Algorithms for Deciding the Safety of States in Fully Observable Non-deterministic Problems: Technical Report

Researchers have made breakthroughs in image generation and safety guarantees for AI decision-making, introducing new algorithms and models that improve performance and reliability.

A team of researchers has developed a new algorithm, iPI, to decide the safety of states in fully observable non-deterministic problems, addressing the lack of safety guarantees in learned action policies^[1]. The iPI algorithm combines the best of both TarjanSafe's effective runtime and a linear-time alternative, guaranteeing a polynomial worst-case runtime. Meanwhile, advancements in image generation have been made with the introduction of Qwen-Image-2.0-RL, a post-training pipeline that applies reinforcement learning from human feedback and on-policy distillation to improve the Qwen-Image-2.0 diffusion model. Qwen-Image-2.0-RL achieved a 57.84 overall score on Qwen-Image-Bench, a 2.61 increase over the base model^[2]. Additionally, HunyuanImage 3.0, a native multimodal model, has been developed, unifying multimodal understanding and generation within an autoregressive framework. The model comprises over 80 billion parameters, with 13 billion activated per token during inference^[3].