Hasil untuk "Earthwork. Foundations"

J. Li, J. Li, J. Li et al.

<p>Accurate determination of the methane isotopic composition (<span class="inline-formula"><i>δ</i>¹³</span>CH<span class="inline-formula">₄</span>) is essential for attributing emission sources of methane (CH<span class="inline-formula">₄</span>). However, for measurements with optical instruments, spectral interference from water vapor and instrumental drift often introduce substantial biases in <span class="inline-formula"><i>δ</i>¹³</span>CH<span class="inline-formula">₄</span> measurements, particularly for humid air measurements. Although multiple calibration strategies exist, a systematic evaluation of their performance under diverse field conditions remains lacking. Here, we evaluate two calibration strategies for a cavity ring-down spectrometer: a delta-based calibration for <span class="inline-formula"><i>δ</i>¹³</span>CH<span class="inline-formula">₄</span> and an isotopologue-specific calibration for <span class="inline-formula">¹²</span>CH<span class="inline-formula">₄</span> and <span class="inline-formula">¹³</span>CH<span class="inline-formula">₄</span>. We performed laboratory experiments over a water vapor range of 0.15 %–4.0 % to establish empirical correction functions, quadratic for <span class="inline-formula">¹²</span>CH<span class="inline-formula">₄</span> and <span class="inline-formula">¹³</span>CH<span class="inline-formula">₄</span>, and linear for <span class="inline-formula"><i>δ</i>¹³</span>CH<span class="inline-formula">₄</span>, to remove humidity-induced biases. These correction functions were then applied to field measurements in both dried air at the SORPES stie and humid air at the Jurong site. At the SORPES site where air samples were dried using a Nafion™ dryer, the mean difference in <span class="inline-formula"><i>δ</i>¹³</span>CH<span class="inline-formula">₄</span> between the two strategies was <span class="inline-formula">∼0.29</span> ‰. In contrast, for humid air at the Jurong site, significant inter-method difference (<span class="inline-formula">Δ<i>δ</i>¹³</span>CH<span class="inline-formula">₄</span>) was observed, with which exhibiting a strong correlation with <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M27" display="inline" overflow="scroll" dspmath="mathml"><mrow><mn mathvariant="normal">1</mn><mo>/</mo><msub><mrow class="chem"><mi mathvariant="normal">CH</mi></mrow><mn mathvariant="normal">4</mn></msub></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="35pt" height="14pt" class="svg-formula" dspmath="mathimg" md5hash="a724ef11153dc67aba6eb50abd0cee99"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="amt-19-1763-2026-ie00001.svg" width="35pt" height="14pt" src="amt-19-1763-2026-ie00001.png"/></svg:svg></span></span>, indicating non-linear spectral effects are most pronounced at lower CH<span class="inline-formula">₄</span> concentrations and compromise the performance of delta-based calibration. Notably, only the isotopologue-specific calibration, coupled with an explicit water vapor correction, delivered stable and accurate <span class="inline-formula"><i>δ</i>¹³</span>CH<span class="inline-formula">₄</span> measurements across all conditions. This work underscores the need for robust calibration strategies to minimize bias in CH<span class="inline-formula">₄</span> isotopic composition measurements.</p>

K.-S. Chung, C.-C. Chang, B.-X. Zhuang et al.

<p>Different configurations within the observation operator cause dual-polarization radar parameters to exhibit various characteristics, which affect the structure of background error covariance as well as the results of data assimilation. Through real case data assimilation experiments, this study evaluates the raindrop-contributed term in the simulated reflectivity (<span class="inline-formula"><i>Z</i>_HH</span>) and differential reflectivity (<span class="inline-formula"><i>Z</i>_DR</span>) to describe the effect of different calculation methods within the operator: the fitting and direct integration methods. In the fitting method, dual-polarization variables are calculated using an analytic function, which assumes a gamma-shaped drop size distribution and fits the relationship between the scattering amplitude (SA) and drop size. In the direct integration method, the quantities of the hydrometeor species and SA are integrated with respect to drop size during the calculation. The results indicate that the fitting method effectively simulates the <span class="inline-formula"><i>Z</i>_HH</span>. However, the limitations of the fitting function may impact the accuracy when represents the structure of <span class="inline-formula"><i>Z</i>_DR</span>. By contrast, the direct integration method effectively simulates polarimetric variables. Validation of the raindrop mass-weighted mean diameter (<span class="inline-formula"><i>D</i>_mr</span>) indicates that assimilation of dual-polarization radar data into the model results in adjustment of the raindrop size distribution regardless of which configuration is used. However, the <span class="inline-formula"><i>D</i>_mr</span>–<span class="inline-formula"><i>Z</i>_DR</span> structure is closer to the observed structure, and the <span class="inline-formula"><i>Z</i>_DR</span> structure is more reasonable when the direct integration method is employed. In summary, different configurations within the operator directly affect the results of data assimilation, and the direct integration method has more reasonable performance with respect to simulating dual-polarization radar variables.</p>

Xinghong Fu, Yanhong Li, Georgios Papaioannou et al.

Learning time series foundation models has been shown to be a promising approach for zero-shot time series forecasting across diverse time series domains. Insofar as scaling has been a critical driver of performance of foundation models in other modalities such as language and vision, much recent work on time series foundation modeling has focused on scaling. This has resulted in time series foundation models with hundreds of millions of parameters that are, while performant, inefficient and expensive to use in practice. This paper describes a simple recipe for learning efficient foundation models for zero-shot time series forecasting that are orders of magnitude smaller. We show that large-scale transformers are not necessary: small hybrid models that interleave long convolution and linear RNN layers (in particular DeltaNet layers) can match the performance of larger transformer-based models while being more than a hundred times smaller. We also describe several data augmentation and inference strategies that further improve performance. This recipe results in Reverso, a family of efficient time series foundation models for zero-shot forecasting that significantly push the performance-efficiency Pareto frontier.

Maxime Rochkoulets, Lovro Vrček, Mile Šikić

Foundation models in genomics have shown mixed success compared to their counterparts in natural language processing. Yet, the reasons for their limited effectiveness remain poorly understood. In this work, we investigate the role of entropy as a fundamental factor limiting the capacities of such models to learn from their training data and develop foundational capabilities. We train ensembles of models on text and DNA sequences and analyze their predictions, static embeddings, and empirical Fisher information flow. We show that the high entropy of genomic sequences -- from the point of view of unseen token prediction -- leads to near-uniform output distributions, disagreement across models, and unstable static embeddings, even for models that are matched in architecture, training and data. We then demonstrate that models trained on DNA concentrate Fisher information in embedding layers, seemingly failing to exploit inter-token relationships. Our results suggest that self-supervised training from sequences alone may not be applicable to genomic data, calling into question the assumptions underlying current methodologies for training genomic foundation models.

J. David, L. D'Angelo, M. Simon et al.

<p>Mass spectrometry techniques traditionally deployed in the field often operate at low mass resolution, making it hard to unambiguously identify and attribute organic molecules. In this regard, in-situ, accurate and precise online mass-spectrometric measurements of organic molecules in atmospheric organic aerosol (OA) are essential for understanding its sources, formation and chemical composition. In this study, we demonstrate the field applicability of a high-resolution (Orbitrap) mass spectrometer with Atmospheric Pressure Chemical Ionization (APCI-Orbitrap-MS) for real-time ambient OA measurements, achieving online, molecular resolution at atmospherically relevant concentrations with a high temporal resolution of 1 s, mass resolution of <span class="inline-formula"><i>R</i></span> <span class="inline-formula">=</span> 120 000 at <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M3" display="inline" overflow="scroll" dspmath="mathml"><mrow><mi>m</mi><mo>/</mo><mi>z</mi></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="23pt" height="14pt" class="svg-formula" dspmath="mathimg" md5hash="d797e7418bb082ad5eec13189d6e5a75"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="amt-18-4573-2025-ie00001.svg" width="23pt" height="14pt" src="amt-18-4573-2025-ie00001.png"/></svg:svg></span></span> 200, and mass accuracy of <span class="inline-formula">±</span>1.5 ppm. These features enable chemically reliable measurements in environments that are exhibiting chemically complex aerosol composition, through molecular-level detection and identification of anthropogenic pollutants, biogenic and biomass burning tracers. As proof of principle, we deployed the APCI-Orbitrap-MS for in-situ measurements in a mobile laboratory container at an urban background station at Campus Riedberg (CR, Frankfurt am Main, Germany) and an agricultural field site in Schivenoglia (SKI, Italy) in the heavily polluted Po Valley. The APCI-Orbitrap-MS showed good agreement with the organic aerosol mass of an aerosol chemical speciation monitor (ACSM), with Pearson's <span class="inline-formula"><i>R</i></span> values of 0.91 and 0.70 for the urban and agricultural sites, respectively. In SKI, we resolved distinct diurnal variations in compounds such as MBTCA (C<span class="inline-formula">₈</span>H<span class="inline-formula">₁₂</span>O<span class="inline-formula">₆</span>), a biogenic marker of photochemical aging, and C<span class="inline-formula">₈</span>H<span class="inline-formula">₁₃</span>O<span class="inline-formula">₈</span>N, an organic nitrate indicative of nighttime chemistry. Additionally, nighttime biomass burning events were detected frequently, with durations ranging from 10 to 40 min, emphasizing the importance of high temporal resolution. During these events we found up to 30 isobaric peaks per unit mass that are baseline-resolved. For the first time, the hydroxypinonyl ester of cis-pinic acid (C<span class="inline-formula">₁₉</span>H<span class="inline-formula">₂₈</span>O<span class="inline-formula">₇</span>) could be measured and confirmed with MS<span class="inline-formula">²</span> experiments in ambient aerosol by an in-situ method at CR. In addition, laboratory experiments were performed to confirm the broad applicability of the APCI-Orbitrap-MS for the real-time detection of biogenic and biomass burning tracers, as well as specific anthropogenic pollutants, such as pesticides, organophosphates or organic esters from aircraft lubrication oil.</p>

Oscar Montiel Ross

This paper develops the foundations of Quantum Granular Computing (QGC), extending classical granular computing including fuzzy, rough, and shadowed granules to the quantum regime. Quantum granules are modeled as effects on a finite dimensional Hilbert space, so granular memberships are given by Born probabilities. This operator theoretic viewpoint provides a common language for sharp (projective) and soft (nonprojective) granules and embeds granulation directly into the standard formalism of quantum information theory. We establish foundational results for effect based quantum granules, including normalization and monotonicity properties, the emergence of Boolean islands from commuting families, granular refinement under Luders updates, and the evolution of granules under quantum channels via the adjoint channel in the Heisenberg picture. We connect QGC with quantum detection and estimation theory by interpreting the effect operators realizing Helstrom minimum error measurement for binary state discrimination as Helstrom type decision granules, i.e., soft quantum counterparts of Bayes optimal decision regions. Building on these results, we introduce Quantum Granular Decision Systems (QGDS) with three reference architectures that specify how quantum granules can be defined, learned, and integrated with classical components while remaining compatible with near term quantum hardware. Case studies on qubit granulation, two qubit parity effects, and Helstrom style soft decisions illustrate how QGC reproduces fuzzy like graded memberships and smooth decision boundaries while exploiting noncommutativity, contextuality, and entanglement. The framework thus provides a unified and mathematically grounded basis for operator valued granules in quantum information processing, granular reasoning, and intelligent systems.

Jianglin Lu, Hailing Wang, Yi Xu et al.

Foundation models learn highly transferable representations through large-scale pretraining on diverse data. An increasing body of research indicates that these representations exhibit a remarkable degree of similarity across architectures and modalities. In this survey, we investigate the representation potentials of foundation models, defined as the latent capacity of their learned representations to capture task-specific information within a single modality while also providing a transferable basis for alignment and unification across modalities. We begin by reviewing representative foundation models and the key metrics that make alignment measurable. We then synthesize empirical evidence of representation potentials from studies in vision, language, speech, multimodality, and neuroscience. The evidence suggests that foundation models often exhibit structural regularities and semantic consistencies in their representation spaces, positioning them as strong candidates for cross-modal transfer and alignment. We further analyze the key factors that foster representation potentials, discuss open questions, and highlight potential challenges.

Ryan Williams

This paper explores the hard problem of consciousness from a different perspective. Instead of drawing distinctions between the physical and the mental, an exploration of a more foundational relationship is examined: the relationship between structure and quality. Information-theoretic measures are developed to quantify the mutual determinability between structure and quality, including a novel Q-S space for analyzing fidelity between the two domains. This novel space naturally points toward a five-fold categorization of possible relationships between structural and qualitative properties, illustrating each through conceptual and formal models. The ontological implications of each category are examined, shedding light on debates around functionalism, emergentism, idealism, panpsychism, and neutral monism. This new line of inquiry has established a framework for deriving theoretical constraints on qualitative systems undergoing evolution that is explored in my companion paper, Qualia & Natural Selection.

Sompote Youwai, David Phim, Vianne Gayl Murcia et al.

This study investigates router-based multi-agent systems for automating foundation design calculations through intelligent task classification and expert selection. Three approaches were evaluated: single-agent processing, multi-agent designer-checker architecture, and router-based expert selection. Performance assessment utilized baseline models including DeepSeek R1, ChatGPT 4 Turbo, Grok 3, and Gemini 2.5 Pro across shallow foundation and pile design scenarios. The router-based configuration achieved performance scores of 95.00% for shallow foundations and 90.63% for pile design, representing improvements of 8.75 and 3.13 percentage points over standalone Grok 3 performance respectively. The system outperformed conventional agentic workflows by 10.0 to 43.75 percentage points. Grok 3 demonstrated superior standalone performance without external computational tools, indicating advances in direct LLM mathematical reasoning for engineering applications. The dual-tier classification framework successfully distinguished foundation types, enabling appropriate analytical approaches. Results establish router-based multi-agent systems as optimal for foundation design automation while maintaining professional documentation standards. Given safety-critical requirements in civil engineering, continued human oversight remains essential, positioning these systems as advanced computational assistance tools rather than autonomous design replacements in professional practice.

Özgün Turgut, Felix S. Bott, Markus Ploner et al.

The success of foundation models in natural language processing and computer vision has motivated similar approaches in time series analysis. While foundational time series models have proven beneficial on a variety of tasks, their effectiveness in medical applications with limited data remains underexplored. In this work, we investigate this question in the context of electroencephalography (EEG) by evaluating general-purpose time series models on age prediction, seizure detection, and classification of clinically relevant EEG events. We compare their diagnostic performance against specialised EEG models and assess the quality of the extracted features. The results show that general-purpose models are competitive and capture features useful to localising demographic and disease-related biomarkers. These findings indicate that foundational time series models can reduce the reliance on large task-specific datasets and models, making them valuable in clinical practice.

Sana Tonekaboni, Lena Stempfle, Adibvafa Fallahpour et al.

Foundation models trained on large-scale de-identified electronic health records (EHRs) hold promise for clinical applications. However, their capacity to memorize patient information raises important privacy concerns. In this work, we introduce a suite of black-box evaluation tests to assess privacy-related memorization risks in foundation models trained on structured EHR data. Our framework includes methods for probing memorization at both the embedding and generative levels, and aims to distinguish between model generalization and harmful memorization in clinically relevant settings. We contextualize memorization in terms of its potential to compromise patient privacy, particularly for vulnerable subgroups. We validate our approach on a publicly available EHR foundation model and release an open-source toolkit to facilitate reproducible and collaborative privacy assessments in healthcare AI.

Ali Movaghar

We introduce Controlled Stochastic Activity Networks (Controlled SANs), a formal extension of classical Stochastic Activity Networks that integrates explicit control actions into a unified semantic framework for modeling distributed real-time systems. Controlled SANs systematically capture dynamic behavior involving nondeterminism, probabilistic branching, and stochastic timing, while enabling policy-driven decision-making within a rigorous mathematical framework. We develop a hierarchical, automata-theoretic semantics for Controlled SANs that encompasses nondeterministic, probabilistic, and stochastic models in a uniform manner. A structured taxonomy of control policies, ranging from memoryless and finite-memory strategies to computationally augmented policies, is formalized, and their expressive power is characterized through associated language classes. To support model abstraction and compositional reasoning, we introduce behavioral equivalences, including bisimulation and stochastic isomorphism. Controlled SANs generalize classical frameworks such as continuous-time Markov decision processes (CTMDPs), providing a rigorous foundation for the specification, verification, and synthesis of dependable systems operating under uncertainty. This framework enables both quantitative and qualitative analysis, advancing the design of safety-critical systems where control, timing, and stochasticity are tightly coupled.

S. Kim, Y. Cho, H. Ki et al.

<p>This study presents advancements in the processing of satellite remote sensing data, focusing mainly on aerosol optical depth (AOD) retrievals from the Geostationary Environment Monitoring Spectrometer (GEMS). The transformation of Level-2 (L2) data, which includes atmospheric-state retrievals, into higher-quality Level-3 (L3) data is crucial in remote sensing. Our contributions lie in two novel improvements to the processing algorithm. First, we improve the inverse-distance-weighting algorithm by incorporating quality flag information into the weight calculation. By assigning weights that are inversely proportional to the number of unreliable grids, the method can provide more accurate L3 products. We validate this approach through simulation studies and apply it to GEMS AOD data across various regions and wavelengths. The use of quality flags in the algorithm can provide a more accurate analysis of remote sensing. Second, we employ a spatiotemporal merging method to address both spatial and temporal variability in AOD data, a departure from previous approaches that solely focused on spatial variability. Our method considers temporal variations spanning previous time intervals. Furthermore, the computed mean fields show similar spatiotemporal patterns to previous studies, confirming their ability to capture real-world phenomena. Lastly, utilizing this procedure, we compute the mean field estimates for GEMS AOD data, which can provide a deeper understanding of the impact of aerosols on climate change and public health.</p>

A. Naseri, J. C. Corbin, J. S. Olfert

<p>It is necessary to measure the mixing states of light-absorbing carbon (LAC) particles to reduce uncertainties in climate forcing due to particulate from wildfires and biomass combustion. For refractory LAC (normally called refractory black carbon; rBC), such measurements can be made using the single particle soot photometer (SP2). The SP2 measures the incandescent mass of individual particles heated by a 1064 nm laser. The SP2 also monitors single-particle light scattering from rBC plus internally mixed material (e.g., <i>coatings</i> of volatile particulate matter). rBC mixing states can be estimated from SP2 measurements by combining the scattering and incandescence signals. This is the basis of the published methods known as (i) scattering–incandescence lag-time, (ii) leading-edge only (LEO), and (iii) normalized derivative methods. More recently, the tandem centrifugal particle mass analyzer–single particle soot photometer (CPMA-SP2) method has been developed. The CPMA-SP2 method does not rely on the SP2 scattering signals and, therefore truly measures the rBC mass fraction, with no assumptions regarding particle composition or morphology. In this study, we provide the first quantitative comparison of the light-scattering and CPMA-SP2 methods for measuring mixing state. We discuss the upper and lower limits of detection (in terms of both rBC and coatings), temporal resolution, role of counting statistics, and errors associated with the measurements. We use a data set of atmospheric particles sampled at a regional background site (Kamloops about 350 km northeast of Vancouver, British Columbia, Canada), where the majority of rBC was emitted by seasonal wildfires. In the overall comparison of measurement methods, the CPMA-SP2 method is found to have significantly better systematic uncertainties than the light-scattering methods for wildfire smoke. For example, the light-scattering methods could not quantify coatings on half of the rBC particles, because their light-scattering signals were below the SP2 detection limit. Consequently, the bias in SP2-only estimates of rBC mixing states depends on the size distribution of the rBC particles. Although more accurate, CPMA-SP2 measurements require significantly more time to acquire, whereas SP2-only light-scattering analyses (both LEO and lag-time) can provide near real-time qualitative information representing large rBC particles.</p>

E. Y. Li, A. Yazdani, A. M. Dillner et al.

<p>Globally, billions of people burn fuels indoors for cooking and heating, which contributes to millions of chronic illnesses and premature deaths annually. Additionally, residential burning contributes significantly to black carbon emissions, which have the highest global warming impacts after carbon dioxide and methane. In this study, we use Fourier transform infrared spectroscopy (FTIR) to analyze fine-particulate emissions collected on Teflon membrane filters from 15 cookstove types and 5 fuel types. Emissions from three fuel types (charcoal, kerosene, and red oak wood) were found to have enough FTIR spectral response for functional group (FG) analysis. We present distinct spectral profiles for particulate emissions of these three fuel types. We highlight the influential FGs constituting organic carbon (OC) using a multivariate statistical method and show that OC estimates by collocated FTIR and thermal–optical transmittance (TOT) are highly correlated, with a coefficient determination of 82.5 %. As FTIR analysis is fast and non-destructive and provides complementary FG information, the analysis method demonstrated herein can substantially reduce the need for thermal–optical measurements for source emissions.</p>

Volodymyr Volianiuk, Dmytro Mishchuk, Eugene Gorbatyuk

Loading and unloading works are an integral part of the construction process. Cranes of various types were mostly use to perform these works. To ensure trouble-free operation and increase the reliability of cranes, when calculating structures and components of their working equipment, it is important to take into account dynamic loads, which are several times higher than static loads. Elements of dynamic loads in the crane suspension are its elastic components (flexible traction bodies) - ropes. The process of lifting a load from a rigid base and picking it up is considered, which divided into three stages: the first is the selection of clearances and the tension of the ropes; the second is the pre-opening stage of lifting the load; the third is the post-detachment stage of lifting the load. For each stage, the initial conditions accepted, the differential equations of the movement of loads compiled, their solution given taking into account many factors, and expressions derived for determining the forces in the load suspension. At the first stage, the duration of the gap selection (tension of the ropes) is determined, at the second stage, the speed of separation of the load from the base is determined, at the third stage, the maximum force in the elastic element determined. The method of determining the forces in the suspension of the load, the duration of the selection of clearances (tension of the ropes), the speed of separation of the load from the base, and the maximum force in the elastic element presented in the work allows you to significantly simplify the solution of complex equations, to determine simple expressions and to determine them with sufficient accuracy for practical calculations values.

Yuhang Zhou, Haolin Li, Siyuan Du et al.

The popularity of large-scale pre-training has promoted the development of medical foundation models. However, some studies have shown that although foundation models exhibit strong general feature extraction capabilities, their performance on specific tasks is still inferior to task-specific methods. In this paper, we explore a new perspective called ``Knowledge Decomposition'' to improve the performance on specific medical tasks, which deconstruct the foundation model into multiple lightweight expert models, each dedicated to a particular task, with the goal of improving specialization while concurrently mitigating resource expenditure. To accomplish the above objective, we design a novel framework named Low-Rank Knowledge Decomposition (LoRKD), which explicitly separates graidents by incorporating low-rank expert modules and the efficient knowledge separation convolution. Extensive experimental results demonstrate that the decomposed models perform well in terms of performance and transferability, even surpassing the original foundation models.

Jafar Isbarov, Kavsar Huseynova, Elvin Mammadov et al.

The emergence of multilingual large language models has enabled the development of language understanding and generation systems in Azerbaijani. However, most of the production-grade systems rely on cloud solutions, such as GPT-4. While there have been several attempts to develop open foundation models for Azerbaijani, these works have not found their way into common use due to a lack of systemic benchmarking. This paper encompasses several lines of work that promote open-source foundation models for Azerbaijani. We introduce (1) a large text corpus for Azerbaijani, (2) a family of encoder-only language models trained on this dataset, (3) labeled datasets for evaluating these models, and (4) extensive evaluation that covers all major open-source models with Azerbaijani support.

Peter Neidlinger, Omar S. M. El Nahhas, Hannah Sophie Muti et al.

Advancements in artificial intelligence have driven the development of numerous pathology foundation models capable of extracting clinically relevant information. However, there is currently limited literature independently evaluating these foundation models on truly external cohorts and clinically-relevant tasks to uncover adjustments for future improvements. In this study, we benchmarked 19 histopathology foundation models on 13 patient cohorts with 6,818 patients and 9,528 slides from lung, colorectal, gastric, and breast cancers. The models were evaluated on weakly-supervised tasks related to biomarkers, morphological properties, and prognostic outcomes. We show that a vision-language foundation model, CONCH, yielded the highest performance when compared to vision-only foundation models, with Virchow2 as close second. The experiments reveal that foundation models trained on distinct cohorts learn complementary features to predict the same label, and can be fused to outperform the current state of the art. An ensemble combining CONCH and Virchow2 predictions outperformed individual models in 55% of tasks, leveraging their complementary strengths in classification scenarios. Moreover, our findings suggest that data diversity outweighs data volume for foundation models. Our work highlights actionable adjustments to improve pathology foundation models.

Chenyang Lei, Liyi Chen, Jun Cen et al.

Foundation models like ChatGPT and Sora that are trained on a huge scale of data have made a revolutionary social impact. However, it is extremely challenging for sensors in many different fields to collect similar scales of natural images to train strong foundation models. To this end, this work presents a simple and effective framework SimMAT to study an open problem: the transferability from vision foundation models trained on natural RGB images to other image modalities of different physical properties (e.g., polarization). SimMAT consists of a modality-agnostic transfer layer (MAT) and a pretrained foundation model. We apply SimMAT to a representative vision foundation model Segment Anything Model (SAM) to support any evaluated new image modality. Given the absence of relevant benchmarks, we construct a new benchmark to evaluate the transfer learning performance. Our experiments confirm the intriguing potential of transferring vision foundation models in enhancing other sensors' performance. Specifically, SimMAT can improve the segmentation performance (mIoU) from 22.15% to 53.88% on average for evaluated modalities and consistently outperforms other baselines. We hope that SimMAT can raise awareness of cross-modal transfer learning and benefit various fields for better results with vision foundation models.