Kakao Brain: Accelerating large-scale natural language processing and AI development with Cloud TPU
Customer Company Size
Large Corporate
Region
- Asia
Country
- Korea
Product
- KoGPT
- Google Cloud TPU
Tech Stack
- Tensor Processing Unit (TPU)
- Generative Pre-trained Transformer 3 (GPT-3)
Implementation Scale
- Enterprise-wide Deployment
Impact Metrics
- Productivity Improvements
- Innovation Output
Technology Category
- Analytics & Modeling - Machine Learning
- Analytics & Modeling - Natural Language Processing (NLP)
Applicable Functions
- Product Research & Development
Use Cases
- Generative AI
- Machine Translation
Services
- Cloud Planning, Design & Implementation Services
About The Customer
Established in 2017, Kakao Brain is a research and development company that develops AI-based technologies in various fields, such as natural language processing. The company recently released a Korean natural language processing model, KoGPT, to help further expand the use and value of AI. As a subsidiary of Kakao Corp., a major South Korean tech giant, Kakao Brain focuses on advancing AI technologies to enhance various applications, including natural language processing. The company is committed to pushing the boundaries of AI research and development, leveraging cutting-edge technologies to create innovative solutions. With a strong emphasis on AI and machine learning, Kakao Brain aims to contribute significantly to the field by developing models that can process and understand complex data, particularly in the Korean language. Their efforts are geared towards creating AI models that can perform tasks such as reading user intentions, writing letters, and even software coding, thereby expanding the scope of AI applications in the Korean language.
The Challenge
In November 2021, Kakao Brain, an artificial intelligence R&D subsidiary of South Korean tech giant Kakao Corp., unveiled KoGPT. A large-scale deep learning-based natural language processing model, KoGPT was developed by adapting Generative Pre-trained Transformer 3 (GPT-3), the most widely used natural language processing model, to the Korean language. When it comes to the English language, GPT-3 is already expanding the scope of application beyond simply translating words into text, by accurately reading a user’s intentions and writing letters, even software coding. This was not available for the Korean language because the process of creating a NLG machine learning model is labor intensive, with rapid learning of large-scale data required. However, KoGPT was able to process six billion model parameters and 200 billion tokens, creating an artificial intelligence model that can understand Korean.
The Solution
Deploying a dedicated machine learning processor optimized for learning large-scale data, according to Woonhyuk Baek, Large-Scale AI Research Scientist at Kakao Brain, Google Cloud TPU plays an important part in accelerating the training process of KoGPT and its massive workloads. Baek goes on to explain that understanding the characteristics of GPU (Graphical Processing Unit) and TPU (Tensor Processing Unit) is the most important starting point for proper utilization. Although TPU has strong AI data processing capabilities, simply replacing all AI systems with TPU immediately will not yield the results wanted. TPU and GPU also have clear areas that complement each other. GPU has the advantages of being able to start a project quickly, and easily respond to a general environment, but it is not easy to scale. Meanwhile, TPU is easy to manage because it can receive resources in units of pods, and the communication speed between each node is fast, which is integral for large-scale data processing. Unlike on-premise and cloud compatible GPU, Cloud TPU was born to accelerate machine learning workloads within the Google Cloud ecosystem. Baek says on-demand TPU devices and pod slices provided ease with workload management, adding that fast networking speeds between TPU nodes made data processing seamless.
Operational Impact
Quantitative Benefit
Case Study missing?
Start adding your own!
Register with your work email and create a new case study profile for your business.
Related Case Studies.
Case Study
Domino’s Enhances Customer Experience with Dialogflow’s Conversational Technology
Domino’s, a global pizza restaurant chain, has been at the forefront of digital innovation, offering customers the ability to order pizzas via 'Dom', an ordering bot available online and through the company’s mobile apps. However, with the changing consumer behavior and the need to innovate, the company faced the challenge of building rich conversational experiences powered by natural language understanding (NLU) and machine learning. The company needed a solution that could scale and accommodate the number of intents needed for a complex ordering process. The challenge was to account for the numerous directions a customer could take when ordering from their extensive menu.
Case Study
Direct Marketing Solution
Marketing to prospects (direct mail, telemarketing, and email) is expensive. Targeting incorrectly can hurt your brand, leaving prospects feeling spammed. Traditional techniques are not very sophisticated, resulting in low response rates which in turn leads to high cost-per-lead/acquisition numbers.
Case Study
Nitrio's Transition to ML-Powered Intent Extraction for Advanced Sales Strategies
Nitrio, an AI company specializing in sales optimization, was facing significant challenges with its Natural Language Processing (NLP) platform. The platform relied heavily on manual rules and heuristics-based models, which led to bottlenecks and scalability issues, hindering Nitrio's growth. The existing platform was unable to ensure the required level of accuracy for sentiment analysis of rep-to-lead messages, resulting in a significant number of messages being outsourced to a third party for manual analysis. This not only increased service costs but also created further bottlenecks and scalability issues. The platform's infrastructure demonstrated tight coupling between services, increasing their dependencies and negatively impacting team performance, causing data quality and consistency issues. Nitrio's platform was designed to efficiently analyze inbound rep-to-lead messages to extract their intent and collect useful data about every sales representative's performance. However, the reliance on manual processes and the inability to ensure 95% certainty in message intent identification were major setbacks.
Case Study
MediaWen International Launches cognitive linguistic solutions with IBM Cloud and IBM Watson technologies
MediaWen International, a B2B and SaaS provider, wanted to launch online linguistic solutions for customers worldwide. The company needed to partner with a global provider that could offer high-performance cloud infrastructure and cognitive computing technologies. The solutions were aimed at providing cognitive subtitling, translation, and dubbing functionality for customers in various sectors including media and entertainment, e-learning, finance, government, and healthcare.
Case Study
Coinbase's Localization Strategy: Redefining Crypto Language for Global Growth
Coinbase, a leading cryptocurrency platform, faced the challenge of localizing and translating its content for global markets. The crypto industry is constantly evolving, with English terminology used to describe crypto concepts frequently changing. This posed a significant challenge for Coinbase's localization experts, who had to redefine existing language and translate it at scale. The company also had to consider the unique jargon of the crypto world, ensuring that their glossaries and style guides included these terms. Another challenge was the translation of certain terms, like 'wallet', which have different meanings in the crypto world and can have varying connotations in different languages. Coinbase also had to ensure that their translated content was relevant, inclusive, and true to the crypto personality and nature.
Case Study
Wine Experiences on Tap
Early in 2014, Vivino reached the point where localization was becoming critical to the company’s success and growth. It boasts an active community of wine-lovers from all over the world and the company knew that localization would be key to further engaging them. Vivino did not have anything in place for translation, so this was an entirely new undertaking. After hearing about Smartling through a partner in the wine industry, Vivino product manager, Christos Iosifidis, took a look at the company’s offerings and agreed that Smartling had the best solution for Vivino’s needs. Christos had experience with translation companies in previous roles and was impressed with the client services Smartling provides.
