Energy-Based Transformers are Scalable Learners and Thinkers

ICLROral2026

Authors: Alexi Gladstone, Ganesh Nanduru, Md Mofijul Islam, Peixuan Han, Hyeonjeong Ha, Aman Chadha, Yilun Du, Heng Ji, Jundong Li, Tariq Iqbal
Affiliation: Flapping Airplanes
Venue: ICLR 2026
Track: Oral

TL;DR

We introduce Energy-Based Transformers, a scalable new approach for learning how to think from unsupervised learning, generalizing current System 2 Thinking/reasoning approaches.

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

energy-based transformer reasoning

← All ICLR 2026 Oral papers · Browse the whole archive