Search for: All records

Mathematical reasoning, a core ability of human intelligence, presents unique challenges for machines in abstract thinking and logical reasoning. Recent large pre-trained language models such as GPT-3 have achieved remarkable progress on mathematical reasoning tasks written in text form, such as math word problems (MWP). However, it is unknown if the models can handle more complex problems that involve math reasoning over heterogeneous information, such as tabular data. To fill the gap, we present Tabular Math Word Problems (TABMWP), a new dataset containing 38,431 open-domain grade-level problems that require mathematical reasoning on both textual and tabular data. Each question in TABMWP is aligned with a tabular context, which is presented as an image, semi-structured text, and a structured table. There are two types of questions: free-text and multi-choice, and each problem is annotated with gold solutions to reveal the multi-step reasoning process. We evaluate different pre-trained models on TABMWP, including the GPT-3 model in a few-shot setting. As earlier studies suggest, since few-shot GPT-3 relies on the selection of in-context examples, its performance is unstable and can degrade to near chance. The unstable issue is more severe when handling complex problems like TABMWP. To mitigate this, we further propose a novel approach, PROMPTPG, which utilizes policy gradient to learn to select in-context examples from a small amount of training data and then constructs the corresponding prompt for the test example. Experimental results show that our method outperforms the best baseline by 5.31% on the accuracy metric and reduces the prediction variance significantly compared to random selection, which verifies its effectiveness in selecting in-context examples. 

Stordalen Mire is a peatland in the discontinuous permafrost zone in arctic Sweden that exhibits a habitat gradient from permafrost palsa, toSphagnumbog underlain by permafrost, toEriophorum‐dominated fully thawed fen. We used three independent approaches to evaluate the annual, multi‐decadal, and millennial apparent carbon accumulation rates (aCAR) across this gradient: seven years of direct semi‐continuous measurement of CO₂and CH₄exchange, and 21 core profiles for²¹⁰Pb and¹⁴C peat dating. Year‐round chamber measurements indicated net carbon balance of −13 ± 8, −49 ± 15, and −91 ± 43 g C m⁻² y⁻¹for the years 2012–2018 in palsa, bog, and fen, respectively. Methane emission offset 2%, 7%, and 17% of the CO₂uptake rate across this gradient. Recent aCAR indicates higher C accumulation rates in surface peats in the palsa and bog compared to current CO₂fluxes, but these assessments are more similar in the fen. aCAR increased from low millennial‐scale levels (17–29 g C m⁻² y⁻¹) to moderate aCAR of the past century (72–81 g C m⁻² y⁻¹) to higher recent aCAR of 90–147 g C m⁻² y⁻¹. Recent permafrost collapse, greater inundation and vegetation response has made the landscape a stronger CO₂sink, but this CO₂sink is increasingly offset by rising CH₄emissions, dominated by modern carbon as determined by¹⁴C. The higher CH₄emissions result in higher net CO_{2‐equivalent}emissions, indicating that radiative forcing of this mire and similar permafrost ecosystems will exert a warming influence on future climate.