Reviews & Analysis

Neural networks may be overconfident before they see real data. By briefly training on random noise, models can learn to be uncertain, leading to better calibration, improved identification of out-of-distribution inputs and thus more reliable predictions.

Combining soft robotics with neuromorphic engineering is a promising approach in embodied intelligence. Giulia d’Angelo et al. contribute to progress in this field by developing a framework for benchmarking neuromorphic controllers on soft robotic platforms.

Identifying cell–cell interactions from imaging-based spatial transcriptomics suffers from limited gene panels. A new self-supervised graph transformer-based method can resolve spatial single-cell-level interactions without requiring known ligand–receptor pairs.

Mixtures are ubiquitous in industrial formulations. A framework unifying predictive and generative machine learning now offers a blueprint for data-driven design of multi-component battery electrolytes.

Capturing the complexity of cardiovascular dynamics demands multiple monitoring modalities, each with inherent trade-offs. Diffusion-based modeling offers a promising route for synthesizing and generating cross-modal data.

Deep generative models that learn intermediate surface-current maps, rather than layouts directly, offer a more stable route to inverse design of tunable and stacked metasurfaces.

Brain-inspired computing can enhance the finite element method, a cornerstone of scientific modelling, by reducing energy costs and reframing numerical simulation through neural dynamics.

Reinforcement learning has a key role in artifical intelligence (AI), but its implementation on neuromorphic hardware typically involves operations executed on conventional digital computers. A study now addresses this issue by implementing an actor–critic network fully in hardware using analogue memristors.

Even before training, convolutional neural networks may reflect the brain’s visual processing principles. A study now shows how structure alone can help to explain the alignment between brains and models.

Durstewitz et al. explore what artificial intelligence can learn from the brain’s ability to adjust quickly to changing environments. By linking neuroscience studies of flexible behaviour with advances in continual and in-context learning, this Perspective outlines ways to strengthen the exchange of ideas between the two fields and advance NeuroAI.

A new benchmark, KaBLE (knowledge and belief language evaluation), indicates that some large language models are unable to accurately distinguish belief from knowledge and fact, calling into question their use in real-word applications such as medicine and law.

Molecular dynamics (MD) simulations are widely used for understanding atomic motion but require substantial computational time. In new research by Nam et al., a generative artificial intelligence framework is developed to accelerate the MD simulations for crystalline materials, by reframing the task as conditional generation of atomic displacement.

Multimodal AI combines different types of data to improve decision-making in fields such as healthcare and engineering, but work so far has focused on vision and language models. To make these systems more usable in the real world, Liu et al. discuss the need to develop approaches with deployment in mind from the start, working closely with experts across relevant disciplines.

Irie and Lake present a metalearning framework that enables artificial neural networks to address classic challenges by providing both incentives to improve specific capabilities and opportunities to practice them.

Although it is possible to use deep learning models to predict static protein conformations from sequencing data, proteins are not static biochemical artefacts. ItsFlexible is a graph-based deep learning tool that is trained on a new dataset of experimentally captured protein motif conformations to classify the dynamic characteristics of proteins.

Designing and optimizing proteins by mutagenesis suffers from the overwhelming space of possible variants. A recent study developed µProtein, a reinforcement learning model coupled with a protein language model as a surrogate oracle, to accelerate this process towards high-functioning proteins.

Ilievski et al. examine differences and similarities in the various ways human and AI systems generalize. The insights are important for effectively supporting alignment in human–AI teams.

The next major challenge for artificial intelligence in drug development lies in proving its value in real-world settings. A new technology not only supports the generation of novel chemical entities but also accelerates a range of real-world molecular design tasks.

Thoughtfully designed negative training datasets may hold the key to more robust machine learning models. Ursu et al. reveal how negative training data composition shapes antibody prediction models and their generalizability. Sometimes, the best way to get better is to train harder.

Despite impressive performances of current large AI models, symbolic and abstract reasoning tasks often elicit failure modes in these systems. In this Perspective, Ito et al. propose to make use of computational complexity theory, formulating algebraic problems as computable circuits to address the challenge of mathematical and symbolic reasoning in AI systems.