Owen Lockwood, Jérémy Béjanin, Joost Bus +4 more
Machine learning's appetite for energy and its impatience for results are both growing, and this work proposes meeting them with physics rather than more silicon of the usual kind. The idea behind thermodynamic computing is to let a physical system's natural stochastic behaviour perform the computation instead of simulating that behaviour digitally.
The specific setting is energy-based computing where the process follows Langevin dynamics with tunable energy potentials. Randomness, which ordinary hardware works hard to suppress and machine learning then reintroduces deliberately, becomes the mechanism rather than the nuisance.
Calling it a blueprint is accurate and appropriately modest. The contribution is a coherent account of what such a stack would need at each layer, which is the necessary step before anyone can judge whether the energy advantage survives the engineering.
To address the escalating energy and latency demands of machine-learning workloads, we introduce a blueprint for an energy-efficient and fast thermodynamic computing stack that leverages stochastic analog processes in physical hardware. In this work, we focus on energy-based thermodynamic computing where the stochastic process is well described by Langevin dynamics with tunable energy potentials. The implementation of such potentials in physical hardware enables us to…
Toward Semantic Communication for Real-time Mobile 3D Reconstruction
arXiv (cs.AI) · July 17, 2026CRAFT: Clustering Rubrics to Diagnose Weak LLM Capabilities and Generate Targeted Fine-Tuning Data
arXiv (cs.CL) · July 17, 2026Rate-Utility Frontiers for Language Encodings: Comparing Tokens, Bytes, and Pixels Under Controlled Linguistic Content
arXiv (cs.AI) · July 17, 2026Harmonizing AI Safety Thresholds