University of Utah engineers encode partial differential equations in light and feed them into newly designed optical neural engine, or ONE, to accelerate machine learning.

Artificial Intelligence (AI) impacts many industries and professions in different ways. A recent study looking at taxi drivers in Yokohama, Japan, shows that AI demand forecasting, which is unrelated to autonomous driving, can improve productivity in less-experienced drivers, helping close the skill gaps. This finding challenges the assumption that AI only favors high-skilled workers in technology-dependent industries.

A study in Science Advances that describes a new soft robotic system was co-led by Harvard Professor L. Mahadevan in collaboration with Professor Ho-Young Kim at Seoul National University. Their work paves new directions for future, low-power swarm robotics.

The new robots, called link-bots, are comprised of centimeter-scale, 3D-printed particles strung into V-shaped chains via notched links and are capable of coordinated, life-like movements without any embedded power or control systems. 

Researchers in the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) have developed a machine learning framework that can predict with quantum-level accuracy how materials respond to electric fields, up to the scale of a million atoms – vastly accelerating simulations beyond quantum mechanical methods, which can model only a few hundred atoms at a time.

Physicists at the University of Oxford have set a new global benchmark for the accuracy of controlling a single quantum bit, achieving the lowest-ever error rate for a quantum logic operation—just 0.000015%, or one error in 6.7 million operations. This record-breaking result represents nearly an order of magnitude improvement over the previous benchmark, set by the same research group a decade ago.

Biological cells exhibit nearly transparent characteristics with weak absorption properties in the visible light spectrum, resulting in extremely low optical contrast between cells and the surrounding medium under traditional bright-field microscopy. To enhance imaging contrast, conventional methods rely on chemical staining or fluorescent labeling, introducing exogenous absorption/fluorescence probes to visualize cellular structures. However, these approaches suffer from drawbacks such as phototoxicity, photobleaching, and poor biocompatibility, severely limiting long-term dynamic observation of living cells. Quantitative phase imaging (QPI) utilizes the inherent physical property of cellular phase (thickness) as an endogenous “probe”, resolving cellular thickness, refractive index, and 3D topography with nanoscale accuracy. It provides a new avenue for dynamic observation of living cells and nanoscale biological studies.

As a significant branch of QPI technology, differential phase contrast (DPC) has attracted considerable attention due to its advantages of being non-interferometric and low-cost. However, its theoretical framework relies on the “weak object approximation”, linking intensity images to sample phase through a linear model. This simplified model introduces two fundamental limitations. First, the phase reconstruction result is highly dependent on the precise modeling of the phase transfer function (PTF) under an ideal pupil. In practical optical systems, however, wavefront aberrations couple with the sample phase, leading to significant reconstruction errors. Second, the conventional half-circle illumination suffers from the problem of PTF response cancellation, resulting in the loss of low-frequency phase information and making it difficult to accurately reconstruct the fine structure of weak phase objects. These limitations significantly compromise the robustness of DPC in non-ideal optical environments and restrict its practical applicability in frontier biological research, such as cellular morphology characterization and tracking of subcellular dynamic processes.