OMRON SINIC X Corporation (Headquarters: Bunkyo-ku, Tokyo; President and CEO: Masaki Suwa; hereinafter referred to as "OSX") is pleased to announce that its research paper has been accepted for the Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS 2023), a top conference in the field of AI and machine learning. Another OSX paper was accepted for oral presentation at the NeurIPS 2023 AI for Science workshop, held with NeurIPS 2023. The papers will be presented at the international conference and workshop, held in New Orleans, Louisiana, USA from December 10 to 16, 2023, respectively.

NeurIPS is one of the world’s leading conferences on AI and machine learning. 26.1% of papers were accepted out of 12,343 submissions in 2023. Of the 153 accepted papers for the workshop, only ten papers were chosen for oral presentation.

A paper accepted for NeurIPS 2023 presents the results of an algorithm that generates near-optimal time-sequential data in offline reinforcement learning, even when optimal time-sequential data cannot be collected for training. Research on a Transformer model, which searches for mathematical formulas from data to promote new scientific discoveries, is presented at the workshop. Some of these research findings have been released in articles with a simple explanation of the corresponding technology and on GitHub as open-source code so that they can be actively used for social implementation. For details, please access the link for each paper.

OSX will continue its efforts to create new value creation by technological innovation through collaboration with universities and external research institutions, and the recruitment of interns.

<Paper accepted for NeurIPS 2023> 　*Presentation time is in local time.

Title	Elastic Decision Transformer
Authors	Yueh-Hua Wu (UC San Diego; OSX intern from December 2022 to March 2023), Xiaolong Wang (UC San Diego), Masashi Hamaya (OSX)
Time of presentation	15:00-17:00, Wednesday, December 13, 2023
Overview	In Reinforcement Learning (RL), an agent is trained on how to act while interacting with the environment to maximize a cumulative reward. RL is used in various fields, including motion control of robots, automated driving in mobility services, game AI, and so on. Offline RL is an RL paradigm that uses only pre-collected datasets. Because this approach eliminates the need to interact with the environment in real time, it can be expected to reduce the cost of data collection. In recent years, Decision Transformer (DT), a framework that extends a Transformer architecture to offline RL, has attracted attention for its high performance. The Transformer architecture is widely used in fields such as natural language processing and computer vision. DT addresses as a time-sequential model to predict and determine the next action based on past time-sequential data. In offline RL, optimal data cannot always be collected. Therefore, it is required to generate optimal (or near-optimal) time-sequential data transitions (trajectory) by stitching together non-optimal time-sequential data (trajectory stitching). However, trajectory stitching may not achieve sufficient performance because DT cannot consider how much reward can be obtained in the short and long term. This paper proposed Elastic Decision Transformer (EDT) that takes a variable length of past time-sequential data (input length) for trajectory stitching in DT. When a short time length is employed as the input, the variations in the output become larger. This leads to the selection of outputs with higher rewards by actively promoting exploration. When the trajectory is already optimal, the model employs a long time length as the input to maintain the stability and consistency. EDT thus improves the performance of trajectory stitching by taking the input length variable for each time step. EDT achieves superior performance in a wide range of application fields, situations, and tasks by conducting benchmark tests of RL in simulation. EDT contributes to diverse applications such as robotics assembly of unknown parts, vehicle control and improved environmental awareness in automated driving.
Details	https://medium.com/sinicx/elastic-decision-transformer-e8578b7d218f https://kristery.github.io/edt/ https://openreview.net/forum?id=RMeQjexaRj

<Paper accepted for NeurIPS 2023 AI for Science workshop>　*Presentation time is in local time.

Title	A Transformer Model for Symbolic Regression towards Scientific Discovery
Authors	Florian Lalande (OIST; OSX intern from April 2023 to September 2023, from November 2023 to December 2023), Yoshitomo Matsubara (Amazon Alexa), Naoya Chiba (Tohoku University), Tatsunori Taniai (OSX), Ryo Igarashi (OSX), Yoshitaka Ushiku (OSX)
Time of presentation	16:05-16:10, Saturday, December 16, 2023
Details	https://github.com/omron-sinicx/transformer4sr https://openreview.net/forum?id=AIfqWNHKjo

About OMRON SINIC X Corporation
OMRON SINIC X Corporation is a strategic subsidiary seeking to realize the "near future designs" that OMRON forecasts. Researchers with cutting-edge knowledge and experience across many technological domains, including AI, Robotics, IoT, and sensing, are affiliated with OSX, and with the aim of solving social issues, they are working to create near future designs by integrating innovative technologies with business models and strategies in technology and IP. The company will also accelerate the creation of near future designs through joint research with universities and external research institutions. For more details, please refer to https://www.omron.com/sinicx/en/

For press inquiries related to this release, please contact the following:: Tech Communications and Collaboration Promo Dept.
Strategy Division
Technology and Intellectual Property H.Q.
OMRON Corporation
TEL: +81-774-74-2010

OMRON SINIC X Research Paper Accepted for NeurIPS 2023, a top AI and Machine learning conference