Communication barriers between deaf and hearing individuals can be mitigated through advancements in sign language recognition (SLR) systems. These SLR systems can also improve the user experience of deaf people when interacting with conversational systems that could emerge in the near future. This work explores a landmark-based approach for word classification within an SLR system. The study investigates the impact of a novel data-cleaning methodology on model performance during training. Specifically, a data cleaning process focused on video trimming and sign placement correction is shown to significantly improve dataset quality, resulting in more accurate classification. This cleaner data not only facilitated a more stable training process for the RNN model but also effectively delayed the onset of overfitting compared to a model trained on the original data. The findings highlight the critical role of data quality, particularly when dealing with the limitations inherent to small datasets commonly encountered in SLR tasks. The contribution of this study lies in demonstrating how targeted data cleaning enhances model stability and performance in resource-limited SLR systems.

Thispaper compares the usability of various Apple MacBook Pro laptops were tested for basic machine learning research applications, including text-based, vision-based, and tabular data. Four tests/benchmarks were conducted using four different MacBook Pro models—M1, M1 Pro, M2, and M2 Pro. A script written in Swift was used to train and evaluate four machine learning models using the Create ML framework, and the process was repeated three times. The script also measured performance metrics, including time results. The results were presented in tables, allowing for a comparison of the performance of each device and the impact of their hardware architectures.

This paper proposes a concept of Digital Stereotypes, observed during research on quantitative overrepresentation of one class over others, and its impact on the results of the training of Deep Learning models. The real-life observed data classes are rarely of the same size, and the intuition of presenting multiple examples of one class and then showing a few counterexamples may be very misleading in multimodal classification. Deep Learning models, when taught with overrepresentation, may produce incorrect inferring results, similar to stereotypes. The generic idea of stereotypes seems to be helpful for categorisation from the training point of view, but it has a negative influence on the inferring result. Authors evaluate a large dataset in various scenarios: overrepresentation of one or two classes, underrepresentation of some classes, and same-size (trimmed) classes. The presented research can be applied to any multiclassification applications, but it may be especially important in AI, where the classification, uncertainty and building new knowledge overlap. This paper presents specific ’decreases in accuracy’ observed within multiclassification of unleveled datasets. The ’decreases in accuracy’, named by the authors ’stereotypes’, can also bring an inspiring insight into other fields and applications, not only multimodal sentiment analysis.