Search

Discounting is standard in economics to consider time preferences of people and account for future market changes. However, so far discounting has mainly been applied to monetary flows and ignored for many ecosystem services. In multi-objective optimization, selectively disregarding time preference for some non-monetary services create bias. Here we study how discounting a range of ecosystem service indicators influences a public planner's optimal land allocation. We used a robust multi-objective optimization approach to model a mixed forestry-avocado farm portfolio in South Africa. The objectives for optimization were the provisioning of various ecosystem services and disservices represented by four indicators: net present value, payback period,carbon sequestration, andfertilizeruse. To account for time preferences concerning indicator flows, we applied specific discount rates to each ecosystem service indicator, depending on its character (non-monetary or monetary indicators). We demonstrate that discounting reduces the standard deviations of the discounted sum of the indicators, which leads to less diversified land-use portfolios. To account for discount rate uncertainty, we introduced three indicator sets simultaneously, each using a different discount rate, which was off setting the effect of decreasing diversification.

Recently introduced language model prompting methods can achieve high accuracy in zero- and few-shot settings while requiring few to no learned task-specific parameters. Nevertheless, these methods still often trail behind full model finetuning. In this work, we investigate if a dedicated continued pretraining stage could improve "promptability", i.e., zero-shot performance with natural language prompts or few-shot performance with prompt tuning. We reveal settings where existing continued pretraining methods lack promptability. We also identify current methodological gaps, which we fill with thorough large-scale experiments. We demonstrate that a simple recipe, continued pretraining that incorporates a trainable prompt during multi-task learning, leads to improved promptability in both zero- and few-shot settings compared to existing methods, up to 31% relative. On the other hand, we find that continued pretraining using MAML-style meta-learning, a method that directly optimizes few-shot promptability, yields subpar performance. We validate our findings with two prompt tuning methods, and, based on our results, we provide concrete recommendations to optimize promptability for different use cases., EMNLP 2022

Auctions have attracted growing attention as bidding mechanisms for soliciting or allocating payments for a wide range of ecosystem services (ES). This paper reviews the latest scientific knowledge on ES auctioning approaches. Using systematically selected academic articles, we trace and discuss the development of ES auction literature across space, time, target ecosystem, and mechanism type. We integrate previous attempts to organize this body of work to produce a composite factor map of entry points to more specialized sub-literatures engaging with current issues in auction design and implementation. The results show that most academic work focuses on reverse auctions, where landowners bid their willingness to accept contracts to protect or promote ES provisioning, but we also locate several forward (i.e. beneficiaries bid theirwillingness to payfor ES) and mixed mechanisms. We critically analyze major advantages and challenges for each approach, emphasizing issues related to transaction costs and accessibility for participants and agencies. Overall, our findings suggest that ES auctions have a robust track record but remain administratively and logistically challenging. Further investment in open-source tools, shared infrastructure, and other efforts to make auctions more accessible to researchers, agencies, and participants alike is strongly indicated.

Pretrained Language Models (LMs) have demonstrated ability to perform numerical reasoning by extrapolating from a few examples in few-shot settings. However, the extent to which this extrapolation relies on robust reasoning is unclear. In this paper, we investigate how well these models reason with terms that are less frequent in the pretraining data. In particular, we examine the correlations between the model performance on test instances and the frequency of terms from those instances in the pretraining data. We measure the strength of this correlation for a number of GPT-based language models (pretrained on the Pile dataset) on various numerical deduction tasks (e.g., arithmetic and unit conversion). Our results consistently demonstrate that models are more accurate on instances whose terms are more prevalent, in some cases above $70\%$ (absolute) more accurate on the top 10\% frequent terms in comparison to the bottom 10\%. Overall, although LMs exhibit strong performance at few-shot numerical reasoning tasks, our results raise the question of how much models actually generalize beyond pretraining data, and we encourage researchers to take the pretraining data into account when interpreting evaluation results.

Additional file 9. Boxplots of cluster sizes formed from ONT MinION simulated datasets.

Additional file 3. Boxplots of clustering specificity on PacBio Sequel simulated datasets.

Additional file 11. Weight threshold determination.

Additional file 2. Sensitivity and specificity of clustering PacBio Sequel simulated datasets.

Additional file 5. Boxplots of clustering specificity on ONT MinION simulated datasets.

Additional file 7. Boxplots of cluster sizes formed from PacBio Sequel simulated datasets.

Additional file 8. Characteristics of clusters formed from ONT MinION simulated datasets.

Additional file 4. Sensitivity and specificity of clustering ONT MinION simulated datasets.