Mr. Machine Learning Engineer

The Hyperconnect Machine Learning Engineer (MLE) finds and solves problems that are difficult to approach with traditional technologies in services connecting people, innovating user experiences through machine learning technology. We develop numerous models across many domains, including video, audio, and text, and aim to make our technology contribute to the growth of real services by resolving research topics encountered while providing them stably through mobile and cloud servers.

Under such a goal, Hyperconnect's ML Engineer has been advancing machine learning technologies that contribute to various Hyperconnect products such as Azar and Hakuna for years and is researching on how to easily utilize these accumulated technologies in various global business services.

As ML Engineer, you need to research and improve cutting-edge models as a scientist and develop computational skills as an engineer to maximize inference performance while considering the temporal/spatial complexity of the resulting model. Based on these abilities, we perform diverse roles, such as identifying problems in real services, reproducing or developing SotA (State of the Art) models to solve problems, deploying the models in both on-device and server environments, and monitoring and continuously improving models to build an AI flywheel. In this process, we actively collaborate with various specialized teams such as backend/frontend/DevOps engineers, data analysts, and product managers. You may want to refer to the following content for a more detailed description of how we work.

•

•

Organizing the results of research into papers or codes is also one of our team goals. While creating machine learning models for use in products, there are many cases in which existing research can be insufficient. Collaborative results with project participants will be organized into meaningful sections of the research and will be released with the code if possible. As a result, we have achieved the following external research results so far.

• 2023

• 2022

• 2022

• 2022

• 2022

• 2022

• 2021

• 2021

• 2021

• 2020

• 2020

• 2019

• 2019

• 2019

• 2nd place in the Low Power Image Recognition Competition (LPIRC) 2018

In order for ML research to proceed well, the infrastructure for deep learning must also be well equipped. Hyperconnect has built its own deep learning research cluster for ML engineers to fully conduct model development and experiments. You can utilize various on-premise equipment, including the . Additionally, we actively use Kubeflow pipeline and BigQuery for production pipelines, data collection and pre-processing, and serving. Also, we are working with various software engineers (backend/frontend/DevOps/MLSE) who will assist in productizing the ML model.

Hyperconnect is making various efforts to apply machine learning technology to its products. The Hyperconnect ML Engineer will perform duties in one of the following three fields.

You will provide a better experience to users and, ultimately contribute to long-term revenue growth by solving various recommendation problems included in the product. We are looking for individuals who can solve the following problems. ()

• Cold-start recommendation problem to give new users a good experience (systems that can understand user preferences only with few-shot data, such as session-based recommendation, graph-based recommendation, contextual bandit, and learning methods to improve the recommendation performance for new users when there is insufficient data)

• Reciprocal recommendation problem where both users can be satisfied

• Real-time recommendation problem that performs inference in a very short time on real-time changing recommendation candidates (session-based recommendation, graph-based recommendation, reinforcement learning, …)

• Recommendation problem considering the trade-off between several objective indicators

• Problem of finding a primary objective indicator that improves long-term indicators

We perform various technologies and research and development for understanding the content of the content in order to satisfy users' experience. We are looking for those who can work together to solve the following problems by accepting unstructured data consisting of video, audio, and natural language as input and extracting useful information for decision making. ()

• The problem of lightweight models and optimizations that can speed up in mobile environments

• The problem of efficient and label-adjustable multi-task or multi-label models

• The problem of using partial multi-modal data

• Problem of detecting anomalous users (ex. spam/fake accounts) in real-time based on user behavior logs and content understanding results flowing in the stream

• Problem of efficient data labeling method through active learning or core-set selection method to reduce data needed for model learning

We offer new experiences that have never been before to users through various generative AI research and development. We make tools that allow users to easily create personalized content and express themselves in the service, and develop new features using generative AI.