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. 

Abstract Extreme rainfall events in the humid-tropical Luquillo Mountains, Puerto Rico export the bulk of suspended sediment and particulate organic carbon. Using 25 years of river carbon and suspended sediment data, which targeted hurricanes and other large rainstorms, we estimated biogenic particulate organic carbon yields of 65 ± 16 tC km⁻²yr⁻¹for the Icacos and 17.7 ± 5.1 tC km⁻²yr⁻¹for the Mameyes rivers. These granitic and volcaniclastic catchments function as substantial atmospheric carbon-dioxide sinks, largely through export of river biogenic particulate organic carbon during extreme rainstorms. Compared to other regions, these high biogenic particulate organic carbon yields are accompanied by lower suspended sediment yields. Accordingly, particulate organic carbon export from these catchments is underpredicted by previous yield relationships, which are derived mainly from catchments with easily erodible sedimentary rocks. Therefore, rivers that drain petrogenic-carbon-poor bedrock require separate accounting to estimate their contributions to the geological carbon cycle. 

PICO bubble chambers have exceptional sensitivity to inelastic dark matter-nucleus interactions due to a combination of their extended nuclear recoil energy detection window from a few keV to O(100 keV) or more and the use of iodine as a heavy target. Inelastic dark matter-nucleus scattering is interesting for studying the properties of dark matter, where many theoretical scenarios have been developed. This study reports the results of a search for dark matter inelastic scattering with the PICO-60 bubble chambers. The analysis reported here comprises physics runs from PICO-60 bubble chambers using CF3I and C3F8. The CF3I run consisted of 36.8 kg of CF3I reaching an exposure of 3415 kg-day operating at thermodynamic thresholds between 7 and 20 keV. The C3F8 runs consisted of 52 kg of C3F8 reaching exposures of 1404 kg-day and 1167 kg-day running at thermodynamic thresholds of 2.45 keV and 3.29 keV, respectively. The analysis disfavors various scenarios, in a wide region of parameter space, that provide a feasible explanation of the signal observed by DAMA, assuming an inelastic interaction, considering that the PICO CF3I bubble chamber used iodine as the target material. 

Monitoring water quality by detecting chemical and biological contaminants is critical to ensuring the provision and discharge of clean water, hence protecting human health and the ecosystem. Among the available analytical techniques, infrared (IR) spectroscopy provides sensitive and selective detection of multiple water contaminants. In this work, we present an application of IR spectroscopy for qualitative and quantitative assessment of chemical and biological water contaminants. We focus on in-line detection of nitrogen pollutants in the form of nitrate and ammonium for wastewater treatment process control and automation. We discuss the effects of water quality parameters such as salinity, pH, and temperature on the IR spectra of nitrogen pollutants. We then focus on application of the sensor for detection of contaminants of emerging concern, such as arsenic and Per- and polyfluoroalkyl substances (PFAS) in drinking water. We demonstrate the use of multivariate statistical analysis for automated data processing in complex fluids. Finally, we discuss application of IR spectroscopy for detecting biological water contaminants. We use the metabolomic signature of E. coli bacteria to determine its presence in water as well as distinguish between different strains of bacteria. Overall, this work shows that IR spectroscopy is a promising technique for monitoring both chemical and biological contaminants in water and has the potential for real-time, inline water quality monitoring. 

The LHCb upgrade represents a major change of the experiment. The detectors have been almost completely renewed to allow running at an instantaneous luminosity five times larger than that of the previous running periods. Readout of all detectors into an all-software trigger is central to the new design, facilitating the reconstruction of events at the maximum LHC interaction rate, and their selection in real time. The experiment's tracking system has been completely upgraded with a new pixel vertex detector, a silicon tracker upstream of the dipole magnet and three scintillating fibre tracking stations downstream of the magnet. The whole photon detection system of the RICH detectors has been renewed and the readout electronics of the calorimeter and muon systems have been fully overhauled. The first stage of the all-software trigger is implemented on a GPU farm. The output of the trigger provides a combination of totally reconstructed physics objects, such as tracks and vertices, ready for final analysis, and of entire events which need further offline reprocessing. This scheme required a complete revision of the computing model and rewriting of the experiment's software. 

The origin of high-energy cosmic rays, atomic nuclei that continuously impact Earth’s atmosphere, is unknown. Because of deflection by interstellar magnetic fields, cosmic rays produced within the Milky Way arrive at Earth from random directions. However, cosmic rays interact with matter near their sources and during propagation, which produces high-energy neutrinos. We searched for neutrino emission using machine learning techniques applied to 10 years of data from the IceCube Neutrino Observatory. By comparing diffuse emission models to a background-only hypothesis, we identified neutrino emission from the Galactic plane at the 4.5σ level of significance. The signal is consistent with diffuse emission of neutrinos from the Milky Way but could also arise from a population of unresolved point sources.