Applied AI letters最新文献

Patents are integral to our shared scientific knowledge, requiring companies and inventors to stay informed about them to conduct research, find licensing opportunities, and manage legal risks. However, the rising rate of filings has made this task increasingly challenging over the years. To address this issue, we introduce ChemQuery, a tool for easily exploring chemistry-related patents using natural language questions. Traditional systems rely on simplistic keyword-based searches to find patents that might be relevant to a user's request. In contrast, ChemQuery uses up-to-date information to return specific answers, along with their sources. It also offers a more comprehensive search experience to the users, thanks to capabilities like extracting molecules from diagrams, integrating information from PubChem, and allowing complex queries about molecular structures. We conduct a thorough empirical evaluation of ChemQuery and compare it with several baseline approaches. The results highlight the practical utility and limitations of our tool.

Current resume screening relies on manual review, causing delays and errors in evaluating large volumes of resumes. Lack of automation and data extraction leads to inefficiencies and potential biases. Recruiters face challenges in identifying qualified candidates due to oversight and time constraints. Inconsistent evaluation criteria hinder decision-making. These issues result in prolonged hiring processes, missed opportunities, and potential bias in candidate selection. The goal of this project is to develop an AI-powered Resume Analysis and Recommendation Tool, catering to the trend of recruiters spending less than 2 min on each CV. The tool will rapidly analyze all resume components while providing personalized predictions and recommendations to applicants for improving their CVs. It will present user-friendly data for recruiters, facilitating export to CSV for integration into their recruitment processes. Additionally, the tool will offer insights and analytics on popular roles and skills within the job market. Its user section will enable applicants to continually test and track their resumes, encouraging repeat usage and driving traffic. Colleges can benefit from gaining insights into students' resumes before placements. Overall, this AI-powered tool aims to enhance the resume evaluation process, benefiting both job seekers and employers. The primary aim of this project is to develop a Resume Analyzer using Python, incorporating advanced libraries such as Pyresparser, NLTK (Natural Language Toolkit), and MySQL. This automated system offers an efficient solution for parsing, analyzing, and extracting essential information from resumes. The user-friendly interface, developed using Streamlit, allows for seamless resume uploading, insightful data visualization, and analytics. The Resume Analyzer significantly streamlines the resume screening process, providing recruiters with valuable insights and enhancing their decision-making capabilities.

Across numerous households in Sub-Saharan Africa, agriculture plays a crucial role. One solution that can effectively bridge the support gap for farmers in the local community is a question–answer system based on agricultural expertise and agro-information. The more recent advancements in question answering research involve the use of large language models that are trained on an extensive amount of data. Due to this, conventional fine-tuning approaches have demonstrated a significant decline in performance when using a significantly smaller amount of data. One proposed alternative to address this decline is to use prompt-based fine-tuning, which allows the model to be fine-tuned with only a few examples, thus addressing the disparities between the objectives of pretraining and fine-tuning. Extensive research has been done on these methods, specifically on text classification and not question answering. In this research, our objective was to study the feasibility of recent few-shot learning approaches such as FewshotQA and Null-prompting for domain-specific agricultural data in four South African languages. We first explored creating a cross-lingual domain-specific extractive question answering dataset through an automated approach using the GPT model. Through exploratory data analysis, the GPT model was able to create a dataset, which requires minor improvements. We then evaluated the overall performance of the different approaches and investigated the effects of adapting these approaches to suit the new dataset. Results show these methods effectively capture semantic relationships and domain-specific terminology but exhibit limitations, including potential biases in automated annotation and plateauing F1 scores. This highlights the need for hybrid approaches that combine artificial intelligence and human supervision. Beyond academic insights, this study has practical significance for industry, demonstrating how prompt-based methods can help tailor AI models to specific use cases in low-resource settings.

The challenge of antimicrobial resistance (AMR) represents one of the most pressing global health crises, particularly, in resource-constrained settings like Africa. In this paper, we explore artificial intelligence (AI) and machine learning (ML) potential in transforming the potential for antimicrobial stewardship (AMS) to improve precision, efficiency, and effectiveness of antibiotic use. The deployment of AI-driven solutions presents unprecedented opportunities for optimizing treatment regimens, predicting resistance patterns, and improving clinical workflows. However, successfully integrating these technologies into Africa's health systems faces considerable obstacles, including limited human capacity and expertise, widespread public distrust, insufficient funding, inadequate infrastructure, fragmented data sources, and weak regulatory and policy enforcement. To harness the full potential of AI and ML in AMS, there is a need to first address these foundational barriers. Capacity-building initiatives are essential to equip healthcare professionals with the skills needed to leverage AI technologies effectively. Public trust must be cultivated through community engagement and transparent communication about the benefits and limitations of AI. Furthermore, technological solutions should be tailored to the unique constraints of resource-limited settings, with a focus on developing low-computational, explainable models that can operate with minimal infrastructure. Financial investment is critical to scaling successful pilot projects and integrating them into national health systems. Effective policy development is equally essential to establishing regulatory frameworks that ensure data security, algorithmic fairness, and ethical AI use. This comprehensive approach will not only improve the deployment of AI systems but also address the underlying issues that exacerbate AMR, such as unauthorized antibiotic sales and inadequate enforcement of guidelines. To effectively and sustainably combat AMR, a concerted effort involving governments, health organizations, communities, and technology developers is essential. Through collaborations and sharing a common goal, we can build resilient and effective AMS programs in Africa.

Text-to-speech (TTS) models have expanded the scope of digital inclusivity by becoming a basis for assistive communication technologies for visually impaired people, facilitating language learning, and allowing for digital textual content consumption in audio form across various sectors. Despite these benefits, the full potential of TTS models is often not realized for the majority of low-resourced African languages because they have traditionally required large amounts of high-quality single-speaker recordings, which are financially costly and time-consuming to obtain. In this paper, we demonstrate that crowdsourced recordings can help overcome the lack of single-speaker data by compensating with data from other speakers of similar intonation (how the voice rises and falls in speech). We fine-tuned an English variational inference with adversarial learning for an end-to-end text-to-speech (VITS) model on over 10 h of speech from six female common voice (CV) speech data speakers for Luganda and Kiswahili. A human mean opinion score evaluation on 100 test sentences shows that the model trained on six speakers sounds more natural than the benchmark models trained on two speakers and a single speaker for both languages. In addition to careful data curation, this approach shows promise for advancing speech synthesis in the context of low-resourced African languages. Our final models for Luganda and Kiswahili are available at https://huggingface.co/marconilab/VITS-commonvoice-females.

The escalating integration of artificial intelligence (AI) in public healthcare has raised a critical concern: the vast amounts of data being generated and utilised by AI language models are not adequately connected to privacy and security considerations. This study addresses the problem by exploring how AI language models can be used to enhance digital security in public healthcare while addressing challenges related to privacy and ethics. The research adopts a three-phase methodology: a bibliometric analysis of literature from the Scopus database to identify research trends, the generation of AI-driven scenarios refined by healthcare professionals and analysing AI responses using grounded theory. Two scenarios, focused on AI-driven clinical decision support systems and AI-powered telemedicine platforms, were validated by healthcare experts and tested using ChatGPT-4 and Gemini, two prominent AI models. While ChatGPT-4 produced contextually specific and diverse responses, Gemini's outputs were inconsistent and repetitive, highlighting discrepancies in their performance. These discrepancies are linked to the data used to train these models, implying that incorporating more specialised healthcare data could enhance performance; however, such data usage must align with ethical guidelines. The analysis found that human, organizational, and technological dimensions are critical for addressing security issues and promoting trust in healthcare systems utilising AI. While AI-generated scenarios are a valuable starting point, they must be mediated by medical professionals to ensure practical applicability. The findings provide a theoretical framework for handling AI-generated issues related to privacy and security concerns, which can be used for future empirical research to enhance digital security in public healthcare.

Businesses have invested billions into artificial intelligence (AI) applications, leading to a sharp rise in the number of AI applications being released to customers. Taking into account previous approaches to attacking machine learning models, we conduct a comparative analysis of adversarial attacks, contrasting large language models (LLMs) being deployed through application programming interfaces (APIs) with the same attacks against locally deployed models to evaluate the significance of security controls in production deployments on attack success in black-box environments. The article puts forward adversarial attacks that are adapted for remote model endpoints in order to create a threat model that can be used by security organizations to prioritize controls when deploying AI systems through APIs. This paper contributes: (1) a public repository of adversarial attacks adapted to handle remote models on https://github.com/l3ra/adversarial-ai, (2) benchmarking results of remote attacks comparing the effectiveness of attacks on remote models with those on local models, and (3) a framework for assessing future AI system deployment controls. By providing a practical framework for benchmarking the security of remote AI systems, this study contributes to the understanding of adversarial attacks in the context of natural language processing models deployed by production applications.

To maximize business value from artificial intelligence and machine learning (ML) systems, understanding what leads to the effective development and deployment of ML systems is crucial. While prior research primarily focused on technical aspects, important issues related to improving decision-making across ML workflows have been overlooked. This paper introduces a “normative-descriptive-prescriptive” decision framework to address this gap. Normative guidelines outline best practices, descriptive dimensions describe actual decision-making, and prescriptive elements provide recommendations to bridge gaps. The three-step framework analyzes decision-making in key ML pipeline phases, identifying gaps and offering prescriptions for improved model building. Key descriptive findings include rushed problem-solving with convenient data, use of inaccurate success metrics, underestimation of downstream impacts, limited roles of subject matter experts, use of non-representative data samples, prioritization of prediction over explanation, lack of formal verification processes, and challenges in monitoring production models. The paper highlights biases, incentive issues, and systematic disconnects in decision-making across the ML pipeline as contributors to descriptive shortcomings. Practitioners can use the framework to pinpoint gaps, develop prescriptive interventions, and build higher quality, ethical, and legally compliant ML systems.

The study suggests using a hybrid convolutional neural networks-support vector machines architecture to extract reliable characteristics from medical images and classify them as an ensemble using four different models. Manual processing of fundus images for the automated identification of ocular disorders is laborious, error-prone, and time-consuming. This necessitates computer-assisted technologies that can automatically identify different ocular illnesses from fundus images. The interpretation of the photos also plays a massive role in the diagnosis. Automating the diagnosing procedure reduces human mistakes and helps with early cataract detection. The oneDNN library available in the oneAPI Environment provided by Intel has been used to optimize all transfer learning models for better performance. The suggested approach is verified through a range of metrics in experiments using the openly accessible Ocular Disease Intelligent Recognition dataset. The MobileNet Model outperformed other transfer learning techniques with an accuracy of 0.9836.

Large Language Models (LLMs) are continuously being evaluated in more diverse contexts. However, they still have challenges when extracting information from highly specific internal technical documentation. One specific area of real-world technical documentation is telecommunications engineering, which could benefit from domain-specific LLMs. In this article, we expand the notion of Technical Language Processing (TLP) to the telecommunications domain by introducing and analyzing the format of technical specifications from a leading telecommunications equipment vendor. Additionally, we highlight the importance of use case definitions by introducing requirement property mapping for maximizing information extraction. Also, we recommend actions to mitigate the effect of the internal specifications format on information extraction, which can lead to LLM-friendly internal specifications. Finally, a PoC is built to showcase the improvements of our proposed framework.