Unassigned · 2026-11-11
An Energy-Based Perspective on Machine Learning
Bo Dai, Yilun Du
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors present a tutorial using energy-based models (EBMs) as a unifying lens. Any EBM is defined by its energy (what it represents), its inference procedure (sampling or optimization), and its learning method (how it fits data). They argue many discriminative, generative, representation, and planning/RL methods can be cast as EBMs that mainly differ in partition-function handling and sampling, and they suggest new algorithms via component substitution and energy composition.
Why it matters
A single EBM framework can organize diverse ML methods—discriminative, generative, representation learning, and decision-making—by mapping them to the same three choices: energy, inference, and learning. The authors also emphasize composability via adding/combining energies, enabling reuse and adapting pretrained components for new tasks at inference time without retraining separate predictors for each combination.
Method
- Define EBM probability densities via an energy: pθ(x) ∝ exp(−Eθ(x)). Joint/conditional forms highlight the role of the partition function Z and unknown/intractable normalization.
- Unify learning methods (maximum likelihood, noise contrastive estimation, pseudolikelihood, score matching, adversarial training) under shared contrastive structure.
- Organize methods by how they differ in handling partition functions and sampling; propose new methods by replacing one EBM component while keeping others, using compositional energy addition at inference time.
Abstract (from arXiv)
Energy-based models (EBMs) define a distribution through a scalar energy, $p_\theta(x)\propto\exp(-E_\theta(x))$, and are usually presented as one model family among many. This tutorial has a broader goal: beyond introducing EBMs, it uses them as a single lens that connects the different aspects of machine learning. Any EBM is specified by three choices, the energy, which determines what the model represents; the inference procedure, which samples from or optimizes the energy; and the learning method, which fits the energy to data. We first develop these components, including sampling, variational inference, and learning by maximum likelihood, noise contrastive estimation, pseudolikelihood, score matching, and adversarial training, and show that the learning methods share one contrastive structure. We then show that discriminative models, generative models such as GANs, normalizing flows, diffusion and autoregressive models, representation learning methods such as contrastive learning, word2vec, BERT, denoising auto-encoders, and diffusion representations, reinforcement learning with EBM policies and world models, inverse reinforcement learning, and planning, as well as inference-time computation with verifiers, are all EBMs that differ in these three choices, mostly in how they handle the partition function and sampling. This view explains existing methods, organizes them into a common framework, and suggests new algorithms by replacing one component while keeping the others, with composition of energies as a principled way to combine learned models at inference time.
Related papers
- VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning
- CAST: Causal Advantage-Structured Training with Spatially Grounded Compositional Rewards for Diffusion Models
- FewStepRL: Fine-tuning Few-step Diffusion Models with Reinforcement Learning
- Statistical Foundations of Reinforcement Learning: A Non-Asymptotic Perspective
- DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-11-10 · Indexed by Pub-AI: 2026-11-11 UTC
An Energy-Based Perspective on Machine Learning
Preprint posted: 2026-11-10 · Original record ↗
- arxiv
- 2611.14383
Explicit source versions (1)
- v1 · 2026-11-10
- Preprint available
Source date: 2026-11-10 · First detected: 2026-11-11 UTC · Historical baseline
Source evidence (2)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.14383v1; arxiv:announce_type=new; item pubDate (official metadata)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Energy-based models (EBMs) define a distribution through a scalar energy, $p_θ(x)\propto\exp(-E_θ(x))$, and are usually presented as one model family among many. This tutorial has a broader goal: beyond introducing EBMs, it uses them as a single lens that connects the different aspects of machine learning. Any EBM is specified by three choices, the energy, which determines what the model represents; the inference procedure, which samples from or optimizes the energy; and the learning method, which fits the energy to data. We first develop these components, including sampling, variational inference, and learning by maximum likelihood, noise contrastive estimation, pseudolikelihood, score matching, and adversarial training, and show that the learning methods share one contrastive structure. We then show that discriminative models, generative models such as GANs, normalizing flows, diffusion and autoregressive models, representation learning methods such as contrastive learning, word2vec, BERT, denoising auto-encoders, and diffusion representations, reinforcement learning with EBM policies and world models, inverse reinforcement learning, and planning, as well as inference-time computation with verifiers, are all EBMs that differ in these three choices, mostly in how they handle the partition function and sampling. This view explains existing methods, organizes them into a common framework, and suggests new algorithms by replacing one component while keeping the others, with composition of energies as a principled way to combine learned models at inference time.
