Search for: All records

External knowledge is often useful for natural language understanding tasks. We introduce a contextual text representation model called Conceptual-Contextual (CC) embeddings, which incorporates structured knowledge into text representations. Unlike entity embedding methods, our approach encodes a knowledge graph into a context model. CC embeddings can be easily reused for a wide range of tasks in a similar fashion to pre-trained language models. Our model effectively encodes the huge UMLS database by leveraging semantic generalizability. Experiments on electronic health records (EHRs) and medical text processing benchmarks showed our model gives a major boost to the performance of supervised medical NLP tasks. 

Unsupervised word embeddings have become a popular approach of word representation in NLP tasks. However there are limitations to the semantics represented by unsupervised embeddings, and inadequate fine-tuning of embeddings can lead to suboptimal performance. We propose a novel learning technique called Delta Embedding Learning, which can be applied to general NLP tasks to improve performance by optimized tuning of the word embeddings. A structured regularization is applied to the embeddings to ensure they are tuned in an incremental way. As a result, the tuned word embeddings become better word representations by absorbing semantic information from supervision without “forgetting.” We apply the method to various NLP tasks and see a consistent improvement in performance. Evaluation also confirms the tuned word embeddings have better semantic properties. 

Silicon as a promising candidate for the next-generation high-capacity lithium-ion battery anode is characterized by outstanding capacity, high abundance, low operational voltage, and environmental benignity. However, large volume changes during Si lithiation and de-lithiation can seriously impair its long-term cyclability. Although extensive research efforts have been made to improve the electrochemical performance of Si-based anodes, there is a lack of efficient fabrication methods that are low cost, scalable, and self-assembled. In this report, co-axial fibrous silicon asymmetric membrane has been synthesized using a scalable and straightforward phase inversion method combined with dip coating as inspired by the hollow fiber membrane technology that has been successfully commercialized over the last decades to provide billions of gallons of purified drinking water worldwide. We demonstrate that ~ 90% initial capacity of co-axial fibrous Si asymmetric membrane electrode can be maintained after 300 cycles applying a current density of 400 mA g−1. The diameter of fibers, size of silicon particles, type of polymers, and exterior coating have been identified as critical factors that can influence the electrode stability, initial capacity, and rate performance. Much enhanced electrochemical performance can be harvested from a sample that has thinner fiber diameter, smaller silicon particle, lower silicon content, and porous carbon coating. This efficient and scalable approach to prepare high-capacity silicon-based anode with outstanding cyclability is fully compatible with industrial roll-to-roll processing technology, thus bearing a great potential for its future commercialization. 

Germanium (Ge) is deemed as one of the most promising alloying anodes for rechargeable lithium‐ion batteries (LIBs) due to its large theoretical capacity, high electrical conductivity, fast lithium‐ion diffusivity, and mechanical robustness. However, Ge‐based anodes suffer from large volume changes during lithiation and delithiation, which can deteriorate their electrochemical performance rapidly. Herein, the large volume change issue is effectively addressed using an asymmetric membrane structure that is prepared using a phase‐inversion method in combination with hydrogen peroxide etching and surface coating. The asymmetric Ge membrane etched by ≈30 wt% H₂O₂at 90 °C for 30 s demonstrates a capacity retention higher than 80% in 50 cycles at 160 mA g⁻¹. Coating the H₂O₂‐etched Ge membrane with carbonaceous membranes can further improve the retention up to 95% in 50 cycles at 160 mA g⁻¹, whereas ≈100% capacity of 700 mAh g⁻¹can be maintained in 170 cycles at 400 mA g⁻¹. A combination of electron microscopy, spectrophotometry, and X‐ray analyses confirms the electrochemical performance of asymmetric Ge membranes as the LIB anode can be significantly affected by membrane geometry, the duration of hydrogen peroxide etching, carbonaceous membrane coating, and Ge concentration.