Featured
-
-
Article
| Open AccessHigher-order Granger reservoir computing: simultaneously achieving scalable complex structures inference and accurate dynamics prediction
For reservoir computing, improving prediction accuracy while maintaining low computing complexity remains a challenge. Inspired by the Granger causality, Li et al. design a data-driven and model-free framework by integrating the inference process and the inferred results on high-order structures.
- Xin Li
- , Qunxi Zhu
- & Wei Lin
-
Article
| Open AccessTacticAI: an AI assistant for football tactics
In modern football games, data-driven analysis serves as a key driver in determining tactics. Wang, Veličković, Hennes et al. develop a geometric deep learning algorithm, named TacticAI, to solve high-dimensional learning tasks over corner kicks and suggest tactics favoured over existing ones 90% of the time.
- Zhe Wang
- , Petar Veličković
- & Karl Tuyls
-
Article
| Open AccessThe bone ecosystem facilitates multiple myeloma relapse and the evolution of heterogeneous drug resistant disease
Here, the authors develop a hybrid agent-based model to quantify the contributions of intrinsic cellular mechanisms and bone ecosystem factors to therapy resistance in multiple myeloma. They show that intrinsic mechanisms are essential for resistance, and that the bone microenvironment provides a protective niche that increases the likelihood.
- Ryan T. Bishop
- , Anna K. Miller
- & David Basanta
-
Article
| Open AccessA control oriented strategy of disruption prediction to avoid the configuration collapse of tokamak reactors
Confining plasma and managing disruptions in tokamak devices is a challenge. Here the authors demonstrate a method predicting and possibly preventing disruptions and macroscopic instabilities in tokamak plasma using data from JET.
- Andrea Murari
- , Riccardo Rossi
- & Michela Gelfusa
-
Article
| Open AccessFast Human Motion reconstruction from sparse inertial measurement units considering the human shape
Inertial Measurement Units-based motion capture effective application in large scale and complex environments depends on improved efficiency and reduced latency. Here, authors propose a full body motion estimation deep neural network based on 6 IMUs, which runs at 65 fps with 15 ms latency on an embedded computer.
- Xuan Xiao
- , Jianjian Wang
- & Jianfu Zhang
-
Article
| Open AccessBidirectional generation of structure and properties through a single molecular foundation model
Multimodal pre-training approaches on the molecule domain were limited. Here, authors propose a multimodal molecular pre-trained model including molecular structure and biochemical properties and apply it to downstream tasks related with both molecule structure and properties.
- Jinho Chang
- & Jong Chul Ye
-
Article
| Open AccessUnderstanding quantum machine learning also requires rethinking generalization
Understanding machine learning models’ ability to extrapolate from training data to unseen data - known as generalisation - has recently undergone a paradigm shift, while a similar understanding for their quantum counterparts is still missing. Here, the authors show that uniform generalization bounds pessimistically estimate the performance of quantum machine learning models.
- Elies Gil-Fuster
- , Jens Eisert
- & Carlos Bravo-Prieto
-
Article
| Open AccessCharting cellular differentiation trajectories with Ricci flow
When stem cells develop into tissues intracellular signalling is rewired, errors in this process lead to cancer. Here, authors applied tools from differential geometry made by Albert Einstein’s General Relativity to understand and predict biological network rewiring in health and disease.
- Anthony Baptista
- , Ben D. MacArthur
- & Christopher R. S. Banerji
-
Article
| Open AccessPredicting multiple observations in complex systems through low-dimensional embeddings
Forecasting the future behaviors based on observed data remains a challenging task especially for large nonlinear systems. The authors propose a data-driven approach combining manifold learning and delay embeddings for prediction of dynamics for all components in high-dimensional systems.
- Tao Wu
- , Xiangyun Gao
- & Jürgen Kurths
-
Article
| Open AccessQuantifying how single dose Ad26.COV2.S vaccine efficacy depends on Spike sequence features
SARS-CoV-2 variants with mutations in spike have emerged during the pandemic. Magaret et al. show that in Latin America, efficacy of the Ad26.COV2.S vaccine against moderate to severe–critical COVID-19 varied by sequence features, antibody escape scores, and neutralization impacting features of the SARS-CoV-2 variant.
- Craig A. Magaret
- , Li Li
- & Peter B. Gilbert
-
Article
| Open AccessUnderstanding the infection severity and epidemiological characteristics of mpox in the UK
Mpox cases without known travel links to endemic countries began to be detected in the UK in mid-2022. In this study, the authors characterise the severity of mpox cases in the UK and estimate the overall infection hospitalisation risk at ~4%.
- Thomas Ward
- , Christopher E. Overton
- & Martyn Fyles
-
Article
| Open AccessDrivers and impact of the early silent invasion of SARS-CoV-2 Alpha
The SARS-CoV-2 Alpha variant of concern emerged in the UK in late 2020 but spread internationally before it was detected. Here, the authors reconstruct the dynamics of dissemination of this variant out of the UK by combining extent of genomic sequencing, travel volume, and local epidemic dynamics in a Bayesian model.
- Benjamin Faucher
- , Chiara E. Sabbatini
- & Chiara Poletto
-
Perspective
| Open AccessEmerging opportunities and challenges for the future of reservoir computing
Reservoir Computing has shown advantageous performance in signal processing and learning tasks due to compact design and ability for fast training. Here, the authors discuss the parallel progress of mathematical theory, algorithm design and experimental realizations of Reservoir Computers, and identify emerging opportunities as well as existing challenges for their large-scale industrial adoption.
- Min Yan
- , Can Huang
- & Jie Sun
-
Article
| Open AccessSystematic analysis of ChatGPT, Google search and Llama 2 for clinical decision support tasks
People will likely use ChatGPT to seek health advice. Here, the authors show promising performance of ChatGPT and open source models, but a lack of high accuracy considering medical question answering. Improvements are expected over time via domain-specific finetuning and integration of regulations.
- Sarah Sandmann
- , Sarah Riepenhausen
- & Julian Varghese
-
Article
| Open AccessA programmable hybrid digital chemical information processor based on the Belousov-Zhabotinsky reaction
Computing platforms based on chemical processes can be an alternative to digital computers in some scenarios but have limited programmability. Here the authors demonstrate a hybrid computing platform combining digital electronics and an oscillatory chemical reaction and demonstrate its computational capabilities.
- Abhishek Sharma
- , Marcus Tze-Kiat Ng
- & Leroy Cronin
-
Article
| Open AccessPrediction of highly stable 2D carbon allotropes based on azulenoid kekulene
Discovery of 2D materials with useful electronic properties is challenging. Here, the authors use DFT to design a stable semiconducting 2D carbon allotrope for optoelectronic applications that has light charge carriers and unusual secondary bandgap.
- Zhenzhe Zhang
- , Hanh D. M. Pham
- & Rustam Z. Khaliullin
-
Article
| Open AccessA comprehensive transformer-based approach for high-accuracy gas adsorption predictions in metal-organic frameworks
Three-dimensional representation learning is efficient in material science. Here, authors proposed a transformer-based framework for multi-purpose gas adsorption prediction. Predicted values correspond with the outcomes of adsorption experiments.
- Jingqi Wang
- , Jiapeng Liu
- & Diannan Lu
-
Article
| Open AccessEmpirical data drift detection experiments on real-world medical imaging data
Data drift is the systematic change in the underlying distribution of input features in prediction models, and can cause deterioration in model performance. Here, the authors highlight the importance of detecting data drift in clinical settings and evaluate methods for detecting drift in medical image data.
- Ali Kore
- , Elyar Abbasi Bavil
- & Mohamed Abdalla
-
Article
| Open AccessLearning stochastic dynamics and predicting emergent behavior using transformers
Learning the dynamics governing a simulation or experiment usually requires coarse graining or projection, as the number of transition rates typically grows exponentially with system size. The authors show that transformers, neural networks introduced initially for natural language processing, can be used to parameterize the dynamics of large systems without coarse graining.
- Corneel Casert
- , Isaac Tamblyn
- & Stephen Whitelam
-
Article
| Open AccessLearning the intrinsic dynamics of spatio-temporal processes through Latent Dynamics Networks
Predicting the evolution of dynamical systems remains challenging, requiring high computational effort or effective reduction of the system into a low-dimensional space. Here, the authors present a data-driven approach for predicting the evolution of systems exhibiting spatiotemporal dynamics in response to external input signals.
- Francesco Regazzoni
- , Stefano Pagani
- & Alfio Quarteroni
-
Article
| Open AccessData leakage inflates prediction performance in connectome-based machine learning models
The effects of data leakage on predictive models in neuroimaging studies are not well understood. Here, the authors show that data leakage via feature selection and repeated subjects drastically inflates prediction performance, whereas other forms of leakage have more minor effects.
- Matthew Rosenblatt
- , Link Tejavibulya
- & Dustin Scheinost
-
Article
| Open AccessTransfer learning with graph neural networks for improved molecular property prediction in the multi-fidelity setting
Modern molecular discovery processes generate millions of measurements at different quality levels. Here, the authors develop a new deep learning method for transfer learning from low-cost and abundant data to enhance the efficiency of drug discovery.
- David Buterez
- , Jon Paul Janet
- & Pietro Lió
-
Article
| Open AccessLarge language models streamline automated machine learning for clinical studies
A knowledge gap persists between machine learning developers and clinicians. Here, the authors show that the Advanced Data Analysis extension of ChatGPT could bridge this gap and simplify complex data analyses, making them more accessible to clinicians.
- Soroosh Tayebi Arasteh
- , Tianyu Han
- & Sven Nebelung
-
Perspective
| Open AccessDiffractive optical computing in free space
Optical computing via free-space-based structured optical materials allows to access optical information without the need for preprocessing or optoelectronic conversion. In this Perspective, the authors describe opportunities and challenges in their use for optical computing, information processing, computational imaging and sensing.
- Jingtian Hu
- , Deniz Mengu
- & Aydogan Ozcan
-
Comment
| Open AccessFudging the volcano-plot without dredging the data
Selecting omic biomarkers using both their effect size and their differential status significance (i.e., selecting the “volcano-plot outer spray”) has long been equally biologically relevant and statistically troublesome. However, recent proposals are paving the way to resolving this dilemma.
- Thomas Burger
-
Article
| Open AccessStructured information extraction from scientific text with large language models
Extracting scientific data from published research is a complex task required specialised tools. Here the authors present a scheme based on large language models to automatise the retrieval of information from text in a flexible and accessible manner.
- John Dagdelen
- , Alexander Dunn
- & Anubhav Jain
-
Article
| Open AccessSequential stacking link prediction algorithms for temporal networks
Link prediction in temporal networks is relevant for many real-world systems, however, current approaches are usually characterized by high computational costs. The authors propose a temporal link prediction framework based on the sequential stacking of static network features, for improved computational speed, appropriate for temporal networks with completely unobserved or partially observed target layers.
- Xie He
- , Amir Ghasemian
- & Peter J. Mucha
-
Article
| Open Accessβ-Variational autoencoders and transformers for reduced-order modelling of fluid flows
Reduced-order models provide better understanding for complex spatio-temporal dynamics of fluid flows with high numbers of degrees of freedom and non-linear interactions. The authors propose a variational autoencoder and transformer framework for learning the temporal dynamics of the nonlinear reduced-order models relevant for fluid dynamics, weather forecasting, and biomedical engineering.
- Alberto Solera-Rico
- , Carlos Sanmiguel Vila
- & Ricardo Vinuesa
-
Article
| Open AccessNeural network enabled nanoplasmonic hydrogen sensors with 100 ppm limit of detection in humid air
Detecting hydrogen gas in humid air is an unresolved challenge of significant importance for the safe implementation of hydrogen (energy) technologies. Here, authors demonstrate how the use of neural networks enables the sensing of hydrogen in highly humid air with a detection limit of 100 ppm.
- David Tomeček
- , Henrik Klein Moberg
- & Christoph Langhammer
-
Article
| Open AccessData-centric artificial olfactory system based on the eigengraph
Sensitivity-dependent data analysis methods disrupted the development of artificial olfactory technologies. Here, authors present a data-centric artificial olfactory system based on eigengraph that reflects the intrinsic electrochemical interaction.
- Seung-Hyun Sung
- , Jun Min Suh
- & Seong Chan Jun
-
Article
| Open AccessCapacitive tendency concept alongside supervised machine-learning toward classifying electrochemical behavior of battery and pseudocapacitor materials
Analysis of capacitive behavior of electrode materials used in batteries and pseudocapacitors is challenging. Here, authors report an electrochemical signal analysis method available as an online tool to classify the charge storage behavior of a material as battery-like or a pseudocapacitor-like.
- Siraprapha Deebansok
- , Jie Deng
- & Olivier Fontaine
-
Article
| Open AccessHigh-efficiency reinforcement learning with hybrid architecture photonic integrated circuit
The Authors present a universal framework that utilizes photonic integrated circuits (PIC) to enhance the efficiency of reinforcement learning (RL). Leveraging the advantages of the hybrid architecture PIC (HyArch PIC), the PIC-RL experiment demonstrates a remarkable 56% improvement in efficiency for synthesizing perovskite materials.
- Xuan-Kun Li
- , Jian-Xu Ma
- & Xian-Min Jin
-
Article
| Open AccessAnticipating regime shifts by mixing early warning signals from different nodes
Early warning signals for rapid regime shifts in complex networks are of importance for ecology, climate and epidemics, where heterogeneities in network nodes and connectivity make construction of early warning signals challenging. The authors propose a method for selecting an optimal set of nodes from which a reliable early warning signal can be obtained.
- Naoki Masuda
- , Kazuyuki Aihara
- & Neil G. MacLaren
-
Article
| Open AccessImproved machine learning algorithm for predicting ground state properties
Recent work proposed a machine learning algorithm for predicting ground state properties of quantum many-body systems that outperforms any non-learning classical algorithm but requires extensive training data. Lewis et al. present an improved algorithm with exponentially reduced training data requirements.
- Laura Lewis
- , Hsin-Yuan Huang
- & John Preskill
-
Article
| Open AccessAdaptive tactile interaction transfer via digitally embroidered smart gloves
Adaptive tactile interactions transfer across users, space, and time, via embroidered smart gloves is reported by the authors. The scalable fabrication and adaptive computation pipeline enable tactile occlusion alleviation, human skills transfer, and interactive teleoperation.
- Yiyue Luo
- , Chao Liu
- & Wojciech Matusik
-
Article
| Open AccessAn actor-model framework for visual sensory encoding
Encoding and downsampling images is key for visual prostheses. Here, the authors show that an actor-model framework using the inherent computation of the retinal network yields better performance in downsampling images compared to learning-free methods.
- Franklin Leong
- , Babak Rahmani
- & Diego Ghezzi
-
Article
| Open AccessRobustly federated learning model for identifying high-risk patients with postoperative gastric cancer recurrence
Medical data faces isolation and cross-center performance issues. Here, the authors propose a robust federated learning model to identify high-risk postoperative gastric cancer recurrence, achieving promising results across data from four independent medical institutions.
- Bao Feng
- , Jiangfeng Shi
- & Wansheng Long
-
Article
| Open AccessA dynamic knowledge graph approach to distributed self-driving laboratories
Global challenges demand global solutions. Here, the authors show a distributed self-driving lab architecture in The World Avatar, linking robots in Cambridge and Singapore for asynchronous multi-objective reaction optimisation.
- Jiaru Bai
- , Sebastian Mosbach
- & Markus Kraft
-
Article
| Open AccessSegment anything in medical images
Segmentation is an important fundamental task in medical image analysis. Here the authors show a deep learning model for efficient and accurate segmentation across a wide range of medical image modalities and anatomies.
- Jun Ma
- , Yuting He
- & Bo Wang
-
Article
| Open AccessMesoscale simulation of biomembranes with FreeDTS
In this work, the authors report the FreeDTS software to simulate biomembranes at the mesoscale. The software provides various membrane simulations, focusing on protein organization and shape remodeling. A versatile tool propelling realistic membrane studies and diverse applications.
- Weria Pezeshkian
- & John H. Ipsen
-
Article
| Open AccessEncoding of multi-modal emotional information via personalized skin-integrated wireless facial interface
Technologies in human emotion recognition are challenged by their capability to accurately extract and exploit the emotional information. Lee et al. report a personalized skin-integrated facial interface to sense and combine facial and vocal expression data, enabling enhanced communication in virtual reality.
- Jin Pyo Lee
- , Hanhyeok Jang
- & Jiyun Kim
-
Article
| Open AccessPersonalising intravenous to oral antibiotic switch decision making through fair interpretable machine learning
The decision to switch patients from intravenous to oral antibiotic therapy is important for the individual and wider society. Here, authors show a machine learning model using routine clinical data can predict when a patient could switch.
- William J. Bolton
- , Richard Wilson
- & Timothy M. Rawson
-
Article
| Open AccessCryo-EM structure and B-factor refinement with ensemble representation
Cryo-EM is the go-to method for visualizing large, flexible biomolecules. Here, authors introduce a new Gaussian mixture modelling method for cryo-EM modelling tasks, including refinement, composite map generation and ensemble representation.
- Joseph G. Beton
- , Thomas Mulvaney
- & Maya Topf
-
Article
| Open AccessTowards provably efficient quantum algorithms for large-scale machine-learning models
It is still unclear whether and how quantum computing might prove useful in solving known large-scale classical machine learning problems. Here, the authors show that variants of known quantum algorithms for solving differential equations can provide an advantage in solving some instances of stochastic gradient descent dynamics.
- Junyu Liu
- , Minzhao Liu
- & Liang Jiang
-
Article
| Open AccessSelective knowledge sharing for privacy-preserving federated distillation without a good teacher
While federated learning is promising for efficient collaborative learning without revealing local data, it remains vulnerable to white-box privacy attacks, suffers from high communication overhead, and struggles to adapt to heterogeneous models. Here, the authors show a federated distillation method to tackle these challenges, which leverages the strengths of knowledge distillation in a federated learning setting.
- Jiawei Shao
- , Fangzhao Wu
- & Jun Zhang
-
Article
| Open AccessGene-SGAN: discovering disease subtypes with imaging and genetic signatures via multi-view weakly-supervised deep clustering
Many diseases can display distinct brain imaging phenotypes across individuals, potentially reflecting disease subtypes. However, biological interpretability is limited if the derived subtypes are not associated with genetic drivers or susceptibility factors. Here, the authors describe a deep-learning method that links imaging phenotypes with genetic factors, thereby conferring genetic correlations to the disease subtypes.
- Zhijian Yang
- , Junhao Wen
- & Christos Davatzikos
-
Article
| Open AccessImproving deep neural network generalization and robustness to background bias via layer-wise relevance propagation optimization
Image background features can undesirably affect deep networks’ decisions. Here, the authors show that the optimization of Layer-wise Relevance Propagation explanation heatmaps can hinder such influence, improving out-of-distribution generalization.
- Pedro R. A. S. Bassi
- , Sergio S. J. Dertkigil
- & Andrea Cavalli
-
Article
| Open AccessTemporal dendritic heterogeneity incorporated with spiking neural networks for learning multi-timescale dynamics
Brain-inspired spiking neural networks have shown their capability for effective learning, however current models may not consider realistic heterogeneities present in the brain. The authors propose a neuron model with temporal dendritic heterogeneity for improved neuromorphic computing applications.
- Hanle Zheng
- , Zhong Zheng
- & Lei Deng
-
Article
| Open AccessLearning low-rank latent mesoscale structures in networks
Network structures can be examined at different scales, and subnetworks in the form of motifs can provide insights into global network properties. The authors propose an approach to decompose a network into a set of latent motifs, which can be used for network comparison, network denoising, and edge inference.
- Hanbaek Lyu
- , Yacoub H. Kureh
- & Mason A. Porter