News & Views in 2019

Photonic computing devices are a compelling alternative to conventional computing setups for machine learning applications, as they are nonlinear, fast and easy to parallelize. Recent work demonstrates the potential of these optical systems to process and classify human motion from video.

Origami engineering has long held the promise of complex and futuristic machines. A new foldable haptics system shows that this paradigm can be functional as well.

Loss-of-function mutations in metal-binding proteins are heavily implicated with numerous diseases, and identifying such ‘cracks’ will be valuable to biologists and medical doctors in the study and treatment of disease. A deep learning approach has been developed to tackle this challenging task.

In cooperative games, humans are biased against AI systems even when such systems behave better than our human counterparts. This raises a question: should AI systems ever be allowed to conceal their true nature and lie to us for our own benefit?

Adversarial attacks make imperceptible changes to a neural network’s inputs so that it recognizes it as something entirely different. This flaw can give us insight into how these networks work and how to make them more robust.

DeepMind’s AlphaFold recently demonstrated the potential of deep learning for protein structure prediction. DeepFragLib, a new protein-specific fragment library built using deep neural networks, may have advanced the field to the next stage.

To prepare robots for working autonomously under real-world conditions, their resilience and capability to recover from damage needs to improve radically. A fresh take on robot design suggests that instead of adapting the robotic control strategy, we could enable robots to change their physical bodies to recover more effectively from damage.

Classic theories of reinforcement learning and neuromodulation rely on reward prediction errors. A new machine learning technique relies on neuromodulatory signals that are optimized for specific tasks, which may lead to better AI and better explanations of neuroscience data.

Humans infer much of the intentions of others by just looking at their gaze. Similarly, we want to understand how machine learning systems solve a problem. New tools are developed to find out what strategies a learning machine is using, such as what it is paying attention to when classifying images.

Memristor-based chips could lead the way to fast, energy-efficient AI hardware accelerators.

To be useful in a variety of daily tasks, robots must be able to interact physically with humans and infer how to be most helpful. A new theory for interactive robot control allows a robot to learn when to assist or challenge a human during reaching movements.