Generating Diverse Translation by Manipulating Multi-Head Attention

Transformer model (Vaswani et al. 2017) has been widely used in machine translation tasks and obtained state-of-the-art results. In this paper, we report an interesting phenomenon in its encoder-decoder multi-head attention: different attention heads of the final decoder layer align to different word translation candidates. We empirically verify this discovery and propose a method to generate diverse translations by manipulating heads. Furthermore, we make use of these diverse translations with the back-translation technique for better data augmentation. Experiment results show that our method generates diverse translations without a severe drop in translation quality. Experiments also show that back-translation with these diverse translations could bring a significant improvement in performance on translation tasks. An auxiliary experiment of conversation response generation task proves the effect of diversity as well.

SOME PECULIARITIES OF THE ORGANISM’S RESPONSES TO A LONG-TERM INHALATION OF SILICA-CONTAINING SUBMICRON (PREDOMINANTLY, NANOSCALE) PARTICLES IN A REAL INDUSTRIAL AEROSOL

Female white rats were exposed in a «nose only» inhalation device to an aerosol containing predominantly submicron (nanoscale included) particles of amorphous silica in a total concentration of 2.6±0.6 or 10.6±2.1 mg/m3, 4 h/day, 5 times a week, during up to 6 months. In an auxiliary experiment with a single-shot intratracheal instillation of these particles, it was shown that they induced a pulmonary cell response comparable with that when administrated a highly cytotoxic and fibrogenic standard quartz dust DQ12. However in a long-term inhalation test, the aerosol investigated proved to be of a very low systemic toxicity and fibrogenicity. This paradox may be explained by a low retention of SiO2 in lungs and other organs due to a relatively high in vivo solubility of those nanoparticles. Nevertheless their genotoxic action and transnasal penetration into the brain urge caution when assessing occupational or environmental hazard of that aerosol.

Demagnetizing Experiment of Magnetism-Concealed Tank Based on Smart Materials

The core component -----a set of coaxial densely winded degaussing loops of a magnetism-concealed tank is designed and made by experiment method. From the data, it can be concluded that the coaxial densely winded degaussing loops can eliminate the tanks magnetic field signs which are engendered by magnetization in the earth magnetic field, specially, the demagnetizing effect near the axis is more obvious. It should be paid attention to the direction of the demagnetizing electric current in experimentation, another important factor to get optimal demagnetizing effect is to seize the right time of demagnetizing electric current. If a demagnetizing electric current is kept too long, as can be seen from the table, the tanks magnetic field signs will be strengthened owing to the counter-rotational magnetization, which accords with the electromagnetic theories. In addition, the degaussing loops work characteristics are given in our auxiliary experiment. In conclusion this experiment design method provides a sustain for magnetism protection theory of equips, and a train of thought is offered for both the magnetism-concealed tanks designing and researching through the experiment process.

27.—The Zeros of Optical Coherence Functions and their Utility in the Study of Spectra

SynopsisThis paper shows that the coherence function associated with a band-limited optical spectrum is expressible as an infinite product involving its zeros. Only a finite number of zeros are physically significant and these can be determined from measurements of the visibility of interference fringes; but an ambiguity remains in the sign of the imaginary part of each zero. If the spectrum is to be recovered using the visibility measurements, an auxiliary experiment is needed to supply the necessary signs, but it need not be especially accurate. If the signs are not known there is only a limited number of different spectra that are compatible with the visibility measurements. Finally it is suggested that wavelength measurements made on an asymmetric spectral line with a Michelson interferometer may yield differing results when used with long and short delays.