[PDG 456] Hierarchical Reasoning Model

Link to article: https://arxiv.org/pdf/2506.21734
Title: Hierarchical Reasoning Model
Content: The Hierarchical Reasoning Model (HRM) is a novel 27-million-parameter recurrent architecture inspired by the human brain's multi-timescale processing, using two interdependent modules for abstract planning and detailed computations to perform complex reasoning in a single forward pass. Despite its small size and minimal training requirements (only 1000 samples, no pre-training), HRM achieves near-perfect performance on challenging tasks like Sudoku and maze-solving while outperforming much larger models on key benchmarks including the Abstraction and Reasoning Corpus.
Slack link: ml-ka.slack.com, channel: #pdg. Please join us -- if you cannot join, please message us here or to mlpaperdiscussiongroupka@gmail.com.

In the Paper Discussion Group (PDG) we discuss recent and fundamental papers in the area of machine learning on a weekly basis. If you are interested, please read the paper beforehand and join us for the discussion. If you have not fully understood the paper, you can still participate – everyone is welcome! You can join the discussion or simply listen in. The discussion is in German or English depending on the participants.