It is challenging to train a good intent classifier for a task-oriented dialogue system with only a few annotations. Recent studies have shown that fine-tuning pre-trained language models with a small amount of labeled utterances from public benchmarks in a supervised manner is extremely helpful. However, we find that supervised pre-training yields an anisotropic feature space, which may suppress the expressive power of the semantic representations. Inspired by recent research in isotropization, we propose to improve supervised pretraining by regularizing the feature space towards isotropy. We propose two regularizers based on contrastive learning and correlation matrix respectively, and demonstrate their effectiveness through extensive experiments. Our main finding is that it is promising to regularize supervised pre-training with isotropization to further improve the performance of few-shot intent detection. The source code can be found at https://github.com/ fanolabs/isoIntentBert-main.
We propose simultaneous confidence bands of the hyperbolic-type for the contrasts between several nonlinear (curvilinear) regression curves. The critical value of a confidence band is determined from the distribution of the maximum of a chisquare random process defined on the domain of explanatory variables. We use the volume-of-tube method to derive an upper tail probability formula of the maximum of a chi-square random process, which is asymptotically exact and sufficiently accurate in commonly used tail regions. Moreover, we prove that the formula obtained is equivalent to the expectation of the Euler-Poincaré characteristic of the excursion set of the chi-square random process, and hence conservative. This result is therefore a generalization of Naiman's inequality for Gaussian random processes. As an illustrative example, growth curves of consomic mice are analyzed.
The alkaline components in red mud represent one of the crucial factors restricting its application, especially for the construction and building industry. The phase state of alkaline components has a significant influence on the dealkalization of red mud. In this work, an environmentally friendly acid leaching strategy is proposed by controlling the phase transformation of red mud during active roasting pretreatment. With a moderate roasting temperature, the alkaline component is prevented from converting into insoluble phases. After acid leaching with a low concentration of 0.1 M, a high dealkalization rate of 92.8% is obtained. Besides, the leachate is neutral (pH = 7) and the valuable metals in red mud are well preserved, manifesting a high selectivity and efficiency of diluted acid leaching. The calcination experiment further confirms the practicability of the strategy in the construction field, where the cementitious minerals can be formed in large quantities. Compared with the traditional acid leaching routes, the diluted acid leaching strategy in this work is acid saving with low valuable element consumption. Meanwhile, the secondary pollution issue can be alleviated. Hence, the findings in this work provide a feasible approach for the separation and recovery of alkali and resource utilization of red mud.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.