#Japan #Tokyo #AI Research #Omron Sinicx #ICLR 2025

Omron Sinicx Presents at ICLR 2025

Omron Sinicx, a pioneering company in the field of cutting-edge technologies including AI and robotics, is set to showcase their latest research findings at the prestigious International Conference on Learning Representations (ICLR) 2025. This conference, held from April 24 to April 28 in Singapore, is recognized globally as a leading event in the machine learning domain, particularly focused on advancements in deep learning and representation learning.

The company will present two key research papers:

Rethinking the Role of Frames for SE(3)-Invariant Crystal Structure Modeling

This study introduces a novel approach to predicting material properties from crystal structures using a transformer-based neural network, CrystalFramer. By expanding the traditional frame method to include the concept of a dynamic frame, the research aims to enhance the precision in capturing three-dimensional information about crystal structures. This advancement follows their previous work on the Crystalformer model, indicating a significant step forward in materials science. The authors, Yusei Ito, Tatsunori Taniai, Ryo Igarashi, Yoshitaka Ushiku, and Kanta Ono from Osaka University, emphasize the importance of developing next-generation materials like superconductors that operate in high-temperature environments and high-performance battery materials. Their work aims to streamline the materials development process using AI technology, which traditionally requires extensive time and trial-and-error.

Read the paper here

Near-Optimal Policy Identification in Robust Constrained Markov Decision Processes via Epigraph Form

The second paper delves into the challenges of applying reinforcement learning in realistic settings. The authors, Toshinori Kitamura, Tadashi Kozuno, Wataru Kumagai, along with collaborators from Kyoto University and The University of Tokyo, propose a unique methodology utilizing epigraph form reinforcement learning. Their research addresses two critical needs: robustness against discrepancies between training data and actual products, and theoretical safety. Through rigorous theoretical exploration and experimental validation, they demonstrate the capability of producing robust and safe policies, marking a significant milestone in the field of reinforcement learning.

Access the paper here

Additionally, during the ICLR 2025 conference, Omron Sinicx will also participate in the AI for Accelerated Materials Discovery (AI4Mat) Workshop, introducing another influential research topic:

Transformer as a Neural Knowledge Graph

This study builds upon their previous research presented at NeurIPS 2024, tackling the issue of limited language data in contrastive learning of crystal structures and language models. The authors propose the use of a Neural Knowledge Graph (NKG) that dynamically integrates relevant knowledge alongside keywords from papers using a transformer. The experimental results indicate that NKG significantly enhances the performance of crystal structure searches based on keywords when compared to traditional methods.

This initiative underscores Omron Sinicx's commitment to advancing material discovery through innovative applications of AI and machine learning. The integration of collaborative research efforts with academic institutions further accelerates their pursuit of futuristic design, aiming to tackle societal challenges through technology.

For more about Omron Sinicx and their innovative projects in AI and materials science, visit their official website here or check out their latest activities here.