Tag
The paper proposes a surrogate-assisted genetic programming framework for lifelong learning of scheduling rules in dynamic flexible job shop scheduling, using an adaptive two-stage training strategy and task-similarity-based weighting to mitigate catastrophic forgetting while adapting to new tasks.
EvoMO-SR is a new LLM-driven symbolic regression framework where a small LLM (Llama-3.1-8B-Instruct) generates equation skeletons whose coefficients are fitted by an external optimizer, using multiobjective survival selection and substructure guidance to control bloating and improve structural recovery.
This paper proposes an adaptive protection mechanism for evolutionary feature construction in symbolic regression to preserve important constructed features during evolution, demonstrating improved performance on regression benchmarks and credit classification tasks.
A tweet from DAIR AI highlights new research by Sakana AI introducing CEDAR, an LLM-agent method that designs, simulates, and refines complex system-dynamics models via Monte Carlo Tree Search over feedback structures, aiming to automate goal-directed design in artificial life.
MEMENTO introduces a memory-guided memetic framework for evolving robot control programs (code-as-policy), outperforming existing methods like Eureka and REvolve in long-horizon embodied tasks, and demonstrating sim-to-real transfer.
QDEvo integrates Quality-Diversity optimization with LLM-driven heuristic search to overcome mode collapse in automated heuristic design, outperforming state-of-the-art methods on benchmarks and real-world applications.
This paper introduces Minimalist Genetic Programming (MGP), a novel algorithm that replaces evolution with a syntactic derivation process inspired by the Minimalist Program from linguistics, using a MERGE operator to construct symbolic expressions. MGP consistently finds exact ground truth models on symbolic regression tasks where standard GP struggles due to bloat.
The article discusses new research from Sakana AI and Meta on self-improving AI agents, specifically the Darwin-Gödel Machine and Hyperagents, which autonomously rewrite their own code and infrastructure to enhance performance without human intervention.
This paper proposes Dual-Scale Evolutionary Policy Training (DEPT) to address the evolution impasse in social language agents, using asymmetric advantage reshaping to restore gradient signals during self-play.