Deloitte (United States)

Community-science platforms such as iNaturalist now contain tens of millions of georeferenced, photographically vouchered biodiversity records, yet extracting reliable quantitative measurements from opportunistic photographs remains methodologically challenging. Here, I evaluate the signal-to-noise ratio of iNaturalist photos for phenotyping Plethodon salamanders across two trait classes: continuous dorsal brightness (a proxy for ecogeographic clines predicted by Gloger's rule and the thermal melanism hypothesis) and discrete color morph frequency in P. cinereus. I optimized a color-extraction pipeline using an agent-guided parameter search adapted from the autoresearch framework (Karpathy 2026; Schmidgall et al. 2025), exploring crop fraction, color space, normalization, and quality-control thresholds across 50 bounded micro-experiments. Applying the production HSV pipeline to 103,653 observations of 34 species, I found negligible geographic structure in dorsal brightness (Rsq = 0.001), even within P. cinereus alone (n = 71,627). Variance decomposition showed that photographer identity explains 23.3% of brightness variance, geography 5.1%, species 1.6%, and time of day 0.3%, with 69.7% residual. In contrast, a hue-threshold morph classifier recovered a significant geographic signal in red-back frequency (Rsq = 0.008, p < 0.001), 7 times stronger than the brightness result, though still weaker than the supervised CNN of Hantak et al. (2022; pseudo-Rsq = 0.04). These results indicate that citizen-science photographs are poorly suited to continuous quantitative phenotyping under current collection conditions, whereas discrete categorical traits remain recoverable with appropriate classifiers. The autoresearch loop clarified the failure mode: no tested parameter configuration recovered a meaningful brightness signal from a dataset dominated by observer effects.

Short Message Service (SMS) spam classification systems based on machine learning and deep learning have achieved high accuracy under standard evaluation settings; however, their vulnerability to adversarial attacks poses significant risks to real-world deployment. Attacks on SMS spam classifiers are realized as precision-level adversarial examples, where carefully crafted perturbations (e.g., character substitution/reordering, word disguising and synonym injection, benign token insertion) maintain human readability while drawing models to misclassify. In this paper, we study universal adversarial attacks on SMS spam clasification models and their adaptation to various feature representations and learning architectures: from classic machine learning models to deep neural networks and transformer-based solutions. And it introduces defense terms like adversarial training, data augmentation, ensemble learning, input sanitization and defensive distillation. Primary challenges such as trade-offs between robustness and accuracy, language diversity, and limited computation in mobile contexts are considered. The findings of the paper show possible future research challenges worth addressing for building robust SMS spam detection systems that are still performing effectively under adversarial conditions.

Introduction: Effectively managing biological risk in a research setting requires a strong organizational commitment to promoting safe practices and adhering to established protocols, a concept known as a “culture of biosafety.” Although a sound culture of biosafety is critical for responsible conduct of research, its assessment remains challenging, relying on self-reporting and qualitative and subjective evaluations. Methods: To determine if a laboratory’s culture of biosafety could be evaluated objectively, we examined historical biosafety violation data at three U.S. universities, including biosafety cabinet recertification violations at one university and a range of violations at two other universities. These data were analyzed to determine the distribution of violations among investigators. Results: At all three institutions, we found that a majority of investigators operate laboratories with low rates of biosafety violations and few repeat violations; however, a minority of investigators operate laboratories that are repeatedly in violation of established biosafety guidelines and standards. Repeat violations were unequally distributed amongst laboratories, with only 20% of laboratories accounting for 70–90% of total repeat violations. Discussion: We found that a vast majority of investigators operate laboratories with commendable cultures of biosafety that manifest as low rates of violations and repeat violations. Conclusion: This research establishes metrics for evaluating a laboratory’s culture of biosafety with an objective evidence basis and transparent approach, laying the foundation for future research to understand the drivers of biosafety culture, create interventions to improve biosafety culture, and creating a near-term mechanism to promote biosafety culture at institutions via data-driven retention and recruitment strategies.

The swift digitalization of agriculture via smart farming technologies has transformed food production, sustainability, and efficiency of resources. Nevertheless, the confluence of IoT, cloud computing, AI, and big data has also precipitated enormous cybersecurity and privacy issues. This review examines 24 seminal studies summarizing security threats, vulnerabilities, and technological frameworks in smart farming ecosystems. The literature identifies an increasing pattern of cyberattacks against IoT-enabled farms, from data breaches and unauthorized access to device-level tampering. Countermeasures like encryption, lightweight authentication, AI-powered threat detection, and edge computing are examined to enhance system robustness. Focus is also given to policymaking, interoperability concerns, and the requirement for energy-efficient, scalable solutions. Novel ideas such as digital twins, blockchain, and large language models (LLMs) are suggested to enhance cyber defense solutions even further. The review highlights a major gap in real-world testing and deployment of security frameworks, emphasizing the need for well-rounded, context-aware strategies that reconcile innovation with security. With the smart farming industry growing worldwide, addressing these security and privacy issues is crucial to ensure resilient, sustainable agricultural practices.

Gamification is viewed as a promising solution for employing game design elements in school leader professional learning and skill acquisition, and this systematic review summarizes research that has employed game-based and gamification approaches in K-12 school leadership development from 2020 to 2024. A total of 159 studies were identified in the ERIC, Google Scholar, and IDESR databases. After removing duplicates and excluding studies based on relevance and methodological quality, fourteen studies were included. The review was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyzes extension for Scoping Reviews (PRISMA-ScR) 2020. Qualitative content analysis and thematic synthesis were used to identify trends, constructs, and implementation outcomes. Leadership training courses involving gamification often used some kind of simulations, serious games or gamified elements to train leadership skills such as decision-making, cooperation or system thinking. Almost all studies affected training participants and engaged them. Those studies also caused them to have self-efficacy and to reflect in learning in a positive way. Researchers need to study more longitudinally and also within different educational contexts to corroborate these early findings. However, this review suggests that gamification may innovate how schools build leadership capacity. More rigorous research using standardized measurements should be conducted to confirm the results and determine the impact of long-term implementation. This research adds to a growing body of literature in online professional learning and has implications for education policy makers and leadership development organizations.

Abstract Document classification serves as foundational step in critical tasks such as information extraction, analysis and decision-making. However, existing approaches often struggle with the variability, volume, and complexity of real-world documents. These methods are further limited by a lack of configurability and explainability, requiring specialized technical expertise to accommodate diverse user needs and often producing results that are difficult to interpret. To address the complexities of modern document processing, this paper introduces a novel zero-shot document classification framework that leverages Large Language Models (LLMs), designed for accessibility and configurability by both technical and non-technical users. Unlike traditional methods, which require extensive labeled data, the zero-shot configuration enables our framework to perform the classification task without any prior exposure to labeled examples of the target categories, relying instead on semantic understanding derived from user-provided label descriptions and document content. Developed and validated using a real-world banking dataset, our framework leverages different strategies for providing context to LLMs during classification. Experimental results demonstrate substantial improvements in both accuracy and efficiency, outperforming current zero-shot methods while also reducing operating costs.

The rapid rise of generative AI has made secure and scalable prompt engineering more important than ever. But there’s a catch — prompts often carry sensitive, proprietary, or regulated information, which brings real security and privacy risks. In response, our research presents a blockchain-powered framework designed for confidential prompt sharing. It brings together smart contracts, zero-knowledge proofs (ZKPs), and a hybrid consensus model that blends Proof-of-Stake with Byzantine Fault Tolerance (PoS-BFT). This combination doesn’t just enhance security — it also boosts energy efficiency, scalability, and resilience. When benchmarked against platforms like Ethereum and Hyperledger Fabric, our framework delivered lower latency, better performance, and strong reliability even under adversarial conditions. Plus, it’s built to work handin-hand with agentic AI workflows and decentralized identity systems, offering a flexible, futureready foundation for secure collaboration between AI agents.

Software services are crucial for reliable communication and networking, therefore, Site Reliability Engineering (SRE) is important to ensure these systems stay reliable and perform well in cloud-native environments. SRE leverages tools like Prometheus and Grafana to monitor system metrics, defining critical Service Level Indicators (SLIs) and Service Level Objectives (SLOs) for maintaining high service standards. However, a significant challenge arises as many developers often lack indepth understanding of these tools and the intricacies involved in defining appropriate SLIs and SLOs. To bridge this gap, we propose a novel SRE platform, called “SRE-Llama”, enhanced by Generative-AI, Federated Learning, Blockchain and NonFungible Tokens (NFTs). This platform aims to automate and simplify the process of monitoring, SLI/SLO generation, and alert management, offers ease in accessibility and efficy for developers. The automation processes are governed by smart contracts on the Blockchain, ensuring transparency and security. The system operates by capturing metrics from cloud-native services and storing them in a time-series database, like Prometheus and Mimir. Utilizing this stored data, our platform employs Federated Learning models to identify the most relevant and impactful SLI metrics for different services and SLO objectives values, addressing concerns around data privacy and decentralized data sources. Subsequently, custom-trained Meta’s Llama-3 LLM is adopted to intelligently generate SLIs, SLOs, Error-budgets, and associated alerting mechanisms based on these identified SLI metrics. The Llama-3-8B LLM has been quantized and finetuned using Quantized Low-Rank Adaptation (QLoRA) to ensure optimal performance on consumer-grade hardware. A unique aspect of our platform is the encoding of generated SLIs and SLOs as NFT objects, which are then stored on a Blockchain. This feature provides immutable record-keeping and facilitates easy verification and auditing of the SRE metrics and objectives. It enhances the traceability and accountability of the SRE processes, offering a verifiable and transparent record of the system’s performance standards. The proposed SRE-Llama platform prototype has been implemented with a use case featuring a customized Open5GS 5G Core.

5G network slicing presents a transformative approach to deliver customizable, isolated virtual networks over shared infrastructure, tailored to meet diverse application requirements such as ultra-low latency, high reliability, or massive connectivity. However, orchestrating these network slices dynamically and efficiently remains a complex challenge due to the need for precise configuration, real-time responsiveness, and seamless coordination across heterogeneous domains (radio access network, core, and transport). In this paper, we present Slice-MCP, a novel 5G network slice orchestration platform that integrates the Model Context Protocol (MCP) with a fine-tuned Meta’s Llama-4 large language model (LLM) to simplify and automate the slice orchestration process. Slice-MCP allows users to invoke natural language queries which are interpreted by the fine-tuned LLM and mapped to the corresponding MCP functions to configure and deploy the appropriate network slices. To ensure accurate intent recognition and robust function mapping, the LLM was fine-tuned on a custom dataset comprising real-world user intents paired with their corresponding MCP server operations. The integration of MCP provides a standardized interface for dynamic context management, enabling a consistent and extensible way to translate high-level user goals into orchestrator-level instructions across network domains. To the best of our knowledge, this is the first work to apply the MCP for 5G network slice orchestration, merging AI-driven natural language interaction with telecom-grade infrastructure management. The effectiveness of the Slice-MCP is demonstrated through a real-world test-bed deployment in a sliced network scenario, utilizing multiple 5G cores (i.e., Open5GS) across Ericsson’s new RAN. Our results demonstrate that Slice-MCP significantly reduces the operational complexity of 5G slice management while enhancing responsiveness, scalability, and user accessibility.