Structured Output
Structured output in machine learning focuses on generating outputs with predefined formats and structures, improving the reliability and usability of AI systems. Current research emphasizes enhancing large language models (LLMs) to produce structured outputs like JSON, code, or tables, often employing techniques like retrieval-augmented generation (RAG) and constrained decoding to improve accuracy and efficiency. This area is crucial for deploying LLMs in real-world applications requiring precise and interpretable results, addressing challenges such as hallucination and bias while improving the overall reliability and trustworthiness of AI systems.
Papers
Controlling the Output of a Generative Model by Latent Feature Vector Shifting
Róbert Belanec, Peter Lacko, Kristína Malinovská
Scalable Federated Learning for Clients with Different Input Image Sizes and Numbers of Output Categories
Shuhei Nitta, Taiji Suzuki, Albert Rodríguez Mulet, Atsushi Yaguchi, Ryusuke Hirai