Premium Only Content

ActInf GuestStream 115.1 ~ Energy-Based Transformers and the Future of Scaling
"Energy-Based Transformers are Scalable Learners and Thinkers"
Alexi Gladstone, Ganesh Nanduru, Md Mofijul Islam, Peixuan Han, Hyeonjeong Ha, Aman Chadha, Yilun Du, Heng Ji, Jundong Li, Tariq Iqbal
https://arxiv.org/abs/2507.02092
Inference-time computation techniques, analogous to human System 2 Thinking, have recently become popular for improving model performances. However, most existing approaches suffer from several limitations: they are modality-specific (e.g., working only in text), problem-specific (e.g., verifiable domains like math and coding), or require additional supervision/training on top of unsupervised pretraining (e.g., verifiers or verifiable rewards). In this paper, we ask the question "Is it possible to generalize these System 2 Thinking approaches, and develop models that learn to think solely from unsupervised learning?" Interestingly, we find the answer is yes, by learning to explicitly verify the compatibility between inputs and candidate-predictions, and then re-framing prediction problems as optimization with respect to this verifier. Specifically, we train Energy-Based Transformers (EBTs) -- a new class of Energy-Based Models (EBMs) -- to assign an energy value to every input and candidate-prediction pair, enabling predictions through gradient descent-based energy minimization until convergence. Across both discrete (text) and continuous (visual) modalities, we find EBTs scale faster than the dominant Transformer++ approach during training, achieving an up to 35% higher scaling rate with respect to data, batch size, parameters, FLOPs, and depth. During inference, EBTs improve performance with System 2 Thinking by 29% more than the Transformer++ on language tasks, and EBTs outperform Diffusion Transformers on image denoising while using fewer forward passes. Further, we find that EBTs achieve better results than existing models on most downstream tasks given the same or worse pretraining performance, suggesting that EBTs generalize better than existing approaches. Consequently, EBTs are a promising new paradigm for scaling both the learning and thinking capabilities of models.
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2507.02092 [cs.LG]
(or arXiv:2507.02092v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2507.02092
Active Inference Institute information:
Website: https://www.activeinference.institute/
Activities: https://activities.activeinference.institute/
Discord: https://discord.activeinference.institute/
Donate: http://donate.activeinference.institute/
YouTube: https://www.youtube.com/c/ActiveInference/
X: https://x.com/InferenceActive
Active Inference Livestreams: https://video.activeinference.institute/
-
1:00:40
Active Inference Institute
2 months agoActInf GuestStream 113.1 ~ Bayesian Mechanics of Economic Choice (Ernesto Moya-Albor et al.)
111 -
3:38:49
Badlands Media
1 day agoThe Narrative Ep. 38: The Sovereign World
104K57 -
2:57:44
The Charlie Kirk Show
12 hours agoWASHINGTON D.C. PRAYER VIGIL FOR CHARLIE KIRK
243K446 -
14:11
Robbi On The Record
13 hours agoThe Trap of Identity Politics: How Division is Killing America
18.8K41 -
1:29:23
Nerdrotic
13 hours ago $19.49 earnedThe Turning Point | New UFO Video with Michael Collins | Forbidden Frontier #117
82.2K29 -
1:08:26
Sarah Westall
10 hours agoSuicide Pacts forming in Youth Social Media Groups - Discord, Reddit, TikTok w/ John Anthony
77.4K26 -
2:25:31
vivafrei
20 hours agoEp. 281: Charlie Kirk; Routh Trial; Charlotte Train; Bolsanaro Defense; SCOTUS & MORE!
157K243 -
2:55:38
Turning Point USA
12 hours agoWASHINGTON D.C. PRAYER VIGIL FOR CHARLIE KIRK
102K44 -
35:54
The Mel K Show
12 hours agoMel K & Tim James | Healing is an Inside Job | 9-14-25
75.6K4 -
3:06:33
IsaiahLCarter
15 hours ago $15.93 earnedCharlie Kirk, American Martyr (with Mikale Olson) || APOSTATE RADIO 028
84.9K30