Humans effortlessly perceive and interpret multiple affective states from multimodal signals, even in complex and dynamic environments. However, affective computing techniques are typically developed for specific tasks in controlled settings, using predefined datasets, architectures, and objectives. As a result, they often lack the flexibility to handle multiple affective states simultaneously, leading to inefficiencies and limiting their applicability in real-world scenarios.

    Recently, foundation models have emerged as a promising solution, demonstrating strong performance across various affective tasks and offering a more comprehensive approach to affective intelligence. However, their adoption also raises critical ethical concerns, including privacy risks, fairness, sustainability, and bias. Real-world cases have shown the potential for LLM-driven interactions to negatively impact mental health, particularly in younger users. Therefore, ensuring their responsible and ethical use is more urgent than ever.

    This workshop aims to advance both the holistic development of affective computing and the understanding of its associated ethical challenges. Two key questions we aim to address:

1. What are the relatedness between different affective states, and how can it benefit their holistic affective modeling?

2. How can holistic affective states be modeled while ensuring responsibility, such as privacy, factuality, and preventing malicious use?