TADIL: Task-Agnostic Domain Learning

GitHub: TADIL
Publications: ICPR 2024 (Oral) - arXiv
BSC: Official software page
Media: Lenovo Press Article
Talk: ContinualAI invited talk (YouTube)

Task-agnostic domain-incremental learning approach using transformer nearest-centroid embeddings. Enables effective domain adaptation without requiring task labels, achieving 6% performance improvement across diverse domains.

• Selected for oral presentation at ICPR 2024
• Featured in ContinualAI invited talk
• Applied to autonomous driving scenarios
• Published in Lenovo Press highlighting autonomous vehicle applications
• Extension published as Adaptive Prompt Evolution (SN Computer Science, 2026)

Key Features:

• Task-ID inference using transformer embeddings
• Domain adaptation without task labels
• Scalable to multiple domains