Artificial intelligence (AI) specialist ACRYL (CEO Jin Park) announced on the 8th that it will present two large-scale language model (LLM)-related papers through joint research with Professor Woo Hong-wook's research team at Sungkyunkwan University at 'NeurIPS 2024' and 'EMNLP 2024'.
NeurIPS and EMNLP are the top academic conferences in the field of AI and natural language processing (NLP), where researchers from around the world gather every year to share and discuss the latest research results.
This paper presentation contains the results of applying the LLMops solution based on ACRYL's flagship product, 'ALLM (ACRYL LLM),' to various industries such as smart home, autonomous driving, healthcare, and robot control.
The research presented by NeurIPS proposes a new ExRAP (Exploratory Retrieval-Augmented Planning) framework to enhance the performance of physical agents that must continuously perform commands in a changing environment.
Overcoming the limitations of existing simple command-based models, the technology that allows agents to explore the environment and efficiently re-plan commands according to context was presented through LLM. ExRAP simultaneously performs common commands and shortens execution time.
In particular, it is explained that it can respond quickly to environmental changes in control systems such as home robots and autonomous vehicles. It emphasized that it is a technological foundation for entering the future robot control market.
(Photo = ACRYL)
In EMNLP, we will introduce the COREN framework that maximizes the learning efficiency of reinforcement learning-based agents by utilizing LLM as a learning reward estimation tool. COREN is a method that does not use LLM directly as an agent, but increases the precision of reward estimation so that agents can quickly learn and make decisions in complex environments.
Using the structure of COREN and LLM, reward values are estimated based on contextual, structural, and temporal consistency. It is explained that it can improve the autonomous learning performance of industrial robots or smart devices used in various industries such as manufacturing, healthcare, and education.
(Photo = ACRYL)
ACRYL CEO Jin Park said, “This paper presentation is an important opportunity to inform the world about ACRYL’s LLM commercialization technology capabilities,” and added, “We will expand our business to the healthcare, autonomous driving, and smart home industries based on technology utilizing LLM.”
He also said, “We plan to develop LLM not only for robot control, but also for information retrieval and decision-making in unstructured data environments, and as an on-device solution to apply it to more industries.”
Meanwhile, this research was conducted with the support of the ‘Human-Centered Artificial Intelligence Core Source Technology Development Project’ promoted by the Ministry of Science and ICT and the Institute of Information and Communications Technology Planning and Evaluation (IITP).
Source: AI Times (https://www.aitimes.com)