you learn

[Nikan Doosti @NIkronic]

@istvan_csanady That's the line of problem I'm trying to address by combining data-driven and explicit (geometry) representation as a differentiable optimization layer, making it fabrication-aware.

Refs about world models:
https://t.co/kB9FoDm75I
https://t.co/ZjgU8YJvPr

[Keenan Crane @keenanisalive]

I don't have a strong opinion about whether video models “understand the world.”

But I do think the first bar should be checking whether you can recover consistent geometry from video—not whether it makes accurate predictions of physics.

(“Accurate physics” is not even well-posed unless the geometry defining the physical experiment is well-defined.)

[Nikan Doosti @NIkronic]

Punishment for being HONEST in academic work:
- https://t.co/FlzYDIETUf holds a research school program, in Poland, Italy, ...
- the entire trip, accommodation, etc. all are funded.

So far, amazing.
- Then you learn that almost everyone have lied in their application, and .... maybe you should've lied to in yours too.

Details:
- For the admission criteria, an MS Excel document was shared and you should evaluate yourself.
- The questions are like "how do you FEEL about your English level?"
- Guess what? IELTS 6.5 has marked.
- I thought ok, that's gonna get rejected, so I filled my application honestly.
- Guess what? I got rejected and those who lied in their entire application got in!
- A friend who had the qualification exactly told me that EVERYONE LIED!

This is super sad, since I'm attending top conferences that the school grant doesn't cover via personal savings, yet if you lie in a obvious manner thanks to astonishingly bad coordinators of this programs (check box excel for funded research school), you get it.

Just a reminder that these are tax-funded programs, so I believe it's good idea that every single person should be able to elaborate how coordinators of these program are spending people's tax!

[Nikan Doosti @NIkronic]

@taiyasaki Before starting my research journey, senior PhDs whom I talked to, told me that you'd think your main job is doing research and publish. Then you're required to do influencer, youtuber, blogger, etc works as an unpaid overtime work :)

[Nikan Doosti @NIkronic]

Unfortunately this is a common problem.
Similar to open review that each author must be verified (e.g., institutional email), each submission should go through all the authors mentioned, and they must one-by-one confirm the summitted article.

It doesn't fix it entirely, but at least protects those who are willing to be responsible. (sorta they get a "verified" status in arxiv submission)

[Justin Solomon @JustinMSolomon]

🚀 I'm delighted to share an **open postdoc position** joint between our group at @MIT_CSAIL and @MITIBMLab, mentored by me, Ankit Gupta, @LChoshen, and friends; the start date can be as early as fall 2026.

[Nikan Doosti @NIkronic]

@CSProfKGD @giffmana ngl, @giffmana's gaming background and this tshirt gives me enough space to state what I think without self-censorship; across science direction, work, etc.

I'm sure even a coffee break with Lucas and Kosta would be memorable one! 😍

[Nikan Doosti @NIkronic]

@LocasaleLab That's why I call "FakedIn" (=LinkedIn) speech a skill for researchers as advanced salesman :)

[Nikan Doosti @NIkronic]

About the reminders, I was wondering the fact that Profs are being bombarded by emails is actually the reason we must send a reminder or two over the course of 3/4 weeks?

PS: this is a totally emotional take, but I spend days preparing my cold emails to Profs and their students. As a result I'd feel down if I consider that I didn't get any response because my email were lost in the spam garbage! :)

[Nikan Doosti @NIkronic]

@nmwsharp Hopefully I can get in and have a talk about inverse design and geometric dl.

[Nikan Doosti @NIkronic]

That's why I think integrating explicit underlying (physics, geometry, etc) representation as a differentiable inductive bias would allow scaling, which is the core of line of research I'd like to pursue.

Simply put, why waste parameters and data on learning things we already know (the said explicit representation).

[Nikan Doosti @NIkronic]

@pmddomingos academics who want to advance science and academics who just want to boost citations, both love AI. Unfortunately the latter are the majority.