Machine learning interatomic potentials (MLIPs) have emerged as transformative tools in atomistic modeling, offering quantum‑level accuracy at a fraction of the computational cost. Traditional first‑principles methods, such as density functional theory (DFT), are highly accurate but limited by system size and timescale constraints. Conversely, empirical force fields enable large‑scale simulations but often lack the transferability and generality required for complex chemical environments. MLIPs fill this gap by learning the potential energy surface from DFT‑level data using machine learning techniques, including neural networks, kernel regression, and Gaussian processes [1,2].

Recent years have witnessed rapid advances in MLIP methodologies. Descriptor‑based approaches, such as Behler–Parrinello neural network potentials [1], the Gaussian Approximation Potential (GAP) [3], and the Moment Tensor Potential (MTP) [4], pioneered the field by providing highly flexible representations of atomic environments. More recently, message‑passing neural network frameworks such as SchNet [5] and NequIP [6] have emerged as state‑of‑the‑art approaches. These architectures respect fundamental symmetries by design, yielding highly accurate, transferable, and data‑efficient models across diverse material classes. Meanwhile, general‑purpose MLIPs trained on massive datasets have demonstrated the feasibility of achieving broad elemental coverage [7]. Recently several novel network architectures have been introduced for atomistic simulations, including polynomial machine learning potentials (MLPs) described by polynomial rotational invariants [8], or matrix products framework aiming for encoding many body PES [9]. These efforts highlight the vibrant advancement of MLIPs aiming for broad chemical coverage as well as their physical interpretability and computational cost. 

From an application standpoint, MLIPs have enabled breakthroughs across disciplines, ranging from physics and chemistry to biology and materials science. In energy materials, MLIPs have accurately reproduced DFT results for lithium‑ion battery cathodes, capturing thermodynamic and structural trends with high fidelity [10]. They have also been used to explore the potential‑energy landscapes of large molecular clusters, creating a database of low energy isomers of (CH₃OH)n (n = 15–20) [11]. In nanomaterials, MLIPs have illuminated the growth mechanism of BN nanotube on CNTs [12], while in alloys, they have revealed dislocation dynamics and defect interactions that govern plastic behavior [13]. These studies underscore the growing role of MLIPs in screening and optimizing next‑generation materials for catalysis, electronics, and energy storage.

With MLIPs achieving increasing sophistication and making inroads across diverse fields, a dedicated workshop will serve as a critical platform for dialogue between model developers and application specialists. The proposed CECAM flagship workshop in Taiwan aims to foster such exchanges, bringing together researchers to discuss state‑of‑the‑art methods, explore challenging applications, and chart a path for the next generation of MLIPs. By facilitating direct interactions between theoreticians and end‑user communities, this event will catalyze collaboration, deepen understanding, and guide the design of robust, interpretable, and universally applicable MLIPs for tackling pressing challenges in chemistry, physics, energy, and materials science.