PowerPoint Presentation

Ulster University.

[Audio] Over project is Semi-Automated Data Labelling.

[Audio] Artificial Intelligence (AI) models are currently revolutionising the industry. The scarcity of labelled data, however, has historically been one of the major hurdles. Without a sufficient amount of labelled data, models can result in being inaccurate or biased..

[Audio] Even when large volumes of unlabelled data are available, the data annotation process is expensive, time-consuming and error-prone, and is exposed to the problem of imbalance and biases..

[Audio] The SADL project aims at investigating methods that can support and partially automate the annotation, including, Label Propagation, Self-Supervised and Semi-Supervised Methods, Reinforcement Learning from Human Feedback (RLHF), Dealing with Class Imbalance and Bias, Few-Shot Learning..

[Audio] The proposed Solution for this project includes, Explainable and trustworthy AI, Monitoring for Biases and Data Imbalance, Reduced Human Interaction in Feedback Loop, AI lifecycle management, Integration of Unsupervised, Self-Supervised, and Few-Shot Learning..

[Audio] With Phase I almost completed and having conducted a literature review of existing technologies, the next steps will include preliminary experimentation evaluating techniques on a common benchmark. Based on the results of these experiments, the combination of identified suitable techniques will be used to design a novel architecture..