Computer science

  • Article
    | Open Access

    Ultrasound is an important imaging modality for the detection and characterization of breast cancer, but it has been noted to have high false-positive rates. Here, the authors present an artificial intelligence system that achieves radiologist-level accuracy in identifying breast cancer in ultrasound imaging.

    • Yiqiu Shen
    • , Farah E. Shamout
    •  & Krzysztof J. Geras
  • Article
    | Open Access

    Some regions on the Moon are permanently covered in shadow and are therefore extremely difficult to see into. We develop a deep learning driven algorithm which enhances images of these regions, allowing us to see inside them with high resolution for the first time.

    • V. T. Bickel
    • , B. Moseley
    •  & M. Shirley
  • Article
    | Open Access

    Finding a biologically-relevant inductive bias for training DNNs on large fitness landscapes is challenging. Here, the authors propose a method called Epistatic Net that improves DNN prediction accuracy and interpretation speed by integrating the knowledge that higher-order epistatic interactions are usually sparse.

    • Amirali Aghazadeh
    • , Hunter Nisonoff
    •  & Kannan Ramchandran
  • Article
    | Open Access

    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

    Accurate seasonal forecasts of sea ice are highly valuable, particularly in the context of sea ice loss due to global warming. A new machine learning tool for sea ice forecasting offers a substantial increase in accuracy over current physics-based dynamical model predictions.

    • Tom R. Andersson
    • , J. Scott Hosking
    •  & Emily Shuckburgh
  • Article
    | Open Access

    Wavefront shaping is used to overcome scattering in biological tissues during imaging, but determining the compensation is slow. Here, the authors use holographic phase stepping interferometry, where new phase information is updated after each measurement, enabling fast improvement of the wavefront correction.

    • Molly A. May
    • , Nicolas Barré
    •  & Alexander Jesacher
  • Article
    | Open Access

    In many machine learning applications, one uses pre-trained neural networks, having limited access to training and test data. Martin et al. show how to predict trends in the quality of such neural networks without access to this information, relevant for reproducibility, diagnostics, and validation.

    • Charles H. Martin
    • , Tongsu (Serena) Peng
    •  & Michael W. Mahoney
  • Article
    | Open Access

    Physical principles underlying machine learning analysis of quantum gas microscopy data are not well understood. Here the authors develop a neural network based approach to classify image data in terms of multi-site correlation functions and reveal the role of fourth-order correlations in the Fermi-Hubbard model.

    • Cole Miles
    • , Annabelle Bohrdt
    •  & Eun-Ah Kim
  • Article
    | Open Access

    Network embedding is a machine learning technique for construction of low-dimensional representations of large networks. Gu et al. propose a method for the identification of an optimal embedding dimension for the encoding of network structural information inspired by natural language processing.

    • Weiwei Gu
    • , Aditya Tandon
    •  & Filippo Radicchi
  • Article
    | Open Access

    Rapid, accurate and specific point-of-care diagnostics can help manage and contain fast-spreading infections. Here, the authors present a nanopore-based system that uses artificial intelligence to discriminate between four coronaviruses in saliva, with little need for sample pre-processing.

    • Masateru Taniguchi
    • , Shohei Minami
    •  & Kazunori Tomono
  • Article
    | Open Access

    Expectations for quantum machine learning are high, but there is currently a lack of rigorous results on which scenarios would actually exhibit a quantum advantage. Here, the authors show how to tell, for a given dataset, whether a quantum model would give any prediction advantage over a classical one.

    • Hsin-Yuan Huang
    • , Michael Broughton
    •  & Jarrod R. McClean
  • Article
    | Open Access

    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

    The implementation of memory-augmented neural networks using conventional computer architectures is challenging due to a large number of read and write operations. Here, Karunaratne, Schmuck et al. propose an architecture that enables analog in-memory computing on high-dimensional vectors at accuracy matching 32-bit software equivalent.

    • Geethan Karunaratne
    • , Manuel Schmuck
    •  & Abbas Rahimi
  • Article
    | Open Access

    Multiphoton microscopy requires precise increases in excitation power with imaging depth to generate contrast without damaging the sample. Here the authors show how an adaptive illumination function can be learned from the sample’s shape and used for in vivo imaging of whole lymph nodes.

    • Henry Pinkard
    • , Hratch Baghdassarian
    •  & Laura Waller
  • Article
    | Open Access

    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

    Most demonstrations of quantum advantages with optics rely on single photons, and are thus difficult to scale up. Here, the authors use coherent states to demonstrate a quantum advantage for the task of verifying the solution to a NP-complete problem when only partial information on the solution is available.

    • Federico Centrone
    • , Niraj Kumar
    •  & Iordanis Kerenidis
  • Article
    | Open Access

    Coronary artery calcium is an accurate predictor of cardiovascular events but this information is not routinely quantified. Here the authors show a robust and time-efficient deep learning system to automatically quantify coronary calcium on CT scans and predict cardiovascular events in a large, multicentre study.

    • Roman Zeleznik
    • , Borek Foldyna
    •  & Hugo J. W. L. Aerts
  • Article
    | Open Access

    While Digital contact tracing (DCT) has been argued to be a valuable complement to manual tracing in the containment of COVID-19, no empirical evidence of its effectiveness is available to date. Here, the authors report the results of a 4-week population-based controlled experiment, where they assessed the impact of the Spanish DCT app.

    • Pablo Rodríguez
    • , Santiago Graña
    •  & Lucas Lacasa
  • Article
    | Open Access

    The advantages coming from involving quantum systems in machine learning are still not fully clear. Here, the authors propose a software/hardware co-design framework towards quantum-friendly neural networks showing quantum advantage, representing data as either random variables or numbers in unitary matrices.

    • Weiwen Jiang
    • , Jinjun Xiong
    •  & Yiyu Shi
  • Article
    | Open Access

    Digital trace data from search engines lacks information about the experiences of the individuals generating the data. Here the authors link search data and human computation to build a tracking model of influenza-like illness.

    • Stefan Wojcik
    • , Avleen S. Bijral
    •  & David Lazer
  • Article
    | Open Access

    The presence of confounding effects is one of the most critical challenges in using deep learning to advance discovery in medical imaging studies. Here, the authors introduce an end-to-end approach for deriving features invariant to confounding factors as inputs to prediction models.

    • Qingyu Zhao
    • , Ehsan Adeli
    •  & Kilian M. Pohl
  • Article
    | Open Access

    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

    Multiplayer games can be used as testbeds for the development of learning algorithms for artificial intelligence. Omidshafiei et al. show how to characterize and compare such games using a graph-based approach, generating new games that could potentially be interesting for training in a curriculum.

    • Shayegan Omidshafiei
    • , Karl Tuyls
    •  & Rémi Munos
  • Article
    | Open Access

    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

    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

    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

    Designing reliable and energy-efficient memristor-based artificial neural networks remains a challenge. Here, the authors demonstrate a technology-agnostic approach, committee machines, which increases the inference accuracy of memristive neural networks that suffer from device variability, faulty devices, random telegraph noise and line resistance.

    • D. Joksas
    • , P. Freitas
    •  & A. Mehonic
  • Article
    | Open Access

    Boolean Networks are a well-established model of biological networks, but usual interpretations can preclude the prediction of behaviours observed in quantitative systems. The authors introduce Most Permissive Boolean Networks, which are shown not to miss any behaviour achievable by the corresponding quantitative model.

    • Loïc Paulevé
    • , Juraj Kolčák
    •  & Stefan Haar
  • Article
    | Open Access

    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

    Extracting central information from ever-growing data generated in our lives calls for new data mining methods. Ferreira et al. show a simple model, called chronnets, that can capture frequent patterns, spatial changes, outliers, and spatiotemporal clusters.

    • Leonardo N. Ferreira
    • , Didier A. Vega-Oliveros
    •  & Elbert E. N. Macau
  • Article
    | Open Access

    It is generally difficult to scale derived estimates and understand the accuracy across locations for passively-collected data sources, such as mobile phones and satellite imagery. Here the authors show that their trained deep learning models are able to explain 70% of the variation in ground-measured village wealth in held-out countries, outperforming previous benchmarks from high-resolution imagery with errors comparable to that of existing ground data.

    • Christopher Yeh
    • , Anthony Perez
    •  & Marshall Burke
  • Article
    | Open Access

    Designing deep learning inference hardware based on in-memory computing remains a challenge. Here, the authors propose a strategy to train ResNet-type convolutional neural networks which results in reduced accuracy loss when transferring weights to in-memory computing hardware based on phase-change memory.

    • Vinay Joshi
    • , Manuel Le Gallo
    •  & Evangelos Eleftheriou
  • Article
    | Open Access

    Multiple access channels model communication from multiple independent senders to a common receiver. By drawing a connection to the study of classical and quantum correlations using nonlocal games, Leditzky et al. reveal remarkably complex behaviour of the entanglement-assisted and unassisted information transmission capabilities of a multiple access channel.

    • Felix Leditzky
    • , Mohammad A. Alhejji
    •  & Graeme Smith
  • Article
    | Open Access

    Unconventional computing architectures might outperform current ones, but their realization has been limited to solving simple specific problems. Here, a network of interconnected Belousov-Zhabotinski reactions, operated by independent magnetic stirrers, performs encoding/decoding operations and data storage.

    • Juan Manuel Parrilla-Gutierrez
    • , Abhishek Sharma
    •  & Leroy Cronin
  • Article
    | Open Access

    Understanding the underlying mechanisms behind the successes of deep networks remains a challenge. Here, the author demonstrates an implicit regularization in training deep networks, showing that the control of complexity in the training is hidden within the optimization technique of gradient descent.

    • Tomaso Poggio
    • , Qianli Liao
    •  & Andrzej Banburski
  • Article
    | Open Access

    The physical limits and reliability of PCR-based random access of DNA encoded data is unknown. Here the authors demonstrate reliable file recovery from as few as ten copies per sequence, providing a data density limit of 17 exabytes per gram.

    • Lee Organick
    • , Yuan-Jyue Chen
    •  & Luis Ceze
  • Article
    | Open Access

    Technologies for acquiring explainable features from medical images need further development. Here, the authors report a deep learning based automated acquisition of explainable features from pathology images, and show a higher accuracy of their method as compared to pathologist based diagnosis of prostate cancer recurrence.

    • Yoichiro Yamamoto
    • , Toyonori Tsuzuki
    •  & Go Kimura
  • Perspective
    | Open Access

    Recent research in motor neuroscience has focused on optimal feedback control of single, simple tasks while robotics and AI are making progress towards flexible movement control in complex environments employing hierarchical control strategies. Here, the authors argue for a return to hierarchical models of motor control in neuroscience.

    • Josh Merel
    • , Matthew Botvinick
    •  & Greg Wayne
  • Article
    | Open Access

    Is there an optimum difficulty level for training? In this paper, the authors show that for the widely-used class of stochastic gradient-descent based learning algorithms, learning is fastest when the accuracy during training is 85%.

    • Robert C. Wilson
    • , Amitai Shenhav
    •  & Jonathan D. Cohen
  • Article
    | Open Access

    Neural signalling is directional, but non-invasive neuroimaging methods are unable to map directed connections between brain regions. Here, the authors show how network communication measures can be used to infer signalling directionality from the undirected topology of brain structural connectomes.

    • Caio Seguin
    • , Adeel Razi
    •  & Andrew Zalesky
  • Perspective
    | Open Access

    Recent gains in artificial neural networks rely heavily on large amounts of training data. Here, the author suggests that for AI to learn from animal brains, it is important to consider that animal behaviour results from brain connectivity specified in the genome through evolution, and not due to unique learning algorithms.

    • Anthony M. Zador