Hasil untuk "Earthwork. Foundations"

A. Filinis, A. Filinis, A. Bell et al.

<p>This paper reports on the instrumental design of a new cryogenically cooled middle-atmosphere water vapor radiometer developed by the University of Bern at the Institute of Applied Physics (IAP). Here, we present the instrument design for the breadboard stage. The key innovation of this new instrument is its cryogenically cooled front-end, which is designed to keep its size compact, reducing the required cooling power compared to existing cryogenically cooled radiometers. The advantage compared to uncooled instruments is the reduced receiver noise temperature and the possibility to extend the altitude coverage of the retrieval of water vapor profiles to even higher altitudes with better temporal resolution. The new radiometer is part of the Swiss H<span class="inline-formula">₂</span>O Hub and is supposed to replace the existing 22 GHz radiometer, MIAWARA, which has been in operation at the University of Bern for over 20 years at the Zimmerwald observatory. The calibration of the new instrument includes tipping curve calibration to determine tropospheric opacity, using the sky as a cold target. An ambient load serves as the hot target for the Hot-Cold calibration, and we also explore the possibility of using frequency-switch calibration to reduce the impact of non-linearities in the receiver chain, allowing for a higher integration time of the line observation compared to other calibration techniques. The combination of a cryogenic front-end and frequency switch microwave radiometers at 22 GHz has not been previously implemented in a single instrument. In addition to detailing the instrumental design and calibration techniques, we present preliminary results of atmospheric spectra obtained with the breadboard setup.</p>

Ayberk Tosun

We develop locale theory constructively and predicatively in univalent foundations (UF), with a particular focus on the theory of spectral and Stone locales. In the context of UF, predicativity refers specifically to the development of mathematics without the use of propositional resizing axioms. The traditional approach to the predicative development of point-free topology is to work with presentations of locales known as formal topologies. Here, we take a different approach: we work directly with frames, keeping careful track of the universes involved and adopting certain size assumptions to ensure that the theory is amenable to predicative development. Although it initially appears that many fundamental constructions of locale theory rely on impredicativity, we show that these can be circumvented under rather natural size assumptions. We first lay the groundwork for the predicative development of locale theory. We then orient our development towards a systematic investigation of the theory of spectral and Stone locales. We establish a categorical equivalence between large, locally small, and small-complete spectral locales and small distributive lattices. Moreover, we exhibit the category of Stone locales as a coreflective subcategory of spectral locales and spectral maps, using the construction known as the patch locale. Finally, we investigate the topology of algebraic DCPOs and Scott domains. We develop the Scott locale of a Scott domain, show that it forms a spectral locale, and then proceed to investigate its patch. Using this, we obtain a topological characterization of de Jong's notion of sharp element: we establish a correspondence between the sharp elements of a Scott domain and the points of the patch of its Scott locale. Our development is completely formalized and has been machine-checked using the Agda proof assistant.

J. Herman, J. Herman, J. Mao et al.

<p>Observations of trace gases, such as <span class="inline-formula">O₃</span>, HCHO, and <span class="inline-formula">NO₂</span>, and their seasonal dependence can be made using satellite and ground-based data from the Ozone Monitoring Instrument (OMI) satellite and Pandora ground-based instruments. Both operate with spectrometers that have similar characteristics in wavelength range and spectral resolution that enable them to retrieve total column amounts of formaldehyde (TCHCHO) and nitrogen dioxide (TCNO<span class="inline-formula">₂</span>) and total column ozone (TCO). The polar orbiting OMI observes at 13:30 <span class="inline-formula">±</span> 0:25 LST (local solar time) plus an occasional second side-scan point 90 min later at mid-latitudes. The ground-based Pandora spectrometer system observes the direct sun all day, with a temporal resolution of 2 min. At most sites, the Pandora data show a strong seasonal dependence for TCO and TCHCHO and less seasonal dependence for TCNO<span class="inline-formula">₂</span>. Use of a low-pass filter LOWESS(3-month) can reveal the seasonal dependence of TCNO<span class="inline-formula">₂</span> for both OMI and Pandora at mid-latitude sites usually correlated with seasonal heating using natural gas or oil. Compared to Pandora, OMI underestimates the amount of NO<span class="inline-formula">₂</span> air pollution that occurs during most days, as the OMI TCNO<span class="inline-formula">₂</span> retrieval occurs around 13:30 <span class="inline-formula">±</span> 0:25 LST, which tends to be near the frequent minimum of the daily TCNO<span class="inline-formula">₂</span> time series. Even when the Pandora data are restricted to between 13:00 and 14:00 LST, OMI retrieves less TCNO<span class="inline-formula">₂</span> than Pandora over urban sites because of OMI's large field of view. The seasonal behavior of TCHCHO is mostly caused by the release of HCHO precursors from plant growth and emissions from lakes that peak in the summer, as observed by Pandora and OMI. Long-term averages show that OMI TCHCHO usually has the same seasonal dependence but differs in magnitude from the amount measured by Pandora and is frequently larger. Comparisons of OMI total column <span class="inline-formula">NO₂</span> and HCHO with Pandora daily time series show both agreement and disagreement at various sites and for different days, with the Pandora results frequently being larger. For ozone, daily time-dependent comparisons of OMI TCO with those retrieved by Pandora show good agreement in most cases. Additional diurnal comparisons are shown of Pandora TCO with hourly retrievals during a day from the EPIC (Earth Polychromatic Imaging Camera) spacecraft instrument orbiting the Earth–Sun Lagrange point <span class="inline-formula"><i>L</i>₁</span>.</p>

A.-K. Kunz, A.-K. Kunz, L. Borchardt et al.

<p><span id="page5350"/>A relaxed eddy accumulation (REA) system was developed and tested, enabling conditional sampling of air for subsequent <span class="inline-formula">¹⁴CO₂</span> analysis. This allows a <span class="inline-formula">¹⁴C</span>-based estimation of fossil fuel <span class="inline-formula">CO₂</span> concentrations in the collected air samples and, thus, an observation-based partitioning of total <span class="inline-formula">CO₂</span> fluxes measured in urban environments by eddy covariance into fossil and non-fossil components. This article describes the REA system, evaluates its performance, and assesses uncertainties in the concentration measurements. In the REA system, two separate inlet lines equipped with fast-response valves and loop systems adapted to the technical requirements enable the conditional collection of air in two sets of aluminum cylinders for updraft and downdraft samples, respectively. The switching between updraft sampling, downdraft sampling, and standby mode is thereby determined by the vertical wind measured at 20 Hz by a co-located ultrasonic 3D anemometer. A logger program provides different options for the definition of a deadband, which is used to increase the concentration differences between updraft and downdraft samples. After the sampling interval, the accumulated air is transferred by an automated 24-port flask sampler into 3 L glass flasks, which can be analyzed in the laboratory, and the cylinders are re-evacuated for the next sampling. The REA system was tested in the laboratory, as well as on a tall tower near the city center of Zurich, Switzerland. Between July 2022 and April 2023, 103 REA updraft and downdraft flask pairs for flux measurements and 9 flask pairs for quality control purposes were selected from the tall tower for laboratory analysis based on suitable micro-meteorological conditions. Uncertainties in the <span class="inline-formula">CO₂</span> concentration differences between updraft and downdraft flasks were estimated by simulations using 20 Hz in situ measurements of a closed-path gas analyzer and an open-path gas analyzer co-located with the ultrasonic anemometer. The measurements show that there is no significant bias in the concentration differences between updraft and downdraft samples and that uncertainties due to the sampling process are negligible when estimating fossil fuel <span class="inline-formula">CO₂</span> signals. In the Zurich measurements, the <span class="inline-formula">CO₂</span> concentration differences between the flask pairs agreed with the differences obtained from in situ measurements within <span class="inline-formula">−</span>0.005 <span class="inline-formula">±</span> 0.227 ppm. The largest source of uncertainty, as well as the main limitation, in the separation of fossil and non-fossil <span class="inline-formula">CO₂</span> signals in Zurich was the small signal-to-noise ratio of the <span class="inline-formula">Δ</span><span class="inline-formula">¹⁴C</span> differences measured by accelerator mass spectrometry between the updraft and downdraft flasks. The novel REA flask sampling system meets the high technical requirements of the REA method and is a promising technology for observation-based estimation of fossil fuel <span class="inline-formula">CO₂</span> fluxes.</p>

A.-M. Sundström, M. Doutriaux-Boucher, S. Jafariserajehlou et al.

<p>The Polar Multi-Sensor Aerosol optical properties product (PMAp) provides global Aerosol Optical Depth (AOD) observations that are retrieved using a combination of measurements from instruments onboard the Metop satellites, including the Global Ozone Monitoring Experiment-2 (GOME-2), the Infrared Atmospheric Sounding Interferometer (IASI), and the Advanced Very High Resolution Radiometer (AVHRR). The PMAp Climate Data Record (CDR), published in 2022, comprises data from the Metop-A and Metop-B satellites covering the period from 2007 to 2019. The PMAp also includes classification for selected aerosol types, including dust. Based on the classification, a dust-dominated total AOD can be extracted. The focus of this work is to assess the dust aerosols in the PMAp CDRs, by analysing the spatio-temporal occurrence of dust and aerosol classification reliability, as well as by carrying out dust-dominated total AOD validation against AErosol RObotic NETwork (AERONET) observations. Our results show that the occurrence and classification of PMAp dust-dominated AOD agrees well with AERONET metrics. For PMAp dust-dominated total AODs, moderate to strong correlations with AERONET (0.45–0.8) are observed, while mean biases exhibit relatively high variability. The root-mean-square errors (RMSEs) typically represent 50 %–80 % of the mean AERONET AOD conditions. As most of the comparisons here occur at relatively high AOD levels over bright land surfaces, where measurement uncertainties and variability are inherently greater, this is somewhat expected. The results also bring up certain challenges, e.g. PMAp AOD overestimation at Central Asian AERONET stations or occasional occurrences of dust-dominated total AODs that appeared as clear outliers in AERONET comparisons. Further investigation is needed to determine their underlying causes. On a larger spatial scale, The PMAp CDRs can capture the expected seasonal variation in dust-affected AODs, such as over the Saharan outflow area, but sampling density can vary across seasons, especially over land. Therefore, full AOD distributions, along with median and mean, should be analyzed to ensure accurate conclusions. Despite challenges, the PMAp CDRs show potential for monitoring global dust aerosol patterns.</p>

R. Anthes, J. Sjoberg, J. Starr et al.

<p>The Radio Occultation Modeling EXperiment (ROMEX) is an international collaboration to test the impact of varying numbers of radio occultation (RO) profiles in operational numerical weather prediction (NWP) models. An average of 35 000 RO profiles d<span class="inline-formula">⁻¹</span> for September–November 2022 from 13 different missions are being used in experiments at major NWP centers. This paper evaluates properties of ROMEX data, with emphasis on the three largest datasets: COSMIC-2 (Constellation Observing System for Meteorology, Ionosphere and Climate-2 or C2), Spire, and Yunyao.</p> <p>The penetration depths (percent of profiles reaching different levels above the surface) of most of the ROMEX datasets are similar, with more than 80 % of all occultations reaching 2 km or lower and more than 50 % reaching 1 km or lower.</p> <p>The relative uncertainties of the C2, Spire, and Yunyao bending angles and refractivities are estimated using the three-cornered hat method. They are similar on the average in the region of overlap (45–45° N). Larger uncertainties occur in the tropics compared to higher latitudes below 20 km. Relatively small variations in longitude exist.</p> <p>We investigate biases in the observations by comparing them to each other and to models. C2 bending angles appear to be biased by about 0.15 % compared to Spire and other ROMEX data between 10 and 30 km altitude. These biases, most of which are representativeness or sampling differences, are caused by the different orbits of C2 and other ROMEX missions around the non-spherical Earth and the associated varying radii of curvature.</p>

Vijay Prakash Dwivedi, Charilaos Kanatsoulis, Shenyang Huang et al.

Graph machine learning has led to a significant increase in the capabilities of models that learn on arbitrary graph-structured data and has been applied to molecules, social networks, recommendation systems, and transportation, among other domains. Data in multi-tabular relational databases can also be constructed as 'relational entity graphs' for Relational Deep Learning (RDL) - a new blueprint that enables end-to-end representation learning without traditional feature engineering. Compared to arbitrary graph-structured data, relational entity graphs have key properties: (i) their structure is defined by primary-foreign key relationships between entities in different tables, (ii) the structural connectivity is a function of the relational schema defining a database, and (iii) the graph connectivity is temporal and heterogeneous in nature. In this paper, we provide a comprehensive review of RDL by first introducing the representation of relational databases as relational entity graphs, and then reviewing public benchmark datasets that have been used to develop and evaluate recent GNN-based RDL models. We discuss key challenges including large-scale multi-table integration and the complexities of modeling temporal dynamics and heterogeneous data, while also surveying foundational neural network methods and recent architectural advances specialized for relational entity graphs. Finally, we explore opportunities to unify these distinct modeling challenges, highlighting how RDL converges multiple sub-fields in graph machine learning towards the design of foundation models that can transform the processing of relational data.

Yuanfeng Ji, Dan Lin, Xiyue Wang et al.

The scarcity of well-annotated diverse medical images is a major hurdle for developing reliable AI models in healthcare. Substantial technical advances have been made in generative foundation models for natural images. Here we develop `ChexGen', a generative vision-language foundation model that introduces a unified framework for text-, mask-, and bounding box-guided synthesis of chest radiographs. Built upon the latent diffusion transformer architecture, ChexGen was pretrained on the largest curated chest X-ray dataset to date, consisting of 960,000 radiograph-report pairs. ChexGen achieves accurate synthesis of radiographs through expert evaluations and quantitative metrics. We demonstrate the utility of ChexGen for training data augmentation and supervised pretraining, which led to performance improvements across disease classification, detection, and segmentation tasks using a small fraction of training data. Further, our model enables the creation of diverse patient cohorts that enhance model fairness by detecting and mitigating demographic biases. Our study supports the transformative role of generative foundation models in building more accurate, data-efficient, and equitable medical AI systems.

Rujia Yang, Geng Chen, Chuan Wen et al.

Following its success in natural language processing and computer vision, foundation models that are pre-trained on large-scale multi-task datasets have also shown great potential in robotics. However, most existing robot foundation models rely solely on 2D image observations, ignoring 3D geometric information, which is essential for robots to perceive and reason about the 3D world. In this paper, we introduce FP3, a first large-scale 3D foundation policy model for robotic manipulation. FP3 builds on a scalable diffusion transformer architecture and is pre-trained on 60k trajectories with point cloud observations. With the model design and diverse pre-training data, FP3 can be efficiently fine-tuned for downstream tasks while exhibiting strong generalization capabilities. Experiments on real robots demonstrate that with only 80 demonstrations, FP3 is able to learn a new task with over 90% success rates in novel environments with unseen objects, significantly surpassing existing robot foundation models.

Arash Afkanpour, Omkar Dige, Fatemeh Tavakoli et al.

Current evaluation frameworks for foundation models rely heavily on static, manually curated benchmarks, limiting their ability to capture the full breadth of model capabilities. This paper introduces Active learning for Capability Evaluation (ACE), a novel framework for scalable, automated, and fine-grained evaluation of foundation models. ACE leverages the knowledge embedded in powerful frontier models to decompose a domain into semantically meaningful capabilities and generates diverse evaluation tasks, significantly reducing human effort. In Mathematics, ACE generated 433 capabilities and 11,800 tasks, covering 94% of Wikipedia-defined skills in the domain while introducing novel, coherent ones. To maximize efficiency, ACE fits a capability model in latent semantic space, allowing reliable approximation of a subject model's performance by evaluating only a subset of capabilities via active learning. It reaches within 0.01 RMSE of exhaustive evaluation by evaluating less than half of capabilities. Compared to static datasets, ACE provides more balanced coverage and uncovers fine-grained differences that aggregate metrics fail to capture. Our results demonstrate that ACE provides a more complete and informative picture of model capabilities, which is essential for safe and well-informed deployment of foundation models.

Anurag Vaidya, Andrew Zhang, Guillaume Jaume et al.

Foundation models are reshaping computational pathology by enabling transfer learning, where models pre-trained on vast datasets can be adapted for downstream diagnostic, prognostic, and therapeutic response tasks. Despite these advances, foundation models are still limited in their ability to encode the entire gigapixel whole-slide images without additional training and often lack complementary multimodal data. Here, we introduce Threads, a slide-level foundation model capable of generating universal representations of whole-slide images of any size. Threads was pre-trained using a multimodal learning approach on a diverse cohort of 47,171 hematoxylin and eosin (H&E)-stained tissue sections, paired with corresponding genomic and transcriptomic profiles - the largest such paired dataset to be used for foundation model development to date. This unique training paradigm enables Threads to capture the tissue's underlying molecular composition, yielding powerful representations applicable to a wide array of downstream tasks. In extensive benchmarking across 54 oncology tasks, including clinical subtyping, grading, mutation prediction, immunohistochemistry status determination, treatment response prediction, and survival prediction, Threads outperformed all baselines while demonstrating remarkable generalizability and label efficiency. It is particularly well suited for predicting rare events, further emphasizing its clinical utility. We intend to make the model publicly available for the broader community.

David Wakeham

Qubits are a great way to build a quantum computer, but a limited way to program one. We replace the usual "states and gates" formalism with a "props and ops" (propositions and operators) model in which (a) the C*-algebra of observables supplies the syntax; (b) states, viewed as linear functionals, give the semantics; and (c) a novel diagrammatic calculus unifies the two. The first part develops the basic objects of the framework, encoding consistent patterns of operator correlation, recovering Hilbert space via the GNS construction, and re-deriving the Bloch sphere as the set of all consistent correlations of operators in the Pauli algebra. We then turn to intervention, showing how measurement modifies state, proving an operator-algebraic version of the Knill-Laflamme conditions, and expressing stabilizer codes with the same diagrammatic machinery. This provides a concise, representation-agnostic account of quantum error correction. The result is a self-contained foundation in which C*-algebras, and their dual Hilbert spaces, offer a rich and universal substrate for quantum programming; forthcoming papers will build a high-level language and quantum software applications on top of this substrate.

Nathaniel H. Park, Eduardo Soares, Victor Y. Shirasuna et al.

From the relative scarcity of training data to the lack of standardized benchmarks, the development of foundation models for polymers face significant and multi-faceted challenges. At the core, many of these issues are tied directly to the structural representation of polymers and here, we present a new foundation model using a SMILES-based polymer graph representation. This approach allows representation of critical polymer architectural features and connectivity that are not available in other SMILES-based representations. The developed polymer foundation model exhibited excellent performance on 28 different benchmark datasets. Critical evaluation of the developed representation against other variations in control experiments reveals this approach to be a highly performant method of representing polymers in language-based foundation models. These control experiments also reveal a strong invariance of all SMILES representations, with many variations achieving state-of-the-art or near state-of-the-art performance, including those which are chemically or semantically invalid. Examination of error sources and attention maps for the evaluated representations corroborate the findings of the control experiments, showing that chemistry language models based on SMILES interpolate over all sequence space for prediction tasks, not only those of semantically valid inputs. Overall, this work highlights the importance of control experiments as a check on human-imposed assumptions that can limit rational design of both chemistry foundation models and their underlying structural representations.

L. Chang, J. Li, J. Li et al.

<p>The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) on board the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite has been widely used in climate and environment studies to obtain the vertical profiles of atmospheric aerosols. To retrieve the vertical profile of aerosol extinction, the CALIOP algorithm assumes column-averaged lidar ratios based on a clustering of aerosol optical properties measured at surface stations. On one hand, these lidar ratio assumptions may not be appropriate or representative at certain locations. One the other hand, the two-wavelength design of CALIOP has the potential to constrain aerosol size information, which has not been considered in the operational algorithm. In this study, we present a modified inversion algorithm to simultaneously retrieve aerosol extinction and effective radius profiles using two-wavelength elastic lidars such as CALIOP. Specifically, a lookup table is built to relate the lidar ratio with the Ångström exponent calculated using aerosol extinction at the two wavelengths, and the lidar ratio is then determined iteratively without a priori assumptions. The retrieved two-wavelength extinction at each layer is then converted to the particle effective radius assuming a lognormal distribution. The algorithm is tested on synthetic data, Raman lidar measurements and then finally the real CALIOP backscatter measurements. Results show improvements over the CALIPSO operational algorithm by comparing with ground-based Raman lidar profiles.</p>

A. Keppens, S. Di Pede, D. Hubert et al.

<p>The Sentinel-5 Precursor (S5P) satellite operated by the European Space Agency has carried the TROPOspheric Monitoring Instrument (TROPOMI) on a Sun-synchronous low-Earth orbit since 13 October 2017. The S5P mission has acquired more than 5 years of TROPOMI nadir ozone profile data retrieved from the level 0 to 1B processor version 2.0 and the level 1B to 2 optimal-estimation-based processor version 2.4.0. The latter is described in detail in this work, followed by the geophysical validation of the resulting ozone profiles for the period May 2018 to April 2023. Comparison of TROPOMI ozone profile data to co-located ozonesonde and lidar measurements used as references concludes to a median agreement better than 5 % to 10 % in the troposphere. The bias goes up to <span class="inline-formula">−</span>15 % in the upper stratosphere (35–45 km) where it can exhibit vertical oscillations. The comparisons show a dispersion of about 30 % in the troposphere and 10 % to 20 % in the upper troposphere to lower stratosphere and in the middle stratosphere, which is close to mission requirements. Chi-square tests of the observed differences confirm on average the validity of the ex ante (prognostic) satellite and ground-based data uncertainty estimates in the middle stratosphere above about 20 km. Around the tropopause and below, the mean chi-square value increases up to about four, meaning that the ex ante TROPOMI uncertainty is underestimated. The information content of the ozone profile retrieval is characterised by about five to six<span id="page3970"/> vertical subcolumns of independent information and a vertical sensitivity (i.e. the fraction of the information that originates from the measurement) nearly equal to unity at altitudes from about 20 to 50 km, decreasing rapidly at altitudes above and below. The barycentre of the retrieved information is usually close to the nominal retrieval altitude in the 20–50 km altitude range, with positive and negative offsets of up to 10 km below and above this range, respectively. The effective vertical resolution of the profile retrieval usually ranges within 10–15 km, with a minimum close to 7 km in the middle stratosphere. Increased sensitivities and higher effective vertical resolutions are observed at higher solar zenith angles (above about 60°), as can be expected, and correlate with higher retrieved ozone concentrations. The vertical sensitivity of the TROPOMI tropospheric ozone retrieval is found to depend on the solar zenith angle, which translates into a seasonal and meridian dependence of the bias with respect to reference measurements. A similar although smaller effect can be seen for the viewing zenith angle. Additionally, the bias is negatively correlated with the surface albedo for the lowest three ozone subcolumns (0–18 km), despite the albedo's apparently slightly positive correlation with the retrieval degrees of freedom in the signal. For the 5 years of TROPOMI ozone profile data that are available now, an overall positive drift is detected for the same three subcolumns, while a negative drift is observed above (24–32 km), resulting in a negligible vertically integrated drift.</p>

P. H. Herig Coimbra, P. H. Herig Coimbra, B. Loubet et al.

<p>Greenhouse gas monitoring is important to ensure climate goals are being achieved. This study unveils the potential of using atmospheric tall towers in direct flux measurements, bridging the gap between atmospheric and ecosystem monitoring networks. The ICOS Cities (PAUL) project aims to monitor CO<span class="inline-formula">₂</span> emissions in urban areas, where concentrated emissions make them key targets for climate change mitigation. This study explores the synergy between ICOS atmospheric and ecosystem networks by utilizing slow-response analysers (<span class="inline-formula">∼</span> 3 s) on tall atmospheric towers for ecosystem studies using the eddy covariance method. A standard setup with an ultrasonic anemometer and an infrared (IR) fast-response CO<span class="inline-formula">₂</span> analyser was installed and compared with measurements from an existing cavity ring-down spectroscopy (CRDS) analyser measuring CO<span class="inline-formula">₂</span>, CO, and CH<span class="inline-formula">₄</span>. Deployed on the 100 m Saclay tower near Paris, covering a 43.9 km<span class="inline-formula">²</span> 80 % footprint with heavy traffic roads, a nearby heating plant, and a forest, the setup addressed technical challenges and height-induced complexities. Corrections for flux attenuation by high-frequency losses were limited to <span class="inline-formula">&lt;</span> 20 % on average for all stabilities and around 11 % for unstable conditions. Elevated mean fluxes for CO<span class="inline-formula">₂</span> (10 <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M9" display="inline" overflow="scroll" dspmath="mathml"><mrow class="unit"><mi mathvariant="normal">µ</mi><mi mathvariant="normal">mol</mi><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">m</mi><mrow><mo>-</mo><mn mathvariant="normal">2</mn></mrow></msup><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">s</mi><mrow><mo>-</mo><mn mathvariant="normal">1</mn></mrow></msup></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="64pt" height="15pt" class="svg-formula" dspmath="mathimg" md5hash="cef32b881b0ccfba22dd5228ab8f6fda"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="amt-17-6625-2024-ie00001.svg" width="64pt" height="15pt" src="amt-17-6625-2024-ie00001.png"/></svg:svg></span></span>) and CH<span class="inline-formula">₄</span> (200 <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M11" display="inline" overflow="scroll" dspmath="mathml"><mrow class="unit"><mi mathvariant="normal">µ</mi><mi mathvariant="normal">mol</mi><mspace linebreak="nobreak" width="0.125em"/><msup><mi mathvariant="normal">m</mi><mrow><mo>-</mo><mn mathvariant="normal">2</mn></mrow></msup><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">s</mi><mrow><mo>-</mo><mn mathvariant="normal">1</mn></mrow></msup></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="64pt" height="15pt" class="svg-formula" dspmath="mathimg" md5hash="efc875b69c70843087788d4fc95915b5"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="amt-17-6625-2024-ie00002.svg" width="64pt" height="15pt" src="amt-17-6625-2024-ie00002.png"/></svg:svg></span></span>) were observed from the heating plant wind direction during December and January. Conversely, the forest direction exhibited the strongest sink among all wind directions, with <span class="inline-formula">−</span>4 <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M13" display="inline" overflow="scroll" dspmath="mathml"><mrow class="unit"><mi mathvariant="normal">µ</mi><mi mathvariant="normal">mol</mi><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">m</mi><mrow><mo>-</mo><mn mathvariant="normal">2</mn></mrow></msup><mspace width="0.125em" linebreak="nobreak"/><msup><mi mathvariant="normal">s</mi><mrow><mo>-</mo><mn mathvariant="normal">1</mn></mrow></msup></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="64pt" height="15pt" class="svg-formula" dspmath="mathimg" md5hash="e7d76965a1a4ff05670f87e685f660ec"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="amt-17-6625-2024-ie00003.svg" width="64pt" height="15pt" src="amt-17-6625-2024-ie00003.png"/></svg:svg></span></span> during July and August. Storage and vertical advection were estimated using the routine three-level profile measurements done in ICOS atmospheric towers. Storage term was of the same magnitude as turbulent flux, increasing at night and de-stocking during the first half of the day. Vertical advection averaged zero on a monthly basis. These results demonstrate the feasibility and versatility of utilizing atmospheric towers for urban emission monitoring, offering valuable insights for emission monitoring strategies worldwide.</p>

Kyudan Jung, Nam-Joon Kim, Hyun Gon Ryu et al.

As the interest in large language models grows, the importance of accuracy in automatic speech recognition has become more pronounced. This is especially true for lectures that include specialized terminology. In such cases, the success rate of traditional ASR models tends to be low, presenting a significant challenge. A method using the word frequency difference approach has been proposed to improve ASR performance for specialized terminology. We investigated this proposal through experiments and data analysis to determine if it effectively addresses the issue. In addition, we introduced the power law as the theoretical foundation for the relative frequency methodology mentioned in this approach.

Fangcen Liu, Chenqiang Gao, Yaming Zhang et al.

In recent years, the foundation models have swept the computer vision field and facilitated the development of various tasks within different modalities. However, it remains an open question on how to design an infrared foundation model. In this paper, we propose InfMAE, a foundation model in infrared modality. We release an infrared dataset, called Inf30 to address the problem of lacking large-scale data for self-supervised learning in the infrared vision community. Besides, we design an information-aware masking strategy, which is suitable for infrared images. This masking strategy allows for a greater emphasis on the regions with richer information in infrared images during the self-supervised learning process, which is conducive to learning the generalized representation. In addition, we adopt a multi-scale encoder to enhance the performance of the pre-trained encoders in downstream tasks. Finally, based on the fact that infrared images do not have a lot of details and texture information, we design an infrared decoder module, which further improves the performance of downstream tasks. Extensive experiments show that our proposed method InfMAE outperforms other supervised methods and self-supervised learning methods in three downstream tasks.

Kanishk Gandhi, Zoe Lynch, Jan-Philipp Fränken et al.

Understanding emotions is fundamental to human interaction and experience. Humans easily infer emotions from situations or facial expressions, situations from emotions, and do a variety of other affective cognition. How adept is modern AI at these inferences? We introduce an evaluation framework for testing affective cognition in foundation models. Starting from psychological theory, we generate 1,280 diverse scenarios exploring relationships between appraisals, emotions, expressions, and outcomes. We evaluate the abilities of foundation models (GPT-4, Claude-3, Gemini-1.5-Pro) and humans (N = 567) across carefully selected conditions. Our results show foundation models tend to agree with human intuitions, matching or exceeding interparticipant agreement. In some conditions, models are ``superhuman'' -- they better predict modal human judgements than the average human. All models benefit from chain-of-thought reasoning. This suggests foundation models have acquired a human-like understanding of emotions and their influence on beliefs and behavior.

Yinghao Ma, Anders Øland, Anton Ragni et al.

In recent years, foundation models (FMs) such as large language models (LLMs) and latent diffusion models (LDMs) have profoundly impacted diverse sectors, including music. This comprehensive review examines state-of-the-art (SOTA) pre-trained models and foundation models in music, spanning from representation learning, generative learning and multimodal learning. We first contextualise the significance of music in various industries and trace the evolution of AI in music. By delineating the modalities targeted by foundation models, we discover many of the music representations are underexplored in FM development. Then, emphasis is placed on the lack of versatility of previous methods on diverse music applications, along with the potential of FMs in music understanding, generation and medical application. By comprehensively exploring the details of the model pre-training paradigm, architectural choices, tokenisation, finetuning methodologies and controllability, we emphasise the important topics that should have been well explored, like instruction tuning and in-context learning, scaling law and emergent ability, as well as long-sequence modelling etc. A dedicated section presents insights into music agents, accompanied by a thorough analysis of datasets and evaluations essential for pre-training and downstream tasks. Finally, by underscoring the vital importance of ethical considerations, we advocate that following research on FM for music should focus more on such issues as interpretability, transparency, human responsibility, and copyright issues. The paper offers insights into future challenges and trends on FMs for music, aiming to shape the trajectory of human-AI collaboration in the music realm.