With the increasing complexity of business processes in today's organizations and the ever-growing amount of structured accounting data, identifying erroneous or fraudulent business transactions and corresponding journal entries poses a major challenge for public accountants at annual audits. In current audit practice, mainly static rules are applied which check only a few attributes of a journal entry for suspicious values. Encouraged by numerous successful adoptions of deep learning in various domains we suggest an approach for applying autoencoder neural networks to detect unusual journal entries within individual financial accounts. The identified journal entries are compared to a list of entries that were manually tagged by two experienced auditors. The comparison shows high f-scores and high recall for all analyzed financial accounts. Additionally, the autoencoder identifies anomalous journal entries that have been overlooked by the auditors. The results underpin the applicability and usefulness of deep learning techniques in financial statement audits.
In the last few decades, text mining has been used to extract knowledge from free texts. Applying neural networks and deep learning to natural language processing (NLP) tasks has led to many accomplishments for real-world language problems over the years. The developments of the last five years have resulted in techniques that have allowed for the practical application of transfer learning in NLP. The advances in the field have been substantial, and the milestone of outperforming human baseline performance based on the general language understanding evaluation has been achieved. This paper implements a targeted literature review to outline, describe, explain, and put into context the crucial techniques that helped achieve this milestone. The research presented here is a targeted review of neural language models that present vital steps towards a general language representation model.
Incomplete time series present a significant challenge for downstream analysis. In the field of time series, Large Language Models are already being used for prediction, classification, and, in rare cases, imputation. This study thoroughly examines the imputation of time series using Large Language Models. Within a defined experimental setup, current state-of-the-art time series imputation methods are compared with the performance of Large Language Models. Parameter-efficient fine-tuning methods are applied to adapt the Large Language Models to the imputation task. The results indicate that the models are suitable for time series imputation. The performance of these models depends on the number of parameters and the type of pre-training. Small specialized models, such as BERT, compete with models like Llama2 and outperform them on selected datasets. Furthermore, it becomes clear that the attention and feedforward network components of Large Language Models are particularly well-suited for adaptation to imputation, and parameter-efficient methods are also performance-enhancing.
Our investigation seeks to enhance the understanding of responsible artificial intelligence. The EU is deeply engaged in discussions concerning AI trustworthiness and has released several relevant documents. It’s crucial to remember that while AI offers immense benefits, it also poses risks, necessitating global oversight. Moreover, there’s a need for a framework that helps enterprises align their AI development with these international standards. This research will aid both policymakers and AI developers in anticipating future challenges and prioritizing their efforts. In our study, we delve into the essence of responsible AI and, to our understanding, introduce a comprehensive definition of the term. Through a thorough literature review, we pinpoint the prevailing trends surrounding responsible AI. Using insights from our analysis, we’ve also deliberated on a prospective framework for responsible AI. Our findings emphasize that human-centeredness should prioritized. This entails adopting AI techniques that prioritize ethical considerations, explainability of models, and aspects like privacy, security, and trustworthiness.
More and more applications nowadays use spatio-temporal data for different purposes. In order to be processed and used efficiently, this unique type of data requires special handling. This paper summarizes methods and approaches for feature selection of spatio-temporal data and machine learning algorithms for spatio-temporal data engineering. Furthermore, it highlights relevant work in specific domains. The range of possible approaches for data processing is quite wide. However, in order to use these approaches with the spatio-temporal data in a meaningful and practical way, individual data processing steps need to be adapted. One of the most important steps is feature engineering.