Track synthesis, materials and catalysis as it is published.
Search papers, labs, and topics across Lattice.
This paper introduces a three-agent LLM system (analyst, coder, recorder) for automated tuning of specialized neural network components within a fixed architecture, operating on an open but semantically narrow code generation space. The framework demonstrates significant performance gains across diverse tasks, including spiking neural network classification and PINN-based differential equation solving, outperforming baselines by substantial margins in accuracy and error reduction. This approach offers a transferable method for optimizing critical hyperparameters and structural elements of neural networks without requiring full script optimization or discrete architecture search.
Automated LLM agents achieve state-of-the-art performance by tuning specialized neural network components, outperforming baselines by up to 2491x on differential equation solving and 13pp on classification.
Many machine learning approaches are two-level: an expert selects a class of methods, within this class the choice of key components is critical and usually manual. This paper addresses the automatic tuning of these components and proposes a framework transferable across models of different nature. The multi-agent large language model (LLM) framework has three roles---analyst, coder, and recorder~---with an experiment log and a sandboxed execution environment. Unlike systems that optimize the entire ML script or search architectures in a discrete space, the framework operates over an open but semantically narrow space: the agent generates code only for the specified components, leaving the overall network structure unchanged. The approach is validated in two directions: tuning a classification spiking neural network with correlation learning, and solving high-order differential equations with Physics-Informed Neural Networks (PINN). Across five tasks---spiking classification on MNIST and Digits, and boundary-value problems for the generalized nonlinear Schr\"odinger equation (GNLSE) and Kuramoto--Sivashinsky (KS) equations---the agent outperforms the baselines: classification accuracy is 13 percentage points (MNIST) and 6 percentage points (Digits) higher; the error against the analytical single-soliton solution is reduced approximately 2491-fold; the invariant-preservation error drops approximately 325-fold in the selected GNLSE run with a Gaussian perturbation and by more than 100-fold on average over repeated initializations; and the mass drift drops 14-fold for the KS problem with a Gaussian perturbation. The framework thus tunes key components on tasks of different nature without changing its internal structure.