<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="3.10.0">Jekyll</generator><link href="https://www.cro.moe/feed.xml" rel="self" type="application/atom+xml" /><link href="https://www.cro.moe/" rel="alternate" type="text/html" /><updated>2025-12-02T07:40:37+00:00</updated><id>https://www.cro.moe/feed.xml</id><title type="html">Uros’ Ongoings</title><subtitle>I am a researcher at the SISSA AstroML group in Trieste. Currently, I&apos;m working on representation learning for irregular time-series.
</subtitle><author><name>Uros Zivanovic</name></author><entry><title type="html">Presenting Romae At Neurips</title><link href="https://www.cro.moe/presenting-romae-at-neurips.html" rel="alternate" type="text/html" title="Presenting Romae At Neurips" /><published>2025-09-22T00:00:00+00:00</published><updated>2025-09-22T00:00:00+00:00</updated><id>https://www.cro.moe/Presenting-RoMAE-at-NeurIPS</id><content type="html" xml:base="https://www.cro.moe/presenting-romae-at-neurips.html"><![CDATA[<p>I’m quite excited to say that our <a href="https://arxiv.org/abs/2505.20535">paper</a> introducing the Rotary Masked Autoencoder (RoMAE)  has been accepted for a poster at this year’s NeurIPS.
For anyone working with irregular time-series or rotary positional embeddings you may find it interesting.
I’m looking forward to using RoMAE in future work for some interesting datasets. 
Specifically, I’d like to do a proper analysis of it’s potential uses with data from the <a href="https://rubinobservatory.org/">LSST</a>.
Particularly exciting is the idea of (<em>nearly</em>) label-free classification through some combination of rep. learning and nearest-neighbors or something similar.
Anyway, check the paper out!</p>]]></content><author><name>Uros Zivanovic</name></author><category term="papers" /><summary type="html"><![CDATA[I’m quite excited to say that our paper introducing the Rotary Masked Autoencoder (RoMAE) has been accepted for a poster at this year’s NeurIPS. For anyone working with irregular time-series or rotary positional embeddings you may find it interesting. I’m looking forward to using RoMAE in future work for some interesting datasets. Specifically, I’d like to do a proper analysis of it’s potential uses with data from the LSST. Particularly exciting is the idea of (nearly) label-free classification through some combination of rep. learning and nearest-neighbors or something similar. Anyway, check the paper out!]]></summary></entry></feed>