Marco Mondelli

Professor @ISTAustria. Machine learning, information theory. Prev: @Stanford, @EPFL

[Elisabetta Cornacchia @Elisabetta68885]

📢 We are pleased to announce the 7th edition of Youth in High Dimensions, taking place on July 20th-23rd at ITCP in Trieste.
Apply via our website by April 18th for financial and/or visa support, and by May 9th for all participants: https://t.co/sykexSX0DG.

[Marco Mondelli @MMondelli]

@MountainOfMoon @mahdisoltanol (cnt.) For square loss, we show that gradient descent needs d samples and we have a mild upper bound on step size.

[Marco Mondelli @MMondelli]

@MountainOfMoon @mahdisoltanol Let me clarify. We look at correlation and square loss. For correlation loss, we show that gradient flow needs d samples, while SGD needs d log d. It was argued that the lower bound on SGD holds for the optimal stable step size of order 1/d (see screenshot above).

[Mahdi Soltanolkotabi @mahdisoltanol]

Fun new theory paper on data reuse:

𝗙𝗨𝗟𝗟-𝗕𝗔𝗧𝗖𝗛 𝗚𝗗 can beat 𝗢𝗡𝗘-𝗣𝗔𝗦𝗦 𝗦𝗚𝗗 by a log-factor in samples.

Same single-index model, same data:

→ 𝗚𝗗 recovers with n ≈ d
→ 𝗦𝗚𝗗 needs n ≳ d log d

https://t.co/ZA0LQCpOU1

[Spurious Correlation & Shortcut Learning Workshop @SCSLWorkshop]

✨️📢 Best paper award:
Spurious Correlations in High Dimensional Regression: The Roles of Regularization, Simplicity Bias and Over-Parameterization
🎉👏 @MMondelli @simone_bombari

[Fanny Yang @FannyYangETH]

Interested in theory for high-dimensional statistical problems? The workshop "Youth in High Dimensions" in #Trieste July 7-9 has an amazing speaker lineup and offers great networking opportunities https://t.co/0d8YO0aKHo - co-organized w/ @MMondelli @sebastiangoldt JeanBarbier

[Marco Mondelli @MMondelli]

@ukmlv @IEEEsps Congratulations!!

[Marco Mondelli @MMondelli]

Congratulations @simone_bombari, very well deserved!

[ISTAustria @ISTAustria]

Congrats to @simone_bombari from the Mondelli Group @MMondelli for becoming the first-ever PhD student at @ISTAustria to receive a @Google #Fellowship. Simone's research will focus on the mathematical foundations of deep learning, differential privacy, and adversarial robustness.

[Marco Mondelli @MMondelli]

@AdelJavanmard Congratulations!

[Ananda Theertha Suresh @th33rtha]

We invite expository articles on generative models for the IEEE BITS magazine: https://t.co/olCzlqBMeM

Feel free to reach out to me or other guest editors @abeirami, @MMondelli , @sametoymac, Vatsal Sharan if you have any questions.

[Marco Mondelli @MMondelli]

@ukmlv @ERC_Research @ISTAustria Thank you!

[Marco Mondelli @MMondelli]

Grateful to @ERC_Research and excited to start the project. This would not have been possible without my group at @ISTAustria and my collaborators, thank you!

Multiple open positions (postdocs, PhDs, interns), strong candidates encouraged to reach out. https://t.co/Fu01ECQVy6

[ISTAustria @ISTAustria]

Three @ISTAustria scientists have won @erc_research Starting Grants #ERCStG! A combined 4.8 million euros will go to @MMondelli, @aliciakmichael, and @lisa_Bugnet. Read more about the scientists and their projects 🔗 https://t.co/Jje0lRqZXv

[Marco Mondelli @MMondelli]

@Grigoris_c @ERC_Research @ISTAustria Thanks!

[Marco Mondelli @MMondelli]

@CevherLIONS @ERC_Research @ISTAustria Thank you Volkan!

[Marco Mondelli @MMondelli]

@sebastiangoldt @Sissaschool bravo, Seb!

[Marco Mondelli @MMondelli]

@stefsmlab @ERC_Research @ISTAustria Thank you!

[Marco Mondelli @MMondelli]

@subhabratasen90 @ERC_Research @ISTAustria Thanks!

[Marco Mondelli @MMondelli]

@joanbruna @ERC_Research @ISTAustria Thank you Joan!

[Marco Mondelli @MMondelli]

@mahdi_tcs @ERC_Research @ISTAustria Thanks!

[Marco Mondelli @MMondelli]

@DimitrisPapail @ERC_Research @ISTAustria very kind, Dimitris, thank you!

[Marco Mondelli @MMondelli]

@mahdisoltanol @ERC_Research @ISTAustria Thank you!