Focus 26 January 2021

AI and machine learning

Computational advances have enabled the deployment of increasingly complex models, which are applied now to a broad-ranging set of fields. This editorial showcase aims at providing a snapshot of the current tools and challenges that are currently holding the promise to change lives in several ways. Herein, we also highlight research on the underlying pursuit of developing the concept of Artificial Intelligence.

Featured articles

Towards universal neural network potential for material discovery applicable to arbitrary combination of 45 elements

Existing neural network potentials are generally designed for narrow target materials. Here the authors develop a neural network potential which is able to handle any combination of 45 elements and show its applicability in multiple domains.
- So Takamoto
- Chikashi Shinagawa
- Takeshi Ibuka
Article Open Access 30 May 2022 Nature Communications
Decoding lip language using triboelectric sensors with deep learning

Lip-language decoding systems are a promising technology to help people lacking a voice live a convenient life with barrier-free communication. Here, authors propose a concept of such system integrating self-powered triboelectric sensors and a well-trained dilated RNN model based on prototype learning.
- Yijia Lu
- Han Tian
- Zhong Lin Wang
Article Open Access 17 Mar 2022 Nature Communications
Forecasting the outcome of spintronic experiments with Neural Ordinary Differential Equations

Deep learning has an increasing impact to assist research. Here, authors show that a dynamical neural network, trained on a minimal amount of data, can predict the behaviour of spintronic devices with high accuracy and an extremely efficient simulation time.
- Xing Chen
- Flavio Abreu Araujo
- Damien Querlioz
Article Open Access 23 Feb 2022 Nature Communications
Uncovering interpretable potential confounders in electronic medical records

Randomized clinical trials are often plagued by selection bias, and expert-selected covariates may insufficiently adjust for confounding factors. Here, the authors develop a framework based on natural language processing to uncover interpretable potential confounders from text.
- Jiaming Zeng
- Michael F. Gensheimer
- Ross D. Shachter
Article Open Access 23 Feb 2022 Nature Communications
Active label cleaning for improved dataset quality under resource constraints

High quality labels are important for model performance, evaluation and selection in medical imaging. As manual labelling is time-consuming and costly, the authors explore and benchmark various resource-effective methods for improving dataset quality.
- Mélanie Bernhardt
- Daniel C. Castro
- Ozan Oktay
Article Open Access 4 Mar 2022 Nature Communications
AI Pontryagin or how artificial neural networks learn to control dynamical systems

Optimal control of complex dynamical systems can be challenging due to cost constraints and analytical intractability. The authors propose a machine-learning-based control framework able to learn control signals and force complex high-dimensional dynamical systems towards a desired target state.
- Lucas Böttcher
- Nino Antulov-Fantulin
- Thomas Asikis
Article Open Access 17 Jan 2022 Nature Communications
Density of states prediction for materials discovery via contrastive learning from probabilistic embeddings

Electrons and phonons give rise to important properties of materials. The machine learning framework Mat2Spec vastly accelerates their computational characterization, enabling discovery of materials for thermoelectrics and solar energy technologies.
- Shufeng Kong
- Francesco Ricci
- John M. Gregoire
Article Open Access 17 Feb 2022 Nature Communications
Artificial-intelligence-driven discovery of catalyst genes with application to CO₂ activation on semiconductor oxides

Here the authors demonstrate an artificial-intelligence based approach to identify catalytic materials features that correlate with mechanisms that trigger, facilitate, or hinder CO2 catalytic reactions.
- Aliaksei Mazheika
- Yang-Gang Wang
- Matthias Scheffler
Article Open Access 20 Jan 2022 Nature Communications
Artificial intelligence-enhanced quantum chemical method with broad applicability

Artificial intelligence is combined with quantum mechanics to break the limitations of traditional methods and create a new general-purpose method for computational chemistry simulations with high accuracy, speed and transferability.
- Peikun Zheng
- Roman Zubatyuk
- Pavlo O. Dral
Article Open Access 2 Dec 2021 Nature Communications
Face detection in untrained deep neural networks

Face-selective neurons are observed in the primate visual pathway and are considered as the basis of face detection in the brain. Here, using a hierarchical deep neural network model of the ventral visual stream, the authors suggest that face selectivity arises in the complete absence of training.
- Seungdae Baek
- Min Song
- Se-Bum Paik
Article Open Access 16 Dec 2021 Nature Communications
A Deep Gravity model for mobility flows generation

The movements of individuals within and among cities influence critical aspects of our society, such as well-being, the spreading of epidemics, and the quality of the environment. Here, the authors use deep neural networks to discover non-linear relationships between geographical variables and mobility flows.
- Filippo Simini
- Gianni Barlacchi
- Luca Pappalardo
Article Open Access 12 Nov 2021 Nature Communications
3D printed biomimetic cochleae and machine learning co-modelling provides clinical informatics for cochlear implant patients

Current spread hampers the efficacy of neuromodulation, while existing animal, in vitro and in silico models have failed to give patient-centric insights. Here the authors employ 3D printing and machine learning to advance clinical predictions of current spread for cochlear implant patients.
- Iek Man Lei
- Chen Jiang
- Yan Yan Shery Huang
Article Open Access 29 Oct 2021 Nature Communications
Neural heterogeneity promotes robust learning

The authors show that heterogeneity in spiking neural networks improves accuracy and robustness of prediction for complex information processing tasks, results in optimal parameter distribution similar to experimental data and is metabolically efficient for learning tasks at varying timescales.
- Nicolas Perez-Nieves
- Vincent C. H. Leung
- Dan F. M. Goodman
Article Open Access 4 Oct 2021 Nature Communications
Embodied intelligence via learning and evolution

The authors propose a new framework, deep evolutionary reinforcement learning, evolves agents with diverse morphologies to learn hard locomotion and manipulation tasks in complex environments, and reveals insights into relations between environmental physics, embodied intelligence, and the evolution of rapid learning.
- Agrim Gupta
- Silvio Savarese
- Li Fei-Fei
Article Open Access 6 Oct 2021 Nature Communications
Next generation reservoir computing

Reservoir computers are artificial neural networks that can be trained on small data sets, but require large random matrices and numerous metaparameters. The authors propose an improved reservoir computer that overcomes these limitations and shows advantageous performance for complex forecasting tasks
- Daniel J. Gauthier
- Erik Bollt
- Wendson A. S. Barbosa
Article Open Access 21 Sep 2021 Nature Communications
Dynamic memory to alleviate catastrophic forgetting in continual learning with medical imaging

In clinical practice, the continuous progress of image acquisition technology or diagnostic procedures and evolving imaging protocols hamper the utility of machine learning, as prediction accuracy on new data deteriorates. Here, the authors propose a continual learning approach to deal with such domain shifts occurring at unknown time points.
- Matthias Perkonigg
- Johannes Hofmanninger
- Georg Langs
Article Open Access 28 Sep 2021 Nature Communications
Element selection for crystalline inorganic solid discovery guided by unsupervised machine learning of experimentally explored chemistry

Machine learning has the potential to significantly speed-up the discovery of new materials in synthetic materials chemistry. Here the authors combine unsupervised machine learning and crystal structure prediction to predict a novel quaternary lithium solid electrolyte that is then synthesized.
- Andrij Vasylenko
- Jacinthe Gamon
- Matthew J. Rosseinsky
Article Open Access 21 Sep 2021 Nature Communications
Understanding the onset of hot streaks across artistic, cultural, and scientific careers

Despite their ubiquitous nature across a wide range of creative domains, it remains unclear if there is any regularity underlying the beginning of successful periods in a career. Here, the authors develop computational methods to trace the career outputs of artists, film directors, and scientists and explore how they move in their creative space along their career trajectory.
- Lu Liu
- Nima Dehmamy
- Dashun Wang
Article Open Access 13 Sep 2021 Nature Communications
Machine learning dismantling and early-warning signals of disintegration in complex systems

Network dismantling allows to find minimum set of units attacking which leads to system’s break down. Grassia et al. propose a deep-learning framework for dismantling of large networks which can be used to quantify the vulnerability of networks and detect early-warning signals of their collapse.
- Marco Grassia
- Manlio De Domenico
- Giuseppe Mangioni
Article Open Access 31 Aug 2021 Nature Communications
Deep neural networks using a single neuron: folded-in-time architecture using feedback-modulated delay loops

Development of deep neural networks benefits from new approaches and perspectives. Stelzer et al. propose to fold a deep neural network of arbitrary size into a single neuron with multiple time-delayed feedback loops which is also of relevance for new hardware implementations and applications.
- Florian Stelzer
- André Röhm
- Serhiy Yanchuk
Article Open Access 27 Aug 2021 Nature Communications
Avalanches and edge-of-chaos learning in neuromorphic nanowire networks

Neuromorphic nanowire networks are found to exhibit neural-like dynamics, including phase transitions and avalanche criticality. Hochstetter and Kuncic et al. show that the dynamical state at the edge-of-chaos is optimal for learning and favours computationally complex information processing tasks.
- Joel Hochstetter
- Ruomin Zhu
- Zdenka Kuncic
Article Open Access 29 Jun 2021 Nature Communications
Advancing diagnostic performance and clinical usability of neural networks via adversarial training and dual batch normalization

Unmasking the decision making process of machine learning models is essential for implementing diagnostic support systems in clinical practice. Here, the authors demonstrate that adversarially trained models can significantly enhance the usability of pathology detection as compared to their standard counterparts.
- Tianyu Han
- Sven Nebelung
- Daniel Truhn
Article Open Access 14 Jul 2021 Nature Communications
Machine learning based energy-free structure predictions of molecules, transition states, and solids

Accurate computational prediction of atomistic structure with traditional methods is challenging. The authors report a kernel-based machine learning model capable of reconstructing 3D atomic coordinates from predicted interatomic distances across a variety of system classes.
- Dominik Lemm
- Guido Falk von Rudorff
- O. Anatole von Lilienfeld
Article Open Access 22 Jul 2021 Nature Communications
A clinical deep learning framework for continually learning from cardiac signals across diseases, time, modalities, and institutions

Deep learning algorithms trained on data streamed temporally from different clinical sites and from a multitude of physiological sensors are generally affected by a degradation in performance. To mitigate this, the authors propose a continual learning strategy that employs a replay buffer.
- Dani Kiyasseh
- Tingting Zhu
- David Clifton
Article Open Access 9 Jul 2021 Nature Communications
Robust learning from noisy, incomplete, high-dimensional experimental data via physically constrained symbolic regression

Reinbold et al. propose a physics-informed data-driven approach that successfully discovers a dynamical model using high-dimensional, noisy and incomplete experimental data describing a weakly turbulent fluid flow. This approach is relevant to other non-equilibrium spatially-extended systems.
- Patrick A. K. Reinbold
- Logan M. Kageorge
- Roman O. Grigoriev
Article Open Access 28 May 2021 Nature Communications
Spectral bias and task-model alignment explain generalization in kernel regression and infinitely wide neural networks

Canatar et al. propose a predictive theory of generalization in kernel regression applicable to real data. This theory explains various generalization phenomena observed in wide neural networks, which admit a kernel limit and generalize well despite being overparameterized.
- Abdulkadir Canatar
- Blake Bordelon
- Cengiz Pehlevan
Article Open Access 18 May 2021 Nature Communications
A graph-based network for predicting chemical reaction pathways in solid-state materials synthesis

Predictive computational approaches are fundamental to accelerating solid-state inorganic synthesis. This work demonstrates a computational tractable approach constructed from available thermochemistry data and based on a graph-based network model for predicting solid-state inorganic reaction pathways.
- Matthew J. McDermott
- Shyam S. Dwaraknath
- Kristin A. Persson
Article Open Access 25 May 2021 Nature Communications
Synaptic metaplasticity in binarized neural networks

Deep neural networks usually rapidly forget the previously learned tasks while training new ones. Laborieux et al. propose a method for training binarized neural networks inspired by neuronal metaplasticity that allows to avoid catastrophic forgetting and is relevant for neuromorphic applications.
- Axel Laborieux
- Maxence Ernoult
- Damien Querlioz
Article Open Access 5 May 2021 Nature Communications
Bias free multiobjective active learning for materials design and discovery

Identifying optimal materials in multiobjective optimization problems represents a challenge for new materials design approaches. Here the authors develop an active-learning algorithm to optimize the Pareto-optimal solutions successfully applied to the in silico polymer design for a dispersant-based application.
- Kevin Maik Jablonka
- Giriprasad Melpatti Jothiappan
- Brian Yoo
Article Open Access 19 Apr 2021 Nature Communications
Qualitative similarities and differences in visual object representations between brains and deep networks

Deep neural networks are widely considered as good models for biological vision. Here, we describe several qualitative similarities and differences in object representations between brains and deep networks that elucidate when deep networks can be considered good models for biological vision and how they can be improved.
- Georgin Jacob
- R. T. Pramod
- S. P. Arun
Article Open Access 25 Mar 2021 Nature Communications
Single-atom level determination of 3-dimensional surface atomic structure via neural network-assisted atomic electron tomography

Precise determination of surface atomic structure of metallic nanoparticles is key to unlock their surface/interface properties. Here the authors introduce a neural network-assisted atomic electron tomography approach that provides a three-dimensional reconstruction of metallic nanoparticles at individual atom level.
- Juhyeok Lee
- Chaehwa Jeong
- Yongsoo Yang
Article Open Access 30 Mar 2021 Nature Communications
Prioritizing non-coding regions based on human genomic constraint and sequence context with deep learning

Intolerance to variation is a strong indicator of disease relevance for coding regions of the human genome. Here, the authors present JARVIS, a deep learning method integrating intolerance to variation in non-coding regions and sequence-specific annotations to infer non-coding variant pathogenicity.
- Dimitrios Vitsios
- Ryan S. Dhindsa
- Slavé Petrovski
Article Open Access 8 Mar 2021 Nature Communications
Predicting orientation-dependent plastic susceptibility from static structure in amorphous solids via deep learning

Predicting a priori local defects in amorphous materials remains a grand challenge. Here authors combine a rotationally non-invariant structure representation with deep-learning to predict the propensity for shear transformations of amorphous solids for different loading orientations, only given the static structure.
- Zhao Fan
- Evan Ma
Article Open Access 8 Mar 2021 Nature Communications
Learning dominant physical processes with data-driven balance models

The dynamics of complex physical systems can be determined by the balance of a few dominant processes. Callaham et al. propose a machine learning approach for the identification of dominant regimes from experimental or numerical data with examples from turbulence, optics, neuroscience, and combustion.
- Jared L. Callaham
- James V. Koch
- Steven L. Brunton
Article Open Access 15 Feb 2021 Nature Communications
Intelligent driving intelligence test for autonomous vehicles with naturalistic and adversarial environment

Tests for autonomous vehicles are usually made in the naturalistic driving environment where safety-critical scenarios are rare. Feng et al. propose a testing approach combining naturalistic and adversarial environment which allows to accelerate testing process and detect dangerous driving events.
- Shuo Feng
- Xintao Yan
- Henry X. Liu
Article Open Access 2 Feb 2021 Nature Communications
Improving the accuracy of medical diagnosis with causal machine learning

In medical diagnosis a doctor aims to explain a patient’s symptoms by determining the diseases causing them, while existing diagnostic algorithms are purely associative. Here, the authors reformulate diagnosis as a counterfactual inference task and derive new counterfactual diagnostic algorithms.
- Jonathan G. Richens
- Ciarán M. Lee
- Saurabh Johri
Article Open Access 11 Aug 2020 Nature Communications
Transforming machine translation: a deep learning system reaches news translation quality comparable to human professionals

The quality of human language translation has been thought to be unattainable by computer translation systems. Here the authors present CUBBITT, a deep learning system that outperforms professional human translators in retaining text meaning in English-to-Czech news translation, and validate the system on English-French and English-Polish language pairs.
- Martin Popel
- Marketa Tomkova
- Zdeněk Žabokrtský
Article Open Access 1 Sep 2020 Nature Communications
Brain-inspired replay for continual learning with artificial neural networks

One challenge that faces artificial intelligence is the inability of deep neural networks to continuously learn new information without catastrophically forgetting what has been learnt before. To solve this problem, here the authors propose a replay-based algorithm for deep learning without the need to store data.
- Gido M. van de Ven
- Hava T. Siegelmann
- Andreas S. Tolias
Article Open Access 13 Aug 2020 Nature Communications
A solution to the learning dilemma for recurrent networks of spiking neurons

Bellec et al. present a mathematically founded approximation for gradient descent training of recurrent neural networks without backwards propagation in time. This enables biologically plausible training of spike-based neural network models with working memory and supports on-chip training of neuromorphic hardware.
- Guillaume Bellec
- Franz Scherr
- Wolfgang Maass
Article Open Access 17 Jul 2020 Nature Communications
Automated extraction of chemical synthesis actions from experimental procedures

Extracting experimental operations for chemical synthesis from procedures reported in prose is a tedious task. Here the authors develop a deep-learning model based on the transformer architecture to translate experimental procedures from the field of organic chemistry into synthesis actions.
- Alain C. Vaucher
- Federico Zipoli
- Teodoro Laino
Article Open Access 17 Jul 2020 Nature Communications
Identifying domains of applicability of machine learning models for materials science

Machine learning models insufficient for certain screening tasks can still provide valuable predictions in specific sub-domains of the considered materials. Here, the authors introduce a diagnostic tool to detect regions of low expected model error as demonstrated for the case of transparent conducting oxides.
- Christopher Sutton
- Mario Boley
- Matthias Scheffler
Article Open Access 4 Sep 2020 Nature Communications
Improving the informativeness of Mendelian disease-derived pathogenicity scores for common disease

Pathogenicity scores are instrumental in prioritizing variants for Mendelian disease, yet their application to common disease is largely unexplored. Here, the authors assess the utility of pathogenicity scores for 41 complex traits and develop a framework to improve their informativeness for common disease.
- Samuel S. Kim
- Kushal K. Dey
- Alkes L. Price
Article Open Access 7 Dec 2020 Nature Communications
Deep learning for genomics using Janggu

Deep learning is becoming a popular approach for understanding biological processes but can be hard to adapt to new questions. Here, the authors develop Janggu, a python library that aims to ease data acquisition and model evaluation and facilitate deep learning applications in genomics.
- Wolfgang Kopp
- Remo Monti
- Altuna Akalin
Article Open Access 13 Jul 2020 Nature Communications
Learning molecular dynamics with simple language model built upon long short-term memory neural network

Artificial neural networks have been successfully used for language recognition. Tsai et al. use the same techniques to link between language processing and prediction of molecular trajectories and show capability to predict complex thermodynamics and kinetics arising in chemical or biological physics.
- Sun-Ting Tsai
- En-Jui Kuo
- Pratyush Tiwary
Article Open Access 9 Oct 2020 Nature Communications
A deep learning system accurately classifies primary and metastatic cancers using passenger mutation patterns

Some cancer patients first present with metastases where the location of the primary is unidentified; these are difficult to treat. In this study, using machine learning, the authors develop a method to determine the tissue of origin of a cancer based on whole sequencing data.
- Wei Jiao
- Gurnit Atwal
- PCAWG Consortium
Article Open Access 5 Feb 2020 Nature Communications
Deep learning radiomics can predict axillary lymph node status in early-stage breast cancer

Breast cancer is frequently diagnosed using ultrasound. Here, the authors show that, in addition to ultrasound, shear wave elastography can be used to diagnose breast cancer and, in conjunction with deep learning and radiomics, can predict whether the disease has spread to axillary lymph nodes.
- Xueyi Zheng
- Zhao Yao
- Jianhua Zhou
Article Open Access 6 Mar 2020 Nature Communications
Designing accurate emulators for scientific processes using calibration-driven deep models

The success of machine learning for scientific discovery normally depends on how well the inherent assumptions match the problem in hand. Here, Thiagarajan et al. alleviate this constraint by allowing the change of optimization criterion in a data-driven approach to emulate complex scientific processes.
- Jayaraman J. Thiagarajan
- Bindya Venkatesh
- Brian Spears
Article Open Access 6 Nov 2020 Nature Communications
Individual differences among deep neural network models

Do artificial neural networks, like brains, exhibit individual differences? Using tools from systems neuroscience, this study reveals substantial variability in network-internal representations, calling into question the neuroscientific practice of using single networks as models of brain function.
- Johannes Mehrer
- Courtney J. Spoerer
- Tim C. Kietzmann
Article Open Access 12 Nov 2020 Nature Communications
Separability and geometry of object manifolds in deep neural networks

Neural activity space or manifold that represents object information changes across the layers of a deep neural network. Here the authors present a theoretical account of the relationship between the geometry of the manifolds and the classification capacity of the neural networks.
- Uri Cohen
- SueYeon Chung
- Haim Sompolinsky
Article Open Access 6 Feb 2020 Nature Communications
Capturing human categorization of natural images by combining deep networks and cognitive models

Theories of human categorization have traditionally been evaluated in the context of simple, low-dimensional stimuli. In this work, the authors use a large dataset of human behavior over 10,000 natural images to re-evaluate these theories, revealing interesting differences from previous results.
- Ruairidh M. Battleday
- Joshua C. Peterson
- Thomas L. Griffiths
Article Open Access 27 Oct 2020 Nature Communications

AI and machine learning

Featured articles

Search

Quick links