This paper proposes K-NRM, a kernel based neural model for document ranking. Given a query and a set of documents, K-NRM uses a translation matrix that models word-level similarities via word embeddings, a new kernel-pooling technique that uses kernels to extract multi-level soft match features, and a learning-to-rank layer that combines those features into the final ranking score. The whole model is trained end-to-end. The ranking layer learns desired feature patterns from the pairwise ranking loss. The kernels transfer the feature patterns into soft-match targets at each similarity level and enforce them on the translation matrix. The word embeddings are tuned accordingly so that they can produce the desired soft matches. Experiments on a commercial search engine's query log demonstrate the improvements of K-NRM over prior feature-based and neural-based states-of-the-art, and explain the source of K-NRM's advantage: Its kernel-guided embedding encodes a similarity metric tailored for matching query words to document words, and provides effective multi-level soft matches

No abstract

Nanoparticles assembled from poly(D,L-lactic acid)-poly(ethylene glycol) (PLA-PEG) block copolymers may have a therapeutic application in site-specific drug delivery. A series of AB block copolymers based on a fixed PEG block (5 kDa) and a varying PLA segment (2-110 kDa) have been synthesized by the ring-opening polymerization of D,L-lactide using stannous octoate as a catalyst. These copolymers assembled to form spherical nanoparticles in aqueous media following precipitation from a water-miscible organic solvent. 1 H NMR studies of the PLA-PEG nanoparticles in D2O confirmed their core-shell structure, with negligible penetration of the hydrated PEG chains into the PLA core. The influence of the PLA block molecular weight on the hydrodynamic size and micellar aggregation number of the assemblies was determined by dynamic and static light scattering techniques. The hydrodynamic radius of the PLA-PEG 2:5-30:5 nanoparticles was solely dependent on the copolymer architecture and scaled linearly as NPLA 1/3 , where NPLA is the number of monomeric units in the PLA block. The PEG chains of the small PLA-PEG 2:5 and 3:5 assemblies appeared to be fairly splayed as a consequence of their relatively low aggregation number and high surface coverage. However, as NPLA was increased to 6 kDa the area available per PEG chain at the periphery of the shell decreased significantly and then remained fairly constant with further increases in the molecular weight of the PLA block. The aggregation number and hence particle size of nanoparticles produced from copolymers with a PLA block of 45 kDa or more was found to also depend on the concentration of copolymer dissolved in the organic phase during preparation. This suggested that that the PEG chains had little influence on the assembly of the higher molecular weight copolymers.

No abstract

Label and label-free methods to image carbon-based nanomaterials exist. However, label-based approaches are limited by the risk of tag detachment over time, and label-free spectroscopic methods have slow imaging speeds, weak photoluminescence signals and strong backgrounds. Here, we present a label-free mass spectrometry imaging method to detect carbon nanotubes, graphene oxide and carbon nanodots in mice. The large molecular weights of nanoparticles are difficult to detect using conventional mass spectrometers, but our method overcomes this problem by using the intrinsic carbon cluster fingerprint signal of the nanomaterials. We mapped and quantified the sub-organ distribution of the nanomaterials in mice. Our results showed that most carbon nanotubes and nanodots were found in the outer parenchyma of the kidney, and all three materials were seen in the red pulp of the spleen. The highest concentrations of nanotubes in the spleen were found within the marginal zone.