Agentic Al systems are increasingly deployed as personal assistants and are likely to become a common object of digital investigations. However, little is known about how their internal state and actions can be reconstructed during forensic analysis. Despite growing popularity, systematic forensic approaches for such systems remain largely unexplored. This paper presents an empirical study of OpenClaw a widely used single-agent assistant. We examine OpenClaw's technical design via static code analysis and apply differential forensic analysis to identify recoverable traces across stages of the agent interaction loop. We classify and correlate these traces to assess their investigative value in a systematic way. Based on these observations, we propose an agent artifact taxonomy that captures recurring investigative patterns. Finally, we highlight a foundational challenge for agentic Al forensics: agent-mediated execution introduces an additional layer of abstraction and substantial nondeterminism in trace generation. The large language model (LLM), the execution environment, and the evolving context can influence tool choice and state transitions in ways that are largely absent from rule-based software. Overall, our results provide an initial foundation for the systematic investigation of agentic Al and outline implications for digital forensic practice and future research.
Cell segmentation is a fundamental task in microscopy image analysis. Several foundation models for cell segmentation have been introduced, virtually all of them are extensions of Segment Anything Model (SAM), improving it for microscopy data. Recently, SAM2 and SAM3 have been published, further improving and extending the capabilities of general-purpose segmentation foundation models. Here, we comprehensively evaluate foundation models for cell segmentation (CellPoseSAM, CellSAM, $μ$SAM) and for general-purpose segmentation (SAM, SAM2, SAM3) on a diverse set of (light) microscopy datasets, for tasks including cell, nucleus and organoid segmentation. Furthermore, we introduce a new instance segmentation strategy called automatic prompt generation (APG) that can be used to further improve SAM-based microscopy foundation models. APG consistently improves segmentation results for $μ$SAM, which is used as the base model, and is competitive with the state-of-the-art model CellPoseSAM. Moreover, our work provides important lessons for adaptation strategies of SAM-style models to microscopy and provides a strategy for creating even more powerful microscopy foundation models. Our code is publicly available at https://github.com/computational-cell-analytics/micro-sam.
This article surveys key conceptual and interpretational developments in quantum mechanics, tracing the theory from its foundational postulates to contemporary discussions of measurement, nonlocality, and the emergence of classicality. Beginning with the structure of Hilbert space and the postulates governing state evolution and measurement, the epistemic stance of the Copenhagen interpretation and its modern reformulations are examined. The Einstein-Podolsky-Rosen argument, Bell's theorem, and Hardy's paradox are then discussed as probes of locality and realism, alongside the deterministic but explicitly nonlocal de Broglie-Bohm theory. The measurement problem and the implications of contextuality are analyzed in relation to objective collapse models, which introduce new physical dynamics to account for definite outcomes. Finally, the role of decoherence in the suppression of interference and the emergence of classical behavior is explored, together with the interpretational frameworks of many-worlds and consistent histories. This material aims to provide a coherent introductory overview of how different interpretations address the central concern of what quantum mechanics tells us about the nature of physical reality.
This study explores computational approaches for measuring moral foundations (MFs) in non-English corpora. Since most resources are developed primarily for English, cross-linguistic applications of moral foundation theory remain limited. Using Chinese as a case study, this paper evaluates the effectiveness of applying English resources to machine translated text, local language lexicons, multilingual language models, and large language models (LLMs) in measuring MFs in non-English texts. The results indicate that machine translation and local lexicon approaches are insufficient for complex moral assessments, frequently resulting in a substantial loss of cultural information. In contrast, multilingual models and LLMs demonstrate reliable cross-language performance with transfer learning, with LLMs excelling in terms of data efficiency. Importantly, this study also underscores the need for human-in-the-loop validation of automated MF assessment, as the most advanced models may overlook cultural nuances in cross-language measurements. The findings highlight the potential of LLMs for cross-language MF measurements and other complex multilingual deductive coding tasks.
Adilet Metinov, Gulida M. Kudakeeva, Gulnara D. Kabaeva
Current AI code generation systems suffer from significant latency bottlenecks due to CPU-GPU data transfers during compilation, execution, and testing phases. We establish theoretical foundations for three complementary approaches to GPU-native compilation that eliminate these transfers: (1) parallel traditional compilation adapted for GPU execution, (2) neural compilation using learned sequence-to-sequence translation with probabilistic verification, and (3) hybrid architectures combining both strategies. We derive latency and energy bounds demonstrating potential speedups of 10-100x for code iteration cycles. Our analysis shows that traditional GPU compilation provides 2-5x improvements through transfer elimination, neural compilation achieves 10-100x speedups via massive parallelism, and hybrid approaches offer practical deployment paths with guaranteed correctness. We formalize the probabilistic verification framework that enables trading compilation accuracy for parallel exploration, and discuss implications for self-improving AI systems and future analog computing substrates.
Developing a single foundation model with the capability to excel across diverse tasks has been a long-standing objective in the field of artificial intelligence. As the wave of general-purpose foundation models sweeps across various domains, their influence has significantly extended to the field of recommendation systems. While recent efforts have explored recommendation foundation models for various generative tasks, they often overlook crucial embedding tasks and struggle with the complexities of multi-task learning, including knowledge sharing & conflict resolution, and convergence speed inconsistencies. To address these limitations, we introduce RecFound, a generative representational learning framework for recommendation foundation models. We construct the first comprehensive dataset for recommendation foundation models covering both generative and embedding tasks across diverse scenarios. Based on this dataset, we propose a novel multi-task training scheme featuring a Task-wise Mixture of Low-rank Experts (TMoLE) to handle knowledge sharing & conflict, a Step-wise Convergence-oriented Sample Scheduler (S2Sched) to address inconsistent convergence, and a Model Merge module to balance the performance across tasks. Experiments demonstrate that RecFound achieves state-of-the-art performance across various recommendation tasks, outperforming existing baselines.
<p>Cloud fraction (CF) is an integral aspect of weather and radiation forecasting, but real time monitoring of CF is still inaccurate, expensive and exclusive to commercial sky imagers. Traditional cloud segmentation methods, which often rely on empirically determined threshold values, struggle under complex atmospheric and cloud conditions. This study investigates the use of a Random Forest (RF) classifier for pixel-wise cloud segmentation using a dataset of semantically annotated images from five geographically diverse locations. The RF model was trained on diverse sky conditions and atmospheric loads, ensuring robust performance across varied environments. The accuracy score was always above 85 % for all the locations along with similarly high F1 score and Receiver Operating Characteristic – Area Under the Curve (ROC-AUC) score establishing the efficiency of the model. Validation experiments conducted at three Atmospheric Radiation Measurement (ARM) sites and two Indian locations, including Gadanki and Merak, demonstrated that the RF classifier outperformed conventional Total Sky Imager (TSI) methods, particularly in high-pollution areas. The model effectively captured long-term weather and cloud patterns, exhibiting strong location-agnostic performance. However, challenges in distinguishing sun glares and cirrus clouds due to annotation limitations were noted. Despite these minor issues, the RF classifier shows significant promise for accurate and adaptable cloud cover estimation, making it a valuable tool in climate studies.</p>
<p>High Spectral Resolution Lidars (HSRLs) have been successfully deployed from a variety of platforms: ground based, airborne, and now satellite. These lidars are uniquely valuable for characterizing atmospheric aerosol and clouds, benefiting from the ability to characterize vertical variability in more detail than any passive instruments, and, compared to elastic backscatter lidars, provide additional channels of measurements that permit the direct retrieval of particulate extinction. Although analytic solutions exist for deriving particulate backscatter, extinction, and linear depolarization ratio, in the case of extinction, the analytic technique greatly magnifies measurement noise. Low signal-to-noise measurements stress the traditional inversion methods. Accordingly, algorithms for the retrieval of HSRL backscatter and extinction are re-examined and optimized to reduce the noise propagation. Here we explore an Optimal Estimation methodology and compare it with an implementation of the direct differentiation method like that historically used for the processing of airborne HSRL data from NASA Langley Research Center.</p>
<p>We have developed an intermediate size (906 L) aerosol processing chamber, and this work reports on the design and initial characterization of dry aerosol experiments. Specifically, we are determining wall-loss and coagulation correction factors using the observed size distribution measurements for surrogates of common aerosol classes: sodium chloride, sucrose, and biomass burning aerosol smoke. Results show that, on average, sodium chloride, sucrose, and smoke wall-loss rates converge to similar values on relatively short time scales (<span class="inline-formula"><i><</i></span> 1 h). The fitted coagulation correction factor, <span class="inline-formula"><math xmlns="http://www.w3.org/1998/Math/MathML" id="M2" display="inline" overflow="scroll" dspmath="mathml"><mrow><msubsup><mi>W</mi><mi mathvariant="normal">C</mi><mrow><mo>-</mo><mn mathvariant="normal">1</mn></mrow></msubsup></mrow></math><span><svg:svg xmlns:svg="http://www.w3.org/2000/svg" width="23pt" height="17pt" class="svg-formula" dspmath="mathimg" md5hash="95e0043826b3994938eca2cc433f4b7a"><svg:image xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="amt-18-5705-2025-ie00001.svg" width="23pt" height="17pt" src="amt-18-5705-2025-ie00001.png"/></svg:svg></span></span>, for smoke particles (1.23 <span class="inline-formula">±</span> 0.312), indicates that on average they adhere to each other more than sodium chloride (0.969 <span class="inline-formula">±</span> 0.524) and sucrose (1.16 <span class="inline-formula">±</span> 1.38). The relative uncertainty is high for the coagulation correction, but it is consistent with our Monte Carlo error analysis. This study lays the foundation for future experiments at elevated humidity and supersaturation conditions to characterize the influence of particle shape on coagulation and cloud parameters.</p>
Using Ni(II) as the catalyst, electron-deficient 3,5-dimethylacryloylpyrazole olefin was reacted with C,N-diarylnitrones alone for 10 min to prepare novel five-member heterocyclic products, 4-3,5-dimethylacryloylpyrazole isoxazolidines with 100% regioselectivity and up to 99% yield. And then, taking these cycloadducts as substrates, six kinds of derivatization reactions, like ring-opening, nucleophilic substitution, addition-elimination and reduction, were studied. Experimental results showed that all kinds of transformations could obtain the target products at a high conversion rate under mild conditions, a finding which provided the basic methods for organic synthesis methodology research based on an isoxazolidine skeleton.
The rise of foundation models marks a paradigm shift in machine learning: instead of training specialized models from scratch, foundation models are first trained on massive datasets before being adapted or fine-tuned to make predictions on smaller datasets. Initially developed for text, foundation models have also excelled at making predictions about social science data. However, while many estimation problems in the social sciences use prediction as an intermediate step, they ultimately require different criteria for success. In this paper, we develop methods for fine-tuning foundation models to perform these estimation problems. We first characterize an omitted variable bias that can arise when a foundation model is only fine-tuned to maximize predictive accuracy. We then provide a novel set of conditions for fine-tuning under which estimates derived from a foundation model are root-n-consistent. Based on this theory, we develop new fine-tuning algorithms that empirically mitigate this omitted variable bias. To demonstrate our ideas, we study gender wage decomposition. This is a statistical estimation problem from econometrics where the goal is to decompose the gender wage gap into components that can and cannot be explained by career histories of workers. Classical methods for decomposing the wage gap employ simple predictive models of wages which condition on coarse summaries of career history that may omit factors that are important for explaining the gap. Instead, we use a custom-built foundation model to decompose the gender wage gap, which captures a richer representation of career history. Using data from the Panel Study of Income Dynamics, we find that career history explains more of the gender wage gap than standard econometric models can measure, and we identify elements of career history that are omitted by standard models but are important for explaining the wage gap.
By treating data and models as the source code, Foundation Models (FMs) become a new type of software. Mirroring the concept of software crisis, the increasing complexity of FMs making FM crisis a tangible concern in the coming decade, appealing for new theories and methodologies from the field of software engineering. In this paper, we outline our vision of introducing Foundation Model (FM) engineering, a strategic response to the anticipated FM crisis with principled engineering methodologies. FM engineering aims to mitigate potential issues in FM development and application through the introduction of declarative, automated, and unified programming interfaces for both data and model management, reducing the complexities involved in working with FMs by providing a more structured and intuitive process for developers. Through the establishment of FM engineering, we aim to provide a robust, automated, and extensible framework that addresses the imminent challenges, and discovering new research opportunities for the software engineering field.
Siqi Lu, Junlin Guo, James R Zimmer-Dauphinee
et al.
Artificial Intelligence (AI) technologies have profoundly transformed the field of remote sensing, revolutionizing data collection, processing, and analysis. Traditionally reliant on manual interpretation and task-specific models, remote sensing research has been significantly enhanced by the advent of foundation models-large-scale, pre-trained AI models capable of performing a wide array of tasks with unprecedented accuracy and efficiency. This paper provides a comprehensive survey of foundation models in the remote sensing domain. We categorize these models based on their architectures, pre-training datasets, and methodologies. Through detailed performance comparisons, we highlight emerging trends and the significant advancements achieved by those foundation models. Additionally, we discuss technical challenges, practical implications, and future research directions, addressing the need for high-quality data, computational resources, and improved model generalization. Our research also finds that pre-training methods, particularly self-supervised learning techniques like contrastive learning and masked autoencoders, remarkably enhance the performance and robustness of foundation models. This survey aims to serve as a resource for researchers and practitioners by providing a panorama of advances and promising pathways for continued development and application of foundation models in remote sensing.
Emir Konuk, Christos Matsoukas, Moein Sorkhei
et al.
We introduce Learning from Offline Foundation Features with Tensor Augmentations (LOFF-TA), an efficient training scheme designed to harness the capabilities of foundation models in limited resource settings where their direct development is not feasible. LOFF-TA involves training a compact classifier on cached feature embeddings from a frozen foundation model, resulting in up to $37\times$ faster training and up to $26\times$ reduced GPU memory usage. Because the embeddings of augmented images would be too numerous to store, yet the augmentation process is essential for training, we propose to apply tensor augmentations to the cached embeddings of the original non-augmented images. LOFF-TA makes it possible to leverage the power of foundation models, regardless of their size, in settings with limited computational capacity. Moreover, LOFF-TA can be used to apply foundation models to high-resolution images without increasing compute. In certain scenarios, we find that training with LOFF-TA yields better results than directly fine-tuning the foundation model.
D. J. V. Robbins, D. J. V. Robbins, C. A. Poulsen
et al.
<p><span id="page3280"/>Extreme biomass burning (BB) events, such as those seen during the 2019–2020 Australian bushfire season, are becoming more frequent and intense with climate change. Ground-based observations of these events can provide useful information on the macro- and micro-physical properties of the plumes, but these observations are sparse, especially in regions which are at risk of intense bushfire events. Satellite observations of extreme BB events provide a unique perspective, with the newest generation of geostationary imagers, such as the Advanced Himawari Imager (AHI), observing entire continents at moderate spatial and high temporal resolution. However, current passive satellite retrieval methods struggle to capture the high values of aerosol optical thickness (AOT) seen during these BB events. Accurate retrievals are necessary for global and regional studies of shortwave radiation, air quality modelling and numerical weather prediction. To address these issues, the Optimal Retrieval of Aerosol and Cloud (ORAC) algorithm has used AHI data to measure extreme BB plumes from the 2019–2020 Australian bushfire season. The sensitivity of the retrieval to the assumed optical properties of BB plumes is explored by comparing retrieved AOT with AErosol RObotic NETwork (AERONET) level-1.5 data over the AERONET site at Tumbarumba, New South Wales, between 1 December 2019 at 00:00 UTC and 3 January 2020 at 00:00 UTC. The study shows that for AOT values <span class="inline-formula">></span> 2, the sensitivity to the assumed optical properties is substantial. The ORAC retrievals and AERONET data are compared against the Japan Aerospace Exploration Agency (JAXA) Aerosol Retrieval Product (ARP), Moderate Resolution Imaging Spectroradiometer (MODIS) Deep Blue over land, MODIS MAIAC, Sentinel-3 SYN and VIIRS Deep Blue products. The comparison shows the ORAC retrieval significantly improves coverage of optically thick plumes relative to the JAXA ARP, with approximately twice as many pixels retrieved and peak retrieved AOT values 1.4 times higher than the JAXA ARP. The ORAC retrievals have accuracy scores of 0.742–0.744 compared to the values of 0.718–0.833 for the polar-orbiting satellite products, despite successfully retrieving approximately 28 times as many pixels over the study period as the most successful polar-orbiting satellite product. The AHI and MODIS satellite products are compared for three case studies covering a range of BB plumes over Australia. The results show good agreement between all products for plumes with AOT values <span class="inline-formula">≤</span> 2. For extreme BB plumes, the ORAC retrieval finds values of AOT <span class="inline-formula">></span> 15, significantly higher than those seen in events classified as extreme by previous studies, although with high uncertainty. A combination of hard limits in the retrieval algorithms and misclassification of BB plumes as cloud prevents the JAXA and MODIS products from returning AOT values significantly greater than 5.</p>
The Wentzel-Kramers-Brillouin (WKB) approximation is frequently used to explore the mechanics of the cochlea. As opposed to numerical strategies, the WKB approximation facilitates analysis of model results through interpretable closed-form equations, and can be implemented with relative ease. As a result, it has maintained relevance in the study of cochlear mechanics for half of a century. Over this time, it has been used to study a variety of phenomena including the limits of frequency tuning, active displacement amplification within the organ of Corti, feedforward mechanisms in the cochlea, and otoacoustic emissions. Despite this ubiquity, it is challenging to find rigorous exposition of the WKB approximation's formulation, derivation and implementation in cochlear mechanics literature. In this tutorial, I discuss the foundations of the WKB approximation in application to models of cochlear macromechanics in 1-D and 2-D. This includes mathematical background, rigorous derivation and details of its implementation in software.
Why are some research studies easy to reproduce while others are difficult? Casting doubt on the accuracy of scientific work is not fruitful, especially when an individual researcher cannot reproduce the claims made in the paper. There could be many subjective reasons behind the inability to reproduce a scientific paper. The field of Machine Learning (ML) faces a reproducibility crisis, and surveying a portion of published articles has resulted in a group realization that although sharing code repositories would be appreciable, code bases are not the end all be all for determining the reproducibility of an article. Various parties involved in the publication process have come forward to address the reproducibility crisis and solutions such as badging articles as reproducible, reproducibility checklists at conferences (\textit{NeurIPS, ICML, ICLR, etc.}), and sharing artifacts on \textit{OpenReview} come across as promising solutions to the core problem. The breadth of literature on reproducibility focuses on measures required to avoid ir-reproducibility, and there is not much research into the effort behind reproducing these articles. In this paper, we investigate the factors that contribute to the easiness and difficulty of reproducing previously published studies and report on the foundational framework to quantify effort of reproducibility.
Discovering causal relationships from observational data is a challenging task that relies on assumptions connecting statistical quantities to graphical or algebraic causal models. In this work, we focus on widely employed assumptions for causal discovery when objects of interest are (multivariate) groups of random variables rather than individual (univariate) random variables, as is the case in a variety of problems in scientific domains such as climate science or neuroscience. If the group-level causal models are derived from partitioning a micro-level model into groups, we explore the relationship between micro and group-level causal discovery assumptions. We investigate the conditions under which assumptions like Causal Faithfulness hold or fail to hold. Our analysis encompasses graphical causal models that contain cycles and bidirected edges. We also discuss grouped time series causal graphs and variants thereof as special cases of our general theoretical framework. Thereby, we aim to provide researchers with a solid theoretical foundation for the development and application of causal discovery methods for variable groups.
<p>The Limb Imaging Fourier Transform Spectrometer Experiment (LIFE) instrument is a balloon-borne prototype of a satellite instrument designed to take vertical images of atmospheric limb emission spectra in the 700–1400 <span class="inline-formula">cm</span><span class="inline-formula"><sup>−1</sup></span> wavenumber range from the upper-troposphere–lower-stratosphere (UTLS) altitude region of the atmosphere. The prototype builds on the success of past and existing instruments while reducing the complexity of the imaging design. This paper details the results of a demonstration flight on a stabilized stratospheric balloon gondola from Timmins, Canada, in August 2019. Retrievals of vertical trace gas profiles for the important greenhouse gases <span class="inline-formula">H<sub>2</sub>O</span>, <span class="inline-formula">O<sub>3</sub></span>, <span class="inline-formula">CH<sub>4</sub></span>, and <span class="inline-formula">N<sub>2</sub>O</span>, as well as <span class="inline-formula">HNO<sub>3</sub></span>, are performed using an optimal estimation approach and the SASKTRAN radiative transfer model. The retrieved profiles are compared to approximately coincident observations made by the Atmospheric Chemistry Experiment Fourier Transform Spectrometer (ACE-FTS) solar occultation and Microwave Limb Sounder (MLS) instruments. An evaluation of the LIFE measurements is performed, and areas of improvement are identified. This work increases the overall technical readiness of the approach for
future balloon, aircraft, and space applications.</p>
<p>The largest inorganic, gas-phase reservoir of chlorine
atoms in the atmosphere is hydrogen chloride (HCl), but challenges in
quantitative sampling of this compound cause difficulties for obtaining
high-quality, high-frequency measurements. In this work, tunable infrared
laser direct absorption spectroscopy (TILDAS) was demonstrated to be a
superior optical method for sensitive, in situ detection of HCl at the
2925.89645 cm<span class="inline-formula"><sup>−1</sup></span> absorption line using a 3 <span class="inline-formula">µ</span>m inter-band cascade
laser. The instrument has an effective path length of 204 m, 1 Hz precision
of 7–8 pptv, and 3<span class="inline-formula"><i>σ</i></span> limit of detection ranging from 21 to 24 pptv. For longer
averaging times, the highest precision obtained was 0.5 pptv with a 3<span class="inline-formula"><i>σ</i></span> limit
of detection of 1.6 pptv at 2.4 min. HCl-TILDAS was also shown to have
high accuracy when compared with a certified gas cylinder, yielding a linear
slope within the expected 5 % tolerance of the reported cylinder
concentration (slope <span class="inline-formula">=</span> 0.964 <span class="inline-formula">±</span> 0.008). The use of heated inlet lines
and active chemical passivation greatly improve the instrument response
times to changes in HCl mixing ratios, with minimum 90 % response times
ranging from 1.2 to 4.4 s depending on inlet flow rate. However, these
response times lengthened at relative humidities <span class="inline-formula">>50</span> %,
conditions under which HCl concentration standards were found to elicit a
significantly lower response (<span class="inline-formula">−5.8</span> %). The addition of high concentrations
of gas-phase nitric acid (<span class="inline-formula">>3.0</span> ppbv) were found to increase HCl
signal (<span class="inline-formula"><10</span> %), likely due to acid displacement with HCl or
particulate chloride adsorbed to inlet surfaces. The equilibrium model
ISORROPIA suggested a potential of particulate chloride partitioning into
HCl gas within the heated inlet system if allowed to thermally equilibrate,
but field results did not demonstrate a clear relationship between
particulate chloride and HCl signal obtained with a denuder installed on the inlet.</p>