We study knowledge-grounded dialogue generation with pre-trained language models. To leverage the redundant external knowledge under capacity constraint, we propose equipping response generation defined by a pretrained language model with a knowledge selection module, and an unsupervised approach to jointly optimizing knowledge selection and response generation with unlabeled dialogues. Empirical results on two benchmarks indicate that our model can significantly outperform state-of-the-art methods in both automatic evaluation and human judgment.

A lot of real-life mobile sensing applications are becoming available. These applications use mobile sensors embedded in smart phones to recognize human activities in order to get a better understanding of human behavior. In this paper, we propose a LSTM-based feature extraction approach to recognize human activities using tri-axial accelerometers data. The experimental results on the (WISDM) Lab public datasets indicate that our LSTM-based approach is practical and achieves 92.1% accuracy.

The electronic structures of cesium carbonate (Cs2CO3) doped 4,7-diphenyl-1,10-phenanthroline (BPhen) films with various doping concentration are characterized by in situ ultraviolet and x-ray photoelectron spectroscopies, in an attempt to understand the mechanism of electron-transport enhancement in Cs2CO3-doped organic electron-transport layer for organic optoelectronic devices. The n-type electrical doping effect is evidenced by the Fermi level shift in the Cs2CO3-doped BPhen films toward unoccupied molecular states with increasing doping concentration, leading to increase in electron concentration in the electron-transport layer and reduction in electron injection barrier height. These findings originate from energetically favorable electron transfer from Cs2CO3 to BPhen.

We present a document-grounded matching network (DGMN) for response selection that can power a knowledge-aware retrieval-based chatbot system. The challenges of building such a model lie in how to ground conversation contexts with background documents and how to recognize important information in the documents for matching. To overcome the challenges, DGMN fuses information in a document and a context into representations of each other, and dynamically determines if grounding is necessary and importance of different parts of the document and the context through hierarchical interaction with a response at the matching step. Empirical studies on two public data sets indicate that DGMN can significantly improve upon state-of-the-art methods and at the same time enjoys good interpretability. * Equal Contribution. † Corresponding author: Rui Yan (ruiyan@pku.edu.cn). A's profile trying new recipes makes me happy. i feel like i need to exercise more. i am an early bird , while my significant other is a night owl. i am a kitty owner. B's profile i might actually be a mermaid. i use all of my time for my education. i am very sociable and love those close to me. i enjoy swimming in the ocean , i feel in tune with its inhabitants. Context A: hi how are you today B: i am good . how are you ? A: pretty good where do you work ? True response i do not work , i am a full time student . what about you? False response i have been working as a salesman for more than 10 years. 1 For space limitation, we only show one false response here.

In this work, we experimentally demonstrated the new functions of trivalent rare earth complex in improving the electroluminescent (EL) performances of iridium complex by codoping trace Eu(TTA)3phen (TTA = thenoyltrifluoroacetone, phen = 1,10-phenanthroline) into a light-emitting layer based on PQ2Ir(dpm) (iridium(III)bis(2-phenylquinoly-N,C(2'))dipivaloylmethane). Compared with a reference device, the codoped devices displayed higher efficiencies, slower efficiency roll-off, higher brightness, and even better color purity. Experimental results demonstrated that Eu(TTA)3phen molecules function as electron trappers due to its low-lying energy levels, which are helpful in balancing holes and electrons and in broadening recombination zone. In addition, the matched triplet energy of Eu(TTA)3phen is instrumental in facilitating energy transfer from host to emitter. Finally, highly efficient red EL devices with the highest current efficiency, power efficiency and brightness up to 58.98 cd A(-1) (external quantum efficiency (EQE) of 21%), 61.73 lm W(-1) and 100870 cd m(-2), respectively, were obtained by appropriately decreasing the doping concentration of iridium complex. At certain brightness of 1000 cd m(-2), EL current efficiency up to 51.94 cd A(-1) (EQE = 18.5%) was retained. Our investigation extends the application of rare earth complexes in EL devices and provides a chance to improve the device performances.