Representational learning forms the backbone of most deep learning applications, and the value of a learned representation is intimately tied to its information content regarding different factors of variation. Finding good representations depends on
the nature of supervision and the learning algorithm. We propose a novel algorithm that relies on a weak form of supervision where the data is partitioned into sets according to certain inactive factors of variation. Our key insight is that by seeking approximate correspondence between elements of different sets, we learn strong representations that exclude the inactive factors of variation and isolate the active factors which vary within all sets. We demonstrate that the method can work in a semi-supervised scenario, and that a portion of the unsupervised data can belong to a different domain entirely. Further control over the content of the learned representations is possible by folding in data augmentation to suppress nuisance factors. We outperform competing baselines on the challenging problem of synthetic-to-real object pose transfer.
We demonstrate experimentally that a granular packing of glass spheres is capable of storing memory of multiple strain states in the dynamic process of stress relaxation. Modeling the system as a non-interacting population of relaxing elements, we fi
nd that the functional form of the predicted relaxation requires a quantitative correction which grows in severity with each additional memory and is suggestive of interactions between elements. Our findings have implications for the broad class of soft matter systems that display memory and anomalous relaxation.
Architectural structures such as masonry walls or columns exhibit a slender verticality, in contrast to the squat, sloped forms obtained with typical unconfined granular materials. Here we demonstrate the ability to create freestanding, weight-bearin
g, similarly slender and vertical structures by the simple pouring of suitably shaped dry particles into a mold that is subsequently removed. Combining experiments and simulations we explore a family of particle types that can entangle through their non-convex, hooked shape. We show that Z-shaped particles produce granular aggregates which can either be fluid and pourable, or solid and rigid enough to maintain vertical interfaces and build freestanding columns of large aspect ratio (>10) that support compressive loads without external confinement. We investigate the stability of such columns with uniaxial compression, bending, and vibration tests and compare with other particle types including U-shaped particles and rods. We find a pronounced anisotropy in the internal stress propagation together with strong strain-stiffening, which stabilizes rather than destabilizes the structures under load.
