On-device query intent prediction with lightweight LLMs to support ubiquitous conversations.

Conversational Agents (CAs) have made their way to providing interactive assistance to users. However, the current dialogue modelling techniques for CAs are predominantly based on hard-coded rules and rigid interaction flows, which negatively affects their flexibility and scalability. Large Language Models (LLMs) can be used as an alternative, but unfortunately they do not always provide good levels of privacy protection for end-users since most of them are running on cloud services. To address these problems, we leverage the potential of transfer learning and study how to best fine-tune lightweight pre-trained LLMs to predict the intent of user queries. Importantly, our LLMs allow for on-device deployment, making them suitable for personalised, ubiquitous, and privacy-preserving scenarios. Our experiments suggest that RoBERTa and XLNet offer the best trade-off considering these constraints. We also show that, after fine-tuning, these models perform on par with ChatGPT. We also discuss the implications of this research for relevant stakeholders, including researchers and practitioners. Taken together, this paper provides insights into LLM suitability for on-device CAs and highlights the middle ground between LLM performance and memory footprint while also considering privacy implications.

Soil contamination by diesel fuel destabilizes the soil microbial pools: Insights from permafrost soil incubations.

Arctic contamination by diesel fuel (DF) is of great concern because of the uncertain feedback of permafrost carbon (C) and soil microbiota to DF in the context of climate change in high latitudes. We conducted a laboratory incubation experiment with a gradient of DF addition ratios to examine the responses of the soil microbiota of the typical permafrost soils in the tundra ecosystems of the Norilsk region (Siberia). The study revealed initial heterogeneity in the microbial activity of the studied soils (Histic Gleyic Cryosols (CR-hi,gl), Turbic Cryosols (CR-tu), Turbic Spodic Folic Cryosols (CR-tu,sd,fo), Gleyic Fluvisols (FL-gl)). We applied the two-pool model for evaluation of the effect of DF on the proportions of C pools and revealed significant differences between soil types in the fast and slow C pools in response to DF addition. The results showed that DF addition treatments had varying effects on the fast and slow C pools, microbial activity, and microbial community structure in the studied soils. For minor exceptions, DF dramatically accelerated C loss from the slow C pool in all soil types. We assume that differences in C pool and microbiota responses to DF addition were caused by soil texture and changes in microbial community structure. We isolated Serratia proteamaculans, S. liquefaciens, S. plymuthica, Rhodococcus erythropolis, Pseudomonas antarctica, P. libanensis, P. brassicacearum, and P. chlororaphis from the DF-polluted soils. These species are recommended for bioremediation to mitigate the DF contamination of permafrost soils, especially regarding climate change and the sustainable well-being of Arctic ecosystems.