- InsightsInsights
- CryptocurrenciesCryptocurrencies
- Stocks
- White Papers
- IndustryIndustry
- GeographyGeography
A few days back LG Group announced that it will invest more than US$100 million in developing mega-scale AI and creating a general-purpose artificial neural network. The LG AI Research, which is the AI R&D Lab of the company said it was planning to build an expert AI Computing system that can replicate the human brain and conduct comprehensive and independent learning. In an online event on May 17, the company disclosed that its AI system will constitute enhanced data inference capabilities, with more than 600 machine learning parameters. The system would be more efficient than the widely used GPT-3.
LG AI Research is based in Seoul and was launched in 2020 with a vision to develop an optimal environment for AI innovations. The initiative aims to leverage advanced AI technology to carry out innovative research with the best AI experts. According to reports, LG AI Research is also planning to develop an AI model with trillion scale parameters by next year along with 16 group affiliates. Further, the report states that the group is experimenting with building an AI model that can perform sales activities for B2B customers and considers the AI system to aid customer counseling and production development. A report by Aju Business Daily revealed that the new AI system put forth by LG AI Research can efficiently analyze customer feelings in conversations when applied to customer counseling chatbots and can exceedingly reduce the process of developing products.
This is not the first innovation LG AI Research lab has flag shipped. The company has advanced the fundamental research in AI to develop state-of-the-art technologies in the research fields. They are consistently working towards developing transformational AI that can be used for real-world applications so that it reaches more people. Parallel computing, continual learning, generative models to solve data shortages, AI for material design, and Explainable AI are some of their AI projects in this milieu.
LG AI Research and the University of Toronto jointly developed an Explainable AI algorithm that helps detect and eliminate defects in display screens. The algorithm is named Semantic Input Sampling for Explanation (SISE) and can be deployed to understand how machine learning makes its decisions, including medical image interpretation. This achievement can be considered as the first major innovation by the research center. This Explainable AI algorithm will provide evidence to humans along with intelligent decisions, which will enable them to comprehend the decision-making process better. This technology could also act as a replacement for human-driven decisions in several significant sectors.
The innovation aims to address the black-box issue in the machine learning models, which can invite biases in the decisions. In the usual scenario, AI models learn from a labeled data set and correlate them to conclude. This method can be challenging for sensitive domains like healthcare, crime detection, and law, financial institutions, etc. This study put forth by the LG AI Research and the University of Toronto has a glass-box model, which displays the decision-making process. It has a hybrid model that combines backpropagation and perturbations. The experimental results, according to the paper, showed that this method could reduce the execution time by up to 30% and enhance comprehensive interpretability by not staking the quality of decision and explanation.
Another research paper published on continual learning enables AI systems to continuously learn sequential data, analyze the data from past learning, and comprehend it. Another experiment by the research team revealed a generative model that can perform state-of-the-art video prediction. The method predicts future frames by first estimating a sequence of semantic structures and then translating them to pixels by video-to-video translation, the publication reveals. It enables a magnitude longer prediction time compared to the conventional existing models. Recently, the company published another research presenting i-Max, a strategy for improving contrastive representation learning through domain-agnostic regularization. It shows how the model consistently improves the quality of learned representations across various domains. LG AI Research is serving its mission by enabling cutting-edge innovations in artificial intelligence and making them available to a wider audience.
