Recent work demonstrates the potential of multilingual pretraining of creating one model that can be used for various tasks in different languages. Previous work in multilingual pretraining has demonstrated that machine translation systems can be created by finetuning on bitext. In this work, we show that multilingual translation models can be created through multilingual finetuning. Instead of finetuning on one direction, a pretrained model is finetuned on many directions at the same time. Compared to multilingual models trained from scratch, starting from pretrained models incorporates the benefits of large quantities of unlabeled monolingual data, which is particularly important for low resource languages where bitext is not available. We demonstrate that pretrained models can be extended to incorporate additional languages without loss of performance. We double the number of languages in mBART to support multilingual machine translation models of 50 languages. Finally, we create the ML50 benchmark, covering low, mid, and high resource languages, to facilitate reproducible research by standardizing training and evaluation data. On ML50, we demonstrate that multilingual finetuning improves on average 1 BLEU over the strongest baselines (being either multilingual from scratch or bilingual finetuning) while improving 9.3 BLEU on average over bilingual baselines from scratch.

Recent work demonstrates the potential of training one model for multilingual machine translation. In parallel, denoising pretraining using unlabeled monolingual data as a starting point for finetuning bitext machine translation systems has demonstrated strong performance gains. However, little has been explored on the potential to combine denoising pretraining with multilingual machine translation in a single model. In this work, we fill this gap by studying how multilingual translation models can be created through multilingual finetuning.Fintuning multilingual model from a denoising pretrained model incorporates the benefits of large quantities of unlabeled monolingual data, which is particularly important for low resource languages where bitext is rare. Further, we create the ML50 benchmark to facilitate reproducible research by standardizing training and evaluation data. On ML50, we show that multilingual finetuning significantly improves over multilingual models trained from scratch and bilingual finetuning for translation into English. We also find that multilingual finetuning can significantly improve over multilingual models trained from scratch for zero-shot translation on non-English directions. Finally, we discuss that the pretraining and finetuning paradigm alone is not enough to address the challenges of multilingual models for to-Many directions performance.

Ga2O3 with a suitable bandgap width is emerging to be a novel candidate for future optoelectronic applications in the solar-blind region (<280 nm). Herein, solar-blind photodetectors are developed by combining layered GaSe flakes with Ga2O3 epitaxial films, forming a van der Waals pn heterojunction. Photoresponse characterizations indicate that the GaSe/Ga2O3 pn junction is highly sensitive to solar-blind ultraviolet light, with the maximum sensitivity at ∼255 nm. The photodetector demonstrates a particularly high responsivity of 51.9 A/W, a pronounced specific detectivity up to 1014 Jones, and a fast response speed of ∼0.3 ms under ultraweak light illumination. These figures of merit are significantly superior to most of the reported Ga2O3-based photodetectors. This work reveals the great potential of GaSe/Ga2O3 van der Waals heterojunction for developing next-generation, solar-blind photodetectors.

Trust is a mechanism for managing the uncertainty about autonomous entities and the information they store, and so can play an important role in any decentralized system. As a result, trust has been widely studied in multiagent systems and related fields such as the semantic web. Here we introduce a formal system of argumentation that can be used to reason using information about trust. This system is described as a set of graphs, which makes it possible to combine our approach with conventional representations of trust between individuals where the relationships between individuals are given in the form of a graph. The resulting system can easily relate the grounds of an argument to the agent that supplied the information, and can be used as the basis to compute Dungian notions of acceptability that take trust into account. We explore some of the properties of these argumentation graphs, examine the computation of trust and belief in the graphs, and illustrate the capabilities of the system on an example from the trust literature.