Neural operators are emerging as a powerful framework for solving partial differential equations using deep learning. This story covers the basics of neural operators. We show how to derive a neura...

In this post I summarize the paper “Simulating Training Dynamics to Reconstruct Training Data from Deep Neural Networks” by Hanling Tian et al. published at ICLR 2025. It’s about how to recover the...

Data compression has always been an important aspect in research. Codecs are tools that compress and decompress digital media, such as image, video, or audio data. Recently, neural codecs have beco...

This post derives the CCDF of the geometric distribution, including its truncated form for finite support. It also provides some insights regarding the usage of the geometric distribution in the Ne...

This post explains the Gumbel-Softmax trick, a method to sample from a categorical distribution using a differentiable function of noise. Starting from the Gumbel-Max trick, which samples exactly b...

Transformers revolutionized sequence modeling with their attention-based, non-recurrent architecture. This tutorial builds up the transformer model step by step—from self-attention and positional e...

I’ve recently attended the Machine Learning Summer School in Kraków (Poland), where I particularly enjoyed Yarin Gal’s talk on uncertainty in deep learning. In this post, I will summarize the main ...

During the second year of my PhD, when I was in need of inspiration, I stumbled upon a remarkable metric that changed everything: the Maximum Mean Discrepancy (MMD). Since the MMD turned things aro...

Diffusion models have emerged as a leading approach in generative modeling, powering advances in image, audio, and text synthesis. This post offers a concise, math-first walkthrough of how they wor...