Abstract
Many have observed that the development and deployment of generative machine learning (ML) and artificial intelligence (AI) models follow a distinctive pattern in which pre-trained models are adapted and fine-tuned for specific downstream tasks. However, there is limited empirical work that examines the structure of these interactions. This paper analyzes 1.86 million models on Hugging Face, a leading peer production platform for model development. Our study of model family trees -- networks that connect fine-tuned models to their base or parent -- reveals sprawling fine-tuning lineages that vary widely in size and structure. Using an evolutionary biology lens to study ML models, we use model metadata and model cards to measure the genetic similarity and mutation of traits over model families. We find that models tend to exhibit a family resemblance, meaning their genetic markers and traits exhibit more overlap when they belong to the same model family. However, these similarities depart in certain ways from standard models of asexual reproduction, because mutations are fast and directed, such that two `sibling' models tend to exhibit more similarity than parent/child pairs. Further analysis of the directional drifts of these mutations reveals qualitative insights about the open machine learning ecosystem: Licenses counter-intuitively drift from restrictive, commercial licenses towards permissive or copyleft licenses, often in violation of upstream license's terms; models evolve from multi-lingual compatibility towards english-only compatibility; and model cards reduce in length and standardize by turning, more often, to templates and automatically generated text. Overall, this work takes a step toward an empirically grounded understanding of model fine-tuning and suggests that ecological models and methods can yield novel scientific insights.

Abstract
A large body of research has employed Machine Learning (ML) models to develop learned operating systems (OSes) and kernels. The latter dynamically adapts to the job load and dynamically adjusts resources (CPU, IO, memory, network bandwidth) allocation to respond to the actual user demand. What this work has in common is that it utilizes ML to improve kernel decisions. To this day, and to the best of our knowledge, no work has taken the opposite direction, i.e., using OS to improve ML. While some work proposes applying system-level optimizations to ML algorithms, they do not tailor the OS to adapt to the ML context. To address this limitation, we take an orthogonal approach in this paper by leveraging the OS to enhance the performance of ML models and algorithms. We explore the path towards an ML-specialized OS, MaLV-OS. MaLV-OS rethinks the OS architecture to make it specifically tailored to ML workloads, especially in virtualized clouds, which are now widely used to run ML applications. MaLV-OS envisioned architecture includes (1) a micro-kernel, Micro-LAKE, which allows kernel space applications to use the GPU, and (2) an MLaaS (ML as a Service) subsystem that gathers ML models to help Micro-LAKE with memory management and CPU scheduling. MaLV-OS architecture also offloads system-sensitive parts of the models to the OS, to lighten the model complexity and programming, and speed up its execution. Finally, MaLV-OS integrates an open-source GPU virtualization software, merged directly into the hypervisor. For more flexibility, MaLV-OS vision is to enable the virtual machine to dynamically select MLaaS policies that can improve the performance of the model the user is running. Because MLaaS is designed as loadable kernel modules, the MaLV-OS architecture enables the dynamic addition of new capabilities to the MLaaS subsystem.

Abstract
Machine learning (ML) has rapidly grown in popularity, becoming vital to many industries. Currently, the research on code smells in ML applications lacks tools and studies that address the identification and validity of ML-specific code smells. This work investigates suitable methods and tools to design and develop a static code analysis tool (MLpylint) based on code smell criteria. This research employed the Design Science Methodology. In the problem identification phase, a literature review was conducted to identify ML-specific code smells. In solution design, a secondary literature review and consultations with experts were performed to select methods and tools for implementing the tool. We evaluated the tool on data from 160 open-source ML applications sourced from GitHub. We also conducted a static validation through an expert survey involving 15 ML professionals. The results indicate the effectiveness and usefulness of the MLpylint. We aim to extend our current approach by investigating ways to introduce MLpylint seamlessly into development workflows, fostering a more productive and innovative developer environment.

Abstract
Test-time scaling (TTS) has emerged as a promising research field for enhancing the effectiveness of large language models (LLMs) without extra training. However, most existing approaches, e.g., Best-of-N and Self-Consistency rely on a single agent interacting with a reward model (SA-SR), constrained by limited capabilities of a single test-time scaling (STTS) paradigm. On the other hand, recent works demonstrate that collective-agent methods can break through the upper bound of single-agent systems by orchestrating diverse models. Thus, in this paper, we take a first step towards exploring Collective Test-Time Scaling (CTTS). Consider the different interaction types of single and multiple models, we design three primary paradigms to investigate the optimal paradigm of CTTS: (1) single agent to multiple reward models (SA-MR); (2) multiple agents to single reward model (MA-SR); and (3) multiple agents to multiple reward models (MA-MR). Extensive experiments demonstrate that MA-MR consistently achieves the best performance. Based on this, we propose a novel framework named CTTS-MM that effectively leverages both multi-agent and multi-reward-model collaboration for enhanced inference. Specifically, for multi-agent collaboration, we propose an Agent Collaboration Search (ACS), which searches for the most effective combination of LLM agents from a large candidate pool; for multi-reward-model collaboration, we propose Mixture of Reword Models (MoR), which consists of a curated question pool and a Prior Reward model Ensemble Selection (PRES) to select the optimal combinations of reward models via Pair-wise Reward Ranking (PRR) metric. Experiments across seven mainstream benchmarks demonstrate that the proposed CTTS-MM consistently obtains superior performance. Code will be released at https://github.com/magent4aci/CTTS-MM.

Abstract
As the demand for jobs in data science increases, so does the demand for universities to develop and facilitate modernized data science curricula to train students for these positions. Yet, the development of these courses remains challenging, especially at the introductory level. To help instructors to meet this demand, we present a flexible blueprint that supports the development of a modernized introductory data science curriculum. This blueprint is narrated through the lens and experience in teaching the introductory data science course at \university{}. This is a large course that serves both STEM and non-STEM majors and includes the incorporation and facilitation of technologies such as R, RStudio, Quarto, Git, and GitHub. We identify and provide discussion around common challenges in teaching a modernized introductory data science course, detail a learning model for students to grow their understanding of data science concepts, and provide reproducible materials to help empower teachers to adopt and adapt such curriculum at their universities.

Abstract
As fault-tolerant quantum computers scale, certifying the accuracy of computations performed with encoded logical qubits will soon become classically intractable. This creates a critical need for scalable, device-independent certification methods. In this work, we introduce logical accreditation, a framework for efficiently certifying quantum computations performed on logical qubits. Our protocol is robust against general noise models, far beyond those typically considered in performance analyses of quantum error-correcting codes. Through numerical simulations, we demonstrate that logical accreditation can scalably certify quantum advantage experiments and indicate the crossover point where encoded computations begin to outperform physical computations. Logical accreditation can also find application in evaluating whether logical circuit error rates are sufficiently low that error mitigation can be efficiently performed, extending the entropy benchmarking method to the regime of fault-tolerant computation, and upper-bounding the infidelity of the logical output state. Underpinning the framework is a novel randomised compiling scheme that converts arbitrary logical circuit noise into stochastic Pauli noise. This scheme includes a method for twirling non-transversal logical gates beyond the standard T-gate, resolving an open problem posed by Piveteau et al. [Piveteau et al. PRL 127, 200505 (2001)]. By bridging fault-tolerant computation and computational certification, logical accreditation offers a practical tool to assess the quality of computations performed on quantum hardware using encoded logical qubits.

Abstract
Open source software (OSS) generates trillions of dollars in economic value and has become essential to technical infrastructures worldwide. As organizations increasingly depend on OSS, understanding project evolution is critical. While existing metrics provide insights into project health, one dimension remains understudied: project resilience -- the ability to return to normal operations after disturbances such as contributor departures, security vulnerabilities, and bug report spikes. We hypothesize that stable commit patterns reflect underlying project characteristics such as mature governance, sustained contributors, and robust development processes that enable resilience. Building on the Composite Stability Index (CSI) framework, we empirically validate commit frequency patterns across 100 highly ranked repositories. Our findings reveal that only 2\% of repositories exhibit daily stability, 29\% achieve weekly stability, and 50\% demonstrate monthly stability, while half remain unstable across all temporal levels. Programming languages and blockchain applications were the most stable. We identified two exemplary repositories that achieved stability at all three granularities, whose governance models, CI cadence, and release policies could serve as reference frameworks. We observed that large yearly commit throughput does not necessarily correlate with stability. Beyond commits, stability can be enriched with issue-resolution times, PR merge rates, and community-engagement metrics to broaden resilience assessment and sharpen stability-based risk evaluation.

Abstract
Generative Machine Learning models have become central to modern systems, powering applications in creative writing, summarization, multi-hop reasoning, and context-aware dialogue. These models underpin large-scale AI assistants, workflow automation, and autonomous decision-making. In such domains, acceptable response is rarely absolute or static, but plural and highly context-dependent. Yet standard evaluation regimes still rely on static, benchmark-style tests, incentivizing optimization toward leaderboard scores rather than alignment with dynamic user needs or evolving realities. GrandJury introduces a formal evaluation protocol combining time-decayed aggregation, complete traceability, with the support of dynamic, transparent task rubric attribution, and multi-rater human judgment. Together, these elements enable pluralistic, accountable evaluation that captures evolving consensus and surfaces disagreement. We provide an open-source implementation (grandjury PyPI package) and a public collection of Large Language Model (LLM) inference outputs to illustrate the need and method. GrandJury provides a new paradigm for AI practitioners when evaluating machine learning outputs without absolute ground truth.

Abstract
Systems like aircraft and spacecraft are expensive to operate in the real world. The design, validation, and testing for such systems therefore relies on a combination of mathematical modeling, abundant numerical simulations, and a relatively small set of real-world experiments. Due to modeling errors, simplifications, and uncertainties, the data synthesized by simulation models often does not match data from the system's real-world operation. We consider the broad research question of whether this model mismatch can be significantly reduced by generative artificial intelligence models (GAIMs). Unlike text- or image-processing, where generative models have attained recent successes, GAIM development for aerospace engineering applications must not only train with scarce operational data, but their outputs must also satisfy governing equations based on natural laws, e.g., conservation laws. The scope of this paper primarily focuses on two case studies of optimally controlled systems that are commonly understood and employed in aircraft guidance, namely: minimum-time navigation in a wind field and minimum-exposure navigation in a threat field. We report GAIMs that are trained with a relatively small set, of the order of a few hundred, of examples and with underlying governing equations. By focusing on optimally controlled systems, we formulate training loss functions based on invariance of the Hamiltonian function along system trajectories. We investigate three GAIM architectures, namely: the generative adversarial network (GAN) and two variants of the variational autoencoder (VAE). We provide architectural details and thorough performance analyses of these models. The main finding is that our new models, especially the VAE-based models, are able to synthesize data that satisfy the governing equations and are statistically similar to the training data despite small volumes of training data.

Abstract
Federated learning (FL) enables collaborative model training across distributed clients while preserving data privacy by keeping data local. Traditional FL approaches rely on a centralized, star-shaped topology, where a central server aggregates model updates from clients. However, this architecture introduces several limitations, including a single point of failure, limited personalization, and poor robustness to distribution shifts or vulnerability to malfunctioning clients. Moreover, update selection in centralized FL often relies on low-level parameter differences, which can be unreliable when client data is not independent and identically distributed, and offer clients little control. In this work, we propose a decentralized, peer-to-peer (P2P) FL framework. It leverages the flexibility of the P2P topology to enable each client to identify and aggregate a personalized set of trustworthy and beneficial updates.This framework is the Local Inference Guided Aggregation for Heterogeneous Training Environments to Yield Enhancement Through Agreement and Regularization (LIGHTYEAR). Central to our method is an agreement score, computed on a local validation set, which quantifies the semantic alignment of incoming updates in the function space with respect to the clients reference model. Each client uses this score to select a tailored subset of updates and performs aggregation with a regularization term that further stabilizes the training. Our empirical evaluation across two datasets shows that the proposed approach consistently outperforms both centralized baselines and existing P2P methods in terms of client-level performance, particularly under adversarial and heterogeneous conditions.

Abstract
Adversarial Training (AT) is one of the most effective methods to train robust Deep Neural Networks (DNNs). However, AT creates an inherent trade-off between clean accuracy and adversarial robustness, which is commonly attributed to the more complicated decision boundary caused by the insufficient learning of hard adversarial samples. In this work, we reveal a counterintuitive fact for the first time: From the perspective of perception consistency, hard adversarial samples that can still attack the robust model after AT are already learned better than those successfully defended. Thus, different from previous views, we argue that it is rather the over-sufficient learning of hard adversarial samples that degrades the decision boundary and contributes to the trade-off problem. Specifically, the excessive pursuit of perception consistency would force the model to view the perturbations as noise and ignore the information within them, which should have been utilized to induce a smoother perception transition towards the decision boundary to support its establishment to an appropriate location. In response, we define a new AT objective named Robust Perception, encouraging the model perception to change smoothly with input perturbations, based on which we propose a novel Robust Perception Adversarial Training (RPAT) method, effectively mitigating the current accuracy-robustness trade-off. Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet with ResNet-18, PreActResNet-18, and WideResNet-34-10 demonstrate the effectiveness of our method beyond four common baselines and 12 state-of-the-art (SOTA) works. The code is available at https://github.com/FlaAI/RPAT.

Abstract
A common use of machine learning (ML) models is predicting the class of a sample. Object detection is an extension of classification that includes localization of the object via a bounding box within the sample. Classification, and by extension object detection, is typically evaluated by counting a prediction as incorrect if the predicted label does not match the ground truth label. This pass/fail scoring treats all misclassifications as equivalent. In many cases, class labels can be organized into a class taxonomy with a hierarchical structure to either reflect relationships among the data or operator valuation of misclassifications. When such a hierarchical structure exists, hierarchical scoring metrics can return the model performance of a given prediction related to the distance between the prediction and the ground truth label. Such metrics can be viewed as giving partial credit to predictions instead of pass/fail, enabling a finer-grained understanding of the impact of misclassifications. This work develops hierarchical scoring metrics varying in complexity that utilize scoring trees to encode relationships between class labels and produce metrics that reflect distance in the scoring tree. The scoring metrics are demonstrated on an abstract use case with scoring trees that represent three weighting strategies and evaluated by the kind of errors discouraged. Results demonstrate that these metrics capture errors with finer granularity and the scoring trees enable tuning. This work demonstrates an approach to evaluating ML performance that ranks models not only by how many errors are made but by the kind or impact of errors. Python implementations of the scoring metrics will be available in an open-source repository at time of publication.

Abstract
Snowflake revolutionized data analytics with an elastic architecture that decouples compute and storage, enabling scalable solutions supporting data architectures like data lake, data warehouse, data lakehouse, and data mesh. Building on this foundation, Snowflake has advanced its AI Data Cloud vision by introducing Snowpark, a managed turnkey solution that supports data engineering and AI and ML workloads using Python and other programming languages. This paper outlines Snowpark's design objectives towards high performance, strong security and governance, and ease of use. We detail the architecture of Snowpark, highlighting its elastic scalability and seamless integration with Snowflake core compute infrastructure. This includes leveraging Snowflake control plane for distributed computing and employing a secure sandbox for isolating Snowflake SQL workloads from Snowpark executions. Additionally, we present core innovations in Snowpark that drive further performance enhancements, such as query initialization latency reduction through Python package caching, improved workload scheduling for customized workloads, and data skew management via efficient row redistribution. Finally, we showcase real-world case studies that illustrate Snowpark's efficiency and effectiveness for large-scale data engineering and AI and ML tasks.

Abstract
We present SAInT, a Python-based tool for visually exploring and understanding the behavior of Machine Learning (ML) models through integrated local and global sensitivity analysis. Our system supports Human-in-the-Loop (HITL) workflows by enabling users - both AI researchers and domain experts - to configure, train, evaluate, and explain models through an interactive graphical interface without programming. The tool automates model training and selection, provides global feature attribution using variance-based sensitivity analysis, and offers per-instance explanation via LIME and SHAP. We demonstrate the system on a classification task predicting survival on the Titanic dataset and show how sensitivity information can guide feature selection and data refinement.

Abstract
PhysiBoSS is an open-source platform that integrates agent-based modeling of cell populations with intracellular stochastic Boolean networks, enabling multiscale simulations of complex biological behaviors. To promote model sharing and versioning, we present the PhysiBoSS-Models database: a curated repository for multiscale models built with PhysiBoSS. By providing a simple Python API, PhysiBoSS-Models provides an easy way to download and simulate preexisting models through tools such as PhysiCell Studio. By providing standardized access to validated models, PhysiBoSS-Models facilitates reuse, validation, and benchmarking, supporting research in biology.

Abstract
Consider a collection of competing machine learning algorithms. Given their performance on a benchmark of datasets, we would like to identify the best performing algorithm. Specifically, which algorithm is most likely to rank highest on a future, unseen dataset. A natural approach is to select the algorithm that demonstrates the best performance on the benchmark. However, in many cases the performance differences are marginal and additional candidates may also be considered. This problem is formulated as subset selection for multinomial distributions. Formally, given a sample from a countable alphabet, our goal is to identify a minimal subset of symbols that includes the most frequent symbol in the population with high confidence. In this work, we introduce a novel framework for the subset selection problem. We provide both asymptotic and finite-sample schemes that significantly improve upon currently known methods. In addition, we provide matching lower bounds, demonstrating the favorable performance of our proposed schemes.

Abstract
This paper argues that Active Inference (AIF) provides a crucial foundation for developing autonomous AI agents capable of learning from experience without continuous human reward engineering. As AI systems begin to exhaust high-quality training data and rely on increasingly large human workforces for reward design, the current paradigm faces significant scalability challenges that could impede progress toward genuinely autonomous intelligence. The proposal for an ``Era of Experience,'' where agents learn from self-generated data, is a promising step forward. However, this vision still depends on extensive human engineering of reward functions, effectively shifting the bottleneck from data curation to reward curation. This highlights what we identify as the \textbf{grounded-agency gap}: the inability of contemporary AI systems to autonomously formulate, adapt, and pursue objectives in response to changing circumstances. We propose that AIF can bridge this gap by replacing external reward signals with an intrinsic drive to minimize free energy, allowing agents to naturally balance exploration and exploitation through a unified Bayesian objective. By integrating Large Language Models as generative world models with AIF's principled decision-making framework, we can create agents that learn efficiently from experience while remaining aligned with human values. This synthesis offers a compelling path toward AI systems that can develop autonomously while adhering to both computational and physical constraints.

Abstract
Despite their significant economic contributions, Small and Medium Enterprises (SMEs) face persistent barriers to securing traditional financing due to information asymmetries. Cash flow lending has emerged as a promising alternative, but its effectiveness depends on accurate modelling of transaction-level data. The main challenge in SME transaction analysis lies in the unstructured nature of textual descriptions, characterised by extreme abbreviations, limited context, and imbalanced label distributions. While consumer transaction descriptions often show significant commonalities across individuals, SME transaction descriptions are typically nonstandard and inconsistent across businesses and industries. To address some of these challenges, we propose a bank categorisation pipeline that leverages synthetic data generation to augment existing transaction data sets. Our approach comprises three core components: (1) a synthetic data generation module that replicates transaction properties while preserving context and semantic meaning; (2) a fine-tuned classification model trained on this enriched dataset; and (3) a calibration methodology that aligns model outputs with real-world label distributions. Experimental results demonstrate that our approach achieves 73.49% (+-5.09) standard accuracy on held-out data, with high-confidence predictions reaching 90.36% (+-6.52) accuracy. The model exhibits robust generalisation across different types of SMEs and transactions, which makes it suitable for practical deployment in cash-flow lending applications. By addressing core data challenges, namely, scarcity, noise, and imbalance, our framework provides a practical solution to build robust classification systems in data-sparse SME lending contexts.