Ilya Schurov

Math, ML and applications. Work at @Radboud_Uni. @[email protected]. In Russian: @ilyaschurov. I condemn Russian aggression in Ukraine.

[Ilya Schurov @ilya_schurov]

I spent some time thinking about the adjoint state method used in the Neural ODEs. Initally, it was a complete mystery for me. After a while, I understood that it's a simple thing (like all other things you finally understand), and it is directly related to backpropagation. A 🧵

[Evgenii Egorov @eeevgen]

I am looking for internship, EU based or remote (I have work permit for EU). I am interested in efficient computation architectures, deep learning for physical simulations. I know and contributed a bit to bayesian deep learning, tensor networks, MCMC methods, SSM&transformers, "equivariance". I can start at January. [email protected] (I also quite enjoy to read a spam).

[Ilya Schurov @ilya_schurov]

@sitnikcode @andreysitnik X asked me to verify my age to show me these “sensitive” photos :)

[Ilya Schurov @ilya_schurov]

In Ukraine, at the same time, incumbents lost the election more often than they won. So, yes, there is a non-elected dictator here, but it is not Zelensky. It is Putin. And it is Putin who started the war

[Ilya Schurov @ilya_schurov]

Since 2012, I was an active electoral observer in Russia. In 2014, when Crimea was already annexed and Russian invasion on Donbass started, I also participated in the presidential elections in Ukraine as an international observer. I can compare the elections between the countries

[Ilya Schurov @ilya_schurov]

Elections in Russia are falsified for a long time. Elections in Ukraine are fair and competitive. In Russia, Putin is stealing elections each time by combination of direct fraud, banning and killing his opponents. Putin changed the constitution to become president five (!) times.

[Ilya Schurov @ilya_schurov]

Similarly, I can dump results of the indermediate step to a file, with a helper function:

def dump(filename):
def wrapper(df):
https://t.co/w8sc2qusga_parquet(filename)
return df
return wrapper

And so on! If you didn't use .pipe before, give it a try!

[Ilya Schurov @ilya_schurov]

I like pandas method chaining and my code usually looks like this:
df.assign(…).query(…).groupby(…)[['some', 'var']].sum().sort_values(…).iloc[:10].mean()
The problem is that it is not easy to get access to the results of intermediate steps. Recently I found a solution!

[Ilya Schurov @ilya_schurov]

Now I wrote a small helper function

def assrt(condition):
def wrapper(df):
assert condition(df)
return df
return wrapper

and use this function with a .pipe method:

df.assign(…).query(…).pipe(assrt(lambda _: not _['column'].isna().any())).groupby(…)…

[Ilya Schurov @ilya_schurov]

AWS Bucket List – Keeps a list of the S3 buckets you swear you’ll organize someday but lets them gather digital dust in the meantime.

[Ilya Schurov @ilya_schurov]

AWS Lambda-fy – Automatically transforms any code you upload into something completely unrecognizable, but somehow still works… sometimes.

[Ilya Schurov @ilya_schurov]

AWS Cloud Naps – Puts your servers to sleep when they’re "overworked" but makes waking them up a total mystery.

[Ilya Schurov @ilya_schurov]

AWS Route 404 – A misdirection service that ensures your users' traffic always ends up in the least expected places for an extra challenge.

[Ilya Schurov @ilya_schurov]

AWS Elastic Mood – Automatically scales your cloud computing resources based on your team’s collective stress levels.

[Ilya Schurov @ilya_schurov]

Okay, probably, AGI is indeed close. I asked ChatGPT to generate funny names for imaginary AWS services, and, you know, some of them are really funny. A thread:

[Ilya Schurov @ilya_schurov]

AWS Cloudburp – A service that "accidentally" reveals hidden cloud storage files you forgot about, kind of like a hiccup in the cloud.

[Konstantin Sonin @k_sonin]

In Russian diaspora, this is a story as big as it can be. The team of the late @navalny, the leader of the Russian opposition, investigated the attempt on @leonidvolkov's life in July 2024, and concluded that it was organized by Leonid Nevzlin, a billionaire exile from @khodorkovsky_en's entourage.

I cannot wrap my head around this information yet. Khodorkovsky spent 10 years in jail; Putin expropriated his multi-billion dollar business. Since 2013, Khodorkovsky was a prominent voice in exile. (Nevzlin fled to Israel in 2003, when Khodorkovsky's empire was under assault from Putin.)

What I can say is that @navalny's team has had an impeccable reputation in investigative journalism. With Alexei and then without him, they uncovered many critical episode's of Putin's regime corruption, their influence networks, their military connections, etc. I'd say that @fbkinfo has been responsible for as many big investigations on Putin than all other world media combined.

Still, it's hard to wrap my head around it.

[Ilya Schurov @ilya_schurov]

This fall I will teach course Scientific Computing at the JetBrains and Constructor University (Bremen) SDT program. The course is aimed at bridging the gap between pure mathematical courses and machine learning. Feel free to subscribe to the course repo https://t.co/X1C6lIjSFD

[Ilya Schurov @ilya_schurov]

@Qusaiabushama @svpino What exactly is problematic in this case? For most of classification models like logistic regression or tree-based methods, there is no assumption that the data is balanced. So, which problem do you encounter, what is the desired behavior and how it differs from the actual one?