Demon in the Machine: Learning to Extract Work and Absorb Entropy from Fluctuating Nanosystems

Article Physics, Multidisciplinary

Bayesian Information Engine that Optimally Exploits Noisy Measurements

Tushar K. Saha et al.

Summary: This paper presents an experimental information engine consisting of a trapped bead in water, discussing its performance under different measurement noise and signal-to-noise ratio conditions.

PHYSICAL REVIEW LETTERS (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

Training neural networks using Metropolis Monte Carlo and an adaptive variant

Stephen Whitelam et al.

Summary: The zero-temperature Metropolis Monte Carlo (MC) algorithm is examined as a tool for training neural networks. It can effectively train neural networks with comparable accuracy to gradient descent (GD), although not necessarily as quickly. The adaptive Monte Carlo algorithm (aMC) is introduced to overcome limitations when the network structure or neuron activations are strongly heterogeneous. The MC method allows training of deep neural networks and recurrent neural networks that cannot be trained by GD due to insignificant or excessive gradients. MC methods offer a complementary approach to gradient-based methods for training neural networks, providing access to different network architectures and principles.

MACHINE LEARNING-SCIENCE AND TECHNOLOGY (2022)

Add to Collection

Article Physics, Fluids & Plasmas

Limited-control optimal protocols arbitrarily far from equilibrium

Adrianne Zhong et al.

Summary: Recent studies have investigated the finite-time dissipation-minimizing protocols for stochastic thermodynamic systems under full external control. However, in real-world scenarios, systems are often subject to limited control. This study presents exact finite-time optimal protocols for the limited control setting, going beyond previous slow-and-fast-driving approximations. The problem is formulated within the framework of optimal control theory, and the exact optimal protocol can be obtained by solving a system of Hamiltonian partial differential equations. The proposed method outperforms previous approaches and demonstrates improved performance in both analytical and numerical examples.

PHYSICAL REVIEW E (2022)

Add to Collection

Review Multidisciplinary Sciences

A feedback control principle common to several biological and engineered systems

Jonathan Y. Suen et al.

Summary: This study investigates the parallels between discrete-event feedback control strategies in biological and engineered systems. The researchers find that common engineering rules are also used by diverse biological systems, and these rules have several goals in optimizing the behavior of the systems.

JOURNAL OF THE ROYAL SOCIETY INTERFACE (2022)

Add to Collection

Article Multidisciplinary Sciences

Anomalous collapses of Nares Strait ice arches leads to enhanced export of Arctic sea ice

G. W. K. Moore et al.

Summary: Ice arches at the northern and southern ends of Nares Strait, a key passage in the Arctic, are forming for shorter durations, leading to increased ice transport and accelerating the export of multi-year ice.

NATURE COMMUNICATIONS (2021)

Add to Collection

Article Multidisciplinary Sciences

Designing self-assembling kinetics with differentiable statistical physics models

Carl P. Goodrich et al.

Summary: The text discusses the importance of the inverse problem of designing component interactions to target emergent structure and kinetics in various fields such as biotechnology, materials science, and statistical physics. By utilizing automatic differentiation, precise kinetic pathways can be designed and manipulated to control the rate of specific structure formation in systems. This approach can also provide insights into the high-dimensional design space and predict the control of multiple kinetic features simultaneously and independently.

PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Data-driven discovery of Koopman eigenfunctions for control

Eurika Kaiser et al.

Summary: Data-driven transformations can reformulate nonlinear systems into a linear framework for prediction, estimation, and control using linear systems theory. By validating eigenfunctions and constructing reduced-order models, improved predictive power is achieved. The KRONIC control architecture is demonstrated on various examples, showing the effectiveness of the approach in different scenarios such as nonlinear systems, Hamiltonian systems, and ocean mixing models.

MACHINE LEARNING-SCIENCE AND TECHNOLOGY (2021)

Add to Collection

Article Physics, Multidisciplinary

Deep reinforcement learning for feedback control in a collective flashing ratchet

Dong-Kyum Kim et al.

Summary: In this study, optimal policies for particle transport system were discovered using deep reinforcement learning, where neural network architecture demonstrated improved performance over previous strategies. Additionally, it was shown that the policies provided by deep RL outperformed previous strategies even in a time-delayed feedback scenario.

PHYSICAL REVIEW RESEARCH (2021)

Add to Collection

Article Physics, Fluids & Plasmas