Hi, I'm Ben Hoover

I build efficient AI using physics

I use energy-based models to build scalable AI that is smaller, faster, and easier to understand.

Featured in
  • Nature Reviews Physics
  • Quanta Magazine
  • Discover Magazine
  • IBM Research Blog
Ben's headshot
Retrieve memories by descending energy

News

Papers, talks and other updates.

Research

Selected papers, tutorials and interactive tools. The CV lists every publication.

Figure from CRAM

2026

CRAM

1980s physics, trained to reason in milliseconds

Benjamin Hoover, Hugh Blayney, James C. R. Whittington

A learned version of Hopfield and Tank's energy from the 1980s solves 99.996% of Sudoku-Extreme and 100% of Maze-Unique test puzzles after 42 ms and 132 ms of training on one GPU, at least 1,000x faster than looped models.

Figure from NRGPT

ICLR'26

NRGPT

A GPT whose tokens roll down an energy landscape

Nima Dehmamy, Benjamin Hoover, Bishwajit Saha, Leo Kozachkov, Jean-Jacques Slotine, Dmitry Krotov

A minimal modification of GPT unifies it with the energy-based model framework, where inference becomes an exploration of tokens on an energy landscape.

Figure from LSRDAM

NeurIPS'25

LSRDAM

Exact recall, plus new memories that emerge

Benjamin Hoover, Zhaoyang Shi, Krishnakumar Balasubramanian, Dmitry Krotov, Parikshit Ram

  • Spotlight (top 3%)

Dense Associative Memories can produce emergent memories when we replace the Gaussian kernel with a KDE-optimal Epanechnikov kernel.

Figure from Tutorial on Associative Memory

ICML'25

Tutorial on Associative Memory

A modern tour of Associative Memory

Dmitry Krotov, Benjamin Hoover, Parikshit Ram, Bao Pham

Energy-based Associative Memory transformed the field of AI, but it is hardly understood. We present a birds eye view of Associative Memory, beginning with the invention of the Hopfield Network and concluding with modern, dense storage versions that have strong connections to Transformers and kernel methods.

Figure from ConceptAttention

ICML'25

ConceptAttention

Diffusion transformers secretly segment concepts

Alec Helbling, Tuna Han Salih Meral, Benjamin Hoover, Pinar Yanardag, Polo Chau

We discover how to extract highly salient features from the learned representations of Diffusion Transformers, using our technique to segment images by semantic concept. We name our technique ConceptAttention, which outperforms all prior methods on single- and multi-class classification.

Figure from DenseAMs meet Random Features

NeurIPS'24

DenseAMs meet Random Features

Add memories without adding parameters

Benjamin Hoover, Polo Chau, Hendrik Strobelt, Parikshit Ram, Dmitry Krotov

DenseAMs can store an exponential number of memories, but each memory adds new parameters to the energy function. We propose a novel, "Distributed representation for DenseAMs" (DrDAM) that allows us to add new memories without increasing the total number of parameters.

Figure from Transformer Explainer

VIS'24

Transformer Explainer

See inside a live GPT-2, in your browser

Aeree Cho, Grace C. Kim, Alexander Karpekov, Alec Helbling, Zijie J. Wang, Benjamin Hoover, +2 more

Transformers are the most powerful AI innovation of the last decade. Learn how they work by interacting with every mechanic from the comfort of your web browser. Taught in Georgia Tech CSE6242 Data and Visual Analytics (typically 250-300 students per semester).

Figure from Diffusion Explainer

VIS'24

Diffusion Explainer

See how Stable Diffusion turns text into images

Seongmin Lee, Benjamin Hoover, Hendrik Strobelt, Zijie J. Wang, ShengYun Peng, Austin Wright, +4 more

Diffusion models are complicated. We break down Stable Diffusion and explain each component of the model visually. Taught in Georgia Tech CSE6242 Data and Visual Analytics (typically 250-300 students per semester).

Figure from Memory in Plain Sight

AMHN@NeurIPS'23

Memory in Plain Sight

Denoising is memory retrieval in disguise

Benjamin Hoover, Hendrik Strobelt, Dmitry Krotov, Judy Hoffman, Zsolt Kira, Polo Chau

We are the first work to discover that diffusion models perform memory retrieval in their denoising dynamics.

Figure from RXNMapper

Sci. Adv.'21

RXNMapper

Transformers learn atom mapping with no labels

Philippe Schwaller, Benjamin Hoover, JL Reymond, Hendrik Strobelt, Teodoro Laino

We discover that Transformers trained on chemical reactions learn, on their own, how atoms physically rearrange.