Can Polat
Ph.D. Candidate at Texas A&M University / Kurban Intelligence Lab / AI for Science

Can Polat.

I build machine learning systems that accelerate materials discovery and the science underneath it.

I earned my B.Sc. in Applied Physics and M.Sc. in Physics and am currently pursuing a Ph.D. in Computer Engineering at Texas A&M University. As a researcher at the Kurban Intelligence Lab, I work at the intersection of generative AI and materials science. My research spans geometric deep learning for crystals and molecules, multimodal representations for chemistry, and the kind of physics-aware modeling that survives outside a benchmark. Our work has appeared at ICLR, ICML, KDD, ACL, CVPR, and Q1 journals with impact factor 13+.

Portrait of Can Polat
Email
can.polat0@yahoo.com

Selected Work

All publications →
Topic Venue

ML for Materials

Building machine-learning systems that represent, screen, and reason about crystals and molecules at the speed and scale wet-lab discovery cannot match. Most of this work lives at the intersection of geometric deep learning, multimodal representations, and the chemistry constraints that let models generalize beyond their training distribution — including the benchmarks and foundation-model stress tests that reveal where they actually hold up.

arXiv
A single design choice determines whether machine learning models of materials make physically impossible predictions
Open MIND
equiparity: code and measurement records for "A single design choice determines whether machine learning models of materials make physically impossible predictions"
arXiv
When do machine-learned exchange-correlation improvements inherit into density-functional tight binding?
KDD
How Far Can You Grow? Characterizing the Extrapolation Frontier of Graph Generative Models for Materials Science
npj Computational Materials
CliffordIP: Clifford algebra equivariant interatomic potentials for heterogeneous catalysis
arXiv
Geometric Algebra Meets Cartesian Tensors: Higher-Order Equivariance for Interatomic Potentials
arXiv
STEMGym: Benchmarking Sequential Decision-Making under Dose Budgets in Autonomous Electron Microscopy
Open MIND
QuantumCanvas: A Multimodal Benchmark for Learning Two-Body Quantum Interactions
Open MIND
RADII: Radius-Resolved Benchmark of Nanoparticle Structures for Generative Models in Materials Science
arXiv
SCALAR: Quantifying Structural Hallucination, Consistency, and Reasoning Gaps in Materials Foundation Models
arXiv
C2NP: A Benchmark for Learning Scale-Dependent Geometric Invariances in 3D Materials Generation
Computational Materials Science
QuantumShellNet: Ground-state eigenvalue prediction of materials using electronic shell structures and fermionic properties via convolutions
arXiv
QuantumCanvas: A Multimodal Benchmark for Visual Learning of Atomic Interactions
Mach. Learn.: Sci. Tech.
CrysMTM: a multiphase, temperature-resolved, multimodal dataset for crystalline materials
Physica Scripta
Enabling ease of access to quantum chemistry with transformer-based text encoding and physics-informed multilayer perceptron
ICML
Beyond Atomic Geometry Representations in Materials Science: A Human-in-the-Loop Multimodal Framework
CVPR
Understanding the Capabilities of Molecular Graph Neural Networks in Materials Science Through Multimodal Learning and Physical Context Encoding
ICLR
TDCM25: A multi-modal, multi-task benchmark for temperature-dependent crystalline materials
ACL
Stress-Testing Multimodal Foundation Models for Crystallographic Reasoning
Mach. Learn.: Sci. Tech.
Multimodal neural network-based predictive modeling of nanoparticle properties from pure compounds

Energy Storage

Designing and screening nanomaterials for hydrogen storage and energy applications — H₂ adsorption on single-atom-doped TiO₂, engineered MgO and copper nanoclusters — pairing electronic-structure calculations with data-efficient predictive models to find the descriptors that govern reversible uptake.

Scientific Reports
Descriptor-guided thermodynamic screening of H₂ adsorption on single-atom-doped anatase TiO₂ nanoparticles with interpretable machine learning
ACS Omega
Engineered MgO Nanoparticles with Tunable Electronic Signatures for Energy Applications
IEEE CPE
Data-Efficient Hydrogen Adsorption Prediction in Copper Nanoclusters: A Computer Vision-Based Transfer Learning Approach
Computational Materials Science
Enhancing the electronic properties of TiO2 nanoparticles through carbon doping: An integrated DFTB and computer vision approach

Explainable AI

Making the models interpretable to the scientists who use them — symbolic regression, feature attribution, and agentic reasoning that turn black-box predictions into orbital-level design rules and quantum-chemistry explanations a domain expert can interrogate and trust.

Int. J. Hydrogen Energy
Multimodal explainable artificial intelligence-driven analysis of quantum size effects in copper nanoclusters for hydrogen storage
arXiv
Grounded verification of chemical and materials reasoning: detection is the bottleneck
Materials & Design
Explainable machine learning reveals orbital design rules for reversible H2 adsorption on single-atom-doped TiO2 nanoparticles
ICML
xChemAgents: Agentic AI for Explainable Quantum Chemistry

Micromachining

Optical systems and the algorithms that interpret what they capture — from my MSc-era work at Elahi Lab on noise-robust focus detection for laser material processing, through interferometric microscopy, to real-time focus positioning on rough surfaces. The thread is the same: when the camera and the model are designed together, things you couldn't measure before become measurable.

IEEE SIU
A Transformer-Based Real-Time Focus Detection Technique for Wide-Field Interferometric Microscopy
Optics & Lasers in Eng.
High-precision laser focus positioning of rough surfaces by deep learning
SPIE
Noise robust focal distance detection in laser material processing using CNNs and Gaussian processes
SPIE
Machine learning-based high-precision and real-time focus detection for laser material processing systems
CLEO
Noise robust high precision and real-time focus detection for laser micromanaging