There are two principle approaches for learning in artificial intelligence: error-driven global learning and neuroscience-oriented local learning. Integrating them into one network may provide complementary learning capabilities for versatile learning scenarios. At the same time, neuromorphic computing holds great promise, but still needs plenty of useful algorithms and algorithm-hardware co-designs to fully exploit its advantages. Here, we present a neuromorphic global-local synergic learning model by introducing a brain-inspired meta-learning paradigm and a differentiable spiking model incorporating neuronal dynamics and synaptic plasticity. It can meta-learn local plasticity and receive top-down supervision information for multiscale learning. We demonstrate the advantages of this model in multiple different tasks, including few-shot learning, continual learning, and fault-tolerance learning in neuromorphic vision sensors. It achieves significantly higher performance than single-learning methods. We further implement the model in the Tianjic neuromorphic platform by exploiting algorithm-hardware co-designs and prove that the model can fully utilize neuromorphic many-core architecture to develop hybrid computation paradigm.

SUMMARYAn overlaid domain decomposition method, called the bridging domain method, for the coupled simulation of molecular/continuum systems is analyzed. In this method, compatibility between the atomistic and the continuum subdomains is enforced by Lagrange multipliers. Two forms are considered: (1) a consistent constraint form, which employs the exact non-diagonal Lagrange multiplier equations and thus enforces compatibility exactly, but is computationally demanding and (2) a diagonalized constraint form, which is computationally more efficient. It is shown that the consistent constraint form conserves linear momentum, angular momentum and energy. The diagonalized constraint form dissipates energy. The diagonalized form is shown to be more effective in suppressing spurious reflections at the interface.

A new technique is presented to study fracture in nanomaterials by coupling quantum mechanics (QM) and continuum mechanics (CM). A key new feature of this method is that broken bonds are identified by a sharp decrease in electron density at the bond midpoint in the QM model. As fracture occurs, the crack tip position and crack path are updated from the broken bonds in the QM model. At each step in the simulation, the QM model is centered on the crack tip to adaptively follow the path. This adaptivity makes it possible to trace paths with complicated geometries. The method is applied to study the propagation of cracks in graphene which are initially perpendicular to zigzag and armchair edges. The simulations demonstrate that the growth of zigzag cracks is self-similar whereas armchair cracks advance in an irregular manner. The critical stress intensity factors for graphene were found to be 4.21 MPa √ m for zigzag cracks and 3.71 MPa √ m for armchair cracks, which is about 10% of that for steel.

A novel hydrogel based on a crosslinked chitosan/gelatin with glutaraldehyde hybrid polymer network was prepared. The gel swells in acidic medium and deswells in alkaline medium. A comparative study of the pH‐dependent release of levamisole, cimetidine and chloramphenicol was carried out.

A nanostructured Cr2O3/WO3 p–n junction photoanode has been prepared for achieving high efficiency in water oxidation.