Prompt Recovery
Prompt recovery focuses on reconstructing the input instructions or prompts used to generate outputs from various models, particularly in image generation and large language models. Current research emphasizes developing robust algorithms, including Q-networks, discrete optimizers, and diffusion models, to effectively recover prompts from limited output information, even in the presence of unforeseen malfunctions or adversarial attacks. This field is crucial for enhancing the safety and reliability of autonomous systems, improving the interpretability of AI models, and addressing concerns about privacy and copyright in AI-generated content.
Papers
Collaborative Adaptation for Recovery from Unforeseen Malfunctions in Discrete and Continuous MARL Domains
Yasin Findik, Hunter Hasenfus, Reza Azadeh
Relational Q-Functionals: Multi-Agent Learning to Recover from Unforeseen Robot Malfunctions in Continuous Action Domains
Yasin Findik, Paul Robinette, Kshitij Jerath, Reza Azadeh
Learning to Recover from Plan Execution Errors during Robot Manipulation: A Neuro-symbolic Approach
Namasivayam Kalithasan, Arnav Tuli, Vishal Bindal, Himanshu Gaurav Singh, Parag Singla, Rohan Paul
Resilient Average Consensus with Adversaries via Distributed Detection and Recovery
Liwei Yuan, Hideaki Ishii