ML Pub Club #28: Neural Algorithmic Reasoning: Building Neural Networks that Can Execute Algorithms

​Some of the most powerful ideas in computer science come from classical algorithms - shortest paths, sorting procedures, dynamic programming, flow optimization. They are reliable, efficient and mathematically grounded.

​Machine learning took a very different path: learning patterns directly from data.

​At the upcoming ML Pub Club, we’ll look at what happens when these two traditions meet.

​This talk introduces neural algorithmic reasoning, an emerging research direction where neural networks are trained to reproduce the behavior of classical algorithms. We’ll review how well current models perform at this task and where this approach is already being explored in practice.

​The session will also present KNARsack, a model developed to imitate the decision-making process of the classical Knapsack algorithm. The work was recently presented at the Learning on Graphs conference in Phoenix.

​Our speaker, Stjepan Požgaj, is a PhD student at FER, University of Zagreb, with a background in Computer Science and Mathematics from PMF. His work explores the intersection of algorithmic reasoning and machine learning.

​🗓️ When: Monday 30th | 18:00
📍 Where: CroAI HQ (Zavrtnica 17, 4th floor)
🔗 Registrations via Luma

​If you’re curious about how algorithms and neural networks might evolve together, join us for this session.