COEFF-KANs: A Paradigm to Address the Electrolyte Field with KANs

TL;DR

The paper tackles the challenge of rapidly predicting Coulombic Efficiency (CE) for electrolyte formulations to reduce experimental validation workload in lithium metal batteries. It introduces COEFF, a two-stage framework that pre-trains a chemical general model and then fine-tunes on domain data, using MoLFormer to extract embeddings for each electrolyte component and a weighted pooling by component ratios to form formulation features, which are then processed by either an MLP or Kolmogorov–Arnold Network (KAN) to predict CE. A key contribution is applying KANs to electrolyte CE prediction, achieving state-of-the-art RMSE on a real-world 13-sample test set and demonstrating data-efficient learning with a relatively small fine-tuning dataset, aided by the CIDO input-disentangled pooling strategy. The results suggest substantial potential to accelerate electrolyte design by reducing experimental workload and enabling broader, more reliable in silico screening across formulations and related properties.

Abstract

To reduce the experimental validation workload for chemical researchers and accelerate the design and optimization of high-energy-density lithium metal batteries, we aim to leverage models to automatically predict Coulombic Efficiency (CE) based on the composition of liquid electrolytes. There are mainly two representative paradigms in existing methods: machine learning and deep learning. However, the former requires intelligent input feature selection and reliable computational methods, leading to error propagation from feature estimation to model prediction, while the latter (e.g. MultiModal-MoLFormer) faces challenges of poor predictive performance and overfitting due to limited diversity in augmented data. To tackle these issues, we propose a novel method COEFF (COlumbic EFficiency prediction via Fine-tuned models), which consists of two stages: pre-training a chemical general model and fine-tuning on downstream domain data. Firstly, we adopt the publicly available MoLFormer model to obtain feature vectors for each solvent and salt in the electrolyte. Then, we perform a weighted average of embeddings for each token across all molecules, with weights determined by the respective electrolyte component ratios. Finally, we input the obtained electrolyte features into a Multi-layer Perceptron or Kolmogorov-Arnold Network to predict CE. Experimental results on a real-world dataset demonstrate that our method achieves SOTA for predicting CE compared to all baselines. Data and code used in this work will be made publicly available after the paper is published.

Figures (5)

• Figure 1: Overview of our method's two-stage paradigm: pre-training a chemical general model and fine-tuning on downstream domain data.
• Figure 2: Overview of our proposed COEFF framework.
• Figure 3: Box plots depict the distribution of LCE outputs for training data based on the formulation constituent count. The outer whiskers represent the minimum and maximum values, the central line represents the median, and the colored box represents the 25th to 75th percentile of the data. Data points outside the outer whiskers are outliers observed in the data.
• Figure 4: Parity plots show predicted LCE values as scatterplots with respect to the actual values.
• Figure 5: RMSE results of LCE prediction at different network depths and widths.