Articles

Photonic computing devices have been proposed as a high-speed and energy-efficient approach to implementing neural networks. Using off-the-shelf components, Antonik et al. demonstrate a reservoir computer that recognizes different forms of human action from video streams using photonic neural networks.

Deep learning is currently transforming digital pathology, helping to make more reliable and faster clinical diagnoses. A promising application is in the recognition of malignant white blood cells—an essential step for detecting acute myeloid leukaemia that is challenging even for trained human examiners. An annotated image dataset of over 18,000 white blood cells is compiled and used to train a convolutional neural network for leukocyte classification. The network classifies the most important cell types with high accuracy and can answer clinically relevant binary questions with human-level performance.

Deep neural networks can be led to misclassify an image when minute changes that are imperceptible to humans are introduced. While for some networks this ability can cast doubt on the reliability of the model, it also offers explainability for networks that use more robust regularization.

To keep radiation therapy from damaging healthy tissue, expert radiologists have to segment CT scans into individual organs. A new deep learning-based method for delineating organs in the area of head and neck performs faster and more accurately than human experts.

Neural network force fields promise to bypass the computationally expensive quantum mechanical calculations typically required to investigate complex materials, such as lithium-ion batteries. Mailoa et al. accelerate these approaches with an architecture that exploits both rotation-invariant and -covariant features separately.

Optoacoustic imaging can achieve high spatial and temporal resolution but image quality is often compromised by suboptimal data acquisition. A new method employing deep learning to recover high-quality images from sparse or limited-view optoacoustic scans has been developed and demonstrated for whole-body mouse imaging in vivo.

Labelling training data to train machine learning models is very time intense. A new method shows that content transformation can be effectively learned from generated data, avoiding the need for any manual labelling in segmentation and classification tasks.

Brain–machine interfaces using steady-state visually evoked potentials (SSVEPs) show promise in therapeutic applications. With a combination of innovations in flexible and soft electronics and in deep learning approaches to classify potentials from two channels and from any subject, a compact, wireless and universal SSVEP interface is designed. Subjects can operate a wheelchair in real time with eye movements while wearing the new brain–machine interface.

A combination of engineering advances shows promise for myoelectric prosthetic hands that are controlled by a user’s remaining muscle activity. Fine finger movements are decoded from surface electromyograms with machine learning algorithms and this is combined with a robotic controller that is active only during object grasping to assist in maximizing contact. This shared control scheme allows user-controlled movements when high dexterity is desired, but also assisted grasping when robustness is required.

Memristive devices can provide energy-efficient neural network implementations, but they must be tailored to suit different network architectures. Wang et al. develop a trainable weight-sharing mechanism for memristor-based CNNs and ConvLSTMs, achieving a 75% reduction in weights without compromising accuracy.

Controlling the flow and representation of information in deep neural networks is fundamental to making networks intelligible. Bergomi et al introduce a mathematical framework in which the space of possible operators representing the data is constrained by using symmetries. This constrained space is still suitable for machine learning: operators can be efficiently computed, approximated and parameterized for optimization.

An approach to protein structure prediction is to assemble candidate structures from template fragments, which are extracted from known protein structures. Wang et al. demonstrate that combining deep neural network architectures with a relatively small but high-resolution fragment dataset can improve the quality of the sample fragment libraries used for protein structure prediction.

When neural networks are retrained to solve more than one problem, they tend to forget what they have learned earlier. Here, the authors propose orthogonal weights modification, a method to avoid this so-called catastrophic forgetting problem. Capitalizing on such an ability, a new module is introduced to enable the network to continually learn context-dependent processing.

Deep neural networks can contain arbitrary mathematical operators, as long as they are derivable. The authors investigate how knowledge about a problem can be incorporated into machine learning through the use of operators that are related to the problem.

For some combinatorial puzzles, solutions can be verified to be optimal, for others, the state space is too large to be certain that a solution is optimal. A new deep learning based search heuristic performs well on the iconic Rubik’s cube and can also generalize to puzzles in which optimal solvers are intractable.

Classifying individual responses in open-ended surveys can be subjective and time-consuming. A network-based survey framework automatically classifies responses in a statistically principled manner.

Neural networks are a promising digital pathology tool but are often criticized for their limited explainability. Faust and others demonstrate how machine-learned features correlate with human-understandable histological patterns and groupings, permitting increased transparency of deep learning tools in medicine.

Artificial intelligence approaches can aid medicinal chemists to creatively look for new chemical entities with drug-like properties. A rule-based approach combined with a machine learning model was trained on successful synthetic routes described in chemical patent literature. This process produced computer-generated compounds that mimic known medicines.

Reducing the radiation dose for medical CT scans can provide a less invasive imaging method, but requires a method for reconstructing an image up to the image quality from a full-dose scan. In this article, Wang and colleagues show that the deep learning approach, combined with the feedback from radiologists, produces higher quality reconstructions than or similar to that using the current commercial methods.

High-throughput brain image registration methods that are independent of any pre-processing steps, and are robust under mild image transformations, could accelerate the study of region-specific changes in brain development. A deep learning-based method is therefore developed for automated registration through segmenting brain regions of interest with minimal human supervision.