Developing accountable models with limited data is a pivotal challenge in machine learning. Common strategies include utilizing large supporting datasets via transfer learning or incorporating domain knowledge. However, domain knowledge is typically added in pre-processing and post-processing stages, insufficient for effective training. Frameworks like active learning offer iterative mechanisms for expert input, but ignore experts’ sophistication and limitations. This paper proposes a framework where expert knowledge input is central, enabling iterative teaching by both data and experts. Experts provide critical knowledge, avoiding overwhelm by extensive training epochs. Precise model explanation definitions and quantifiable expert knowledge distinctions are required. By integrating both strategies, bias and scarcity are addressed, empowering experts to monitor, understand, and influence learning. Comparisons with similar frameworks and experimental results demonstrate improved performance with less data, facilitating collaboration between machine learning and industries.