Machine Learning Research Engineer (HYPERCONNECT AI | Military Service Exemption for R&D Personnel Transfer/Switch Possible)

AI Lab

Hyperconnect AI Lab innovates user experience by identifying and solving problems in products that connect people, which are difficult to approach with existing technologies but can be solved with machine learning.

To this end, we aim for the technology created by AI Lab to contribute to the growth of actual products by developing numerous models in various domains, including video/audio/natural language/recommendation, and solving problems encountered while stably providing them through mobile and cloud servers.

Under this goal, Hyperconnect AI Lab has been developing machine learning technologies that contribute to Hyperconnect's products, including Azar, for several years.

ML Research Engineer Role Introduction

An ML Research Engineer at AI Lab requires research capabilities as a scientist who researches and improves cutting-edge models, and engineering capabilities as an engineer who makes technical decisions by comprehensively considering realistic constraints.

Based on these capabilities, they discover/listen to problems encountered in actual products, translate them into AI problems, and solve them.

In this process, they actively collaborate with and receive help from various specialized organizations such as backend/frontend/DevOps/ML software engineers, data scientists/analysts, and PMs. For more detailed stories about how we work, please refer to the following:

• AI in Social Discovery(Blending Research and Production)
• [[How AI Lab Works] Head of AI - Shurain Interview](https://career.hyperconnect.com/post/64d1ffea7b0f3d00016fc4e7)

ML Research Engineers view the product problem-solving process as a research process. They proactively manage the A-Z of the problem-solving process, from problem definition, stakeholder persuasion, goal setting, deriving SotA models, schedule management, performance analysis, to future strategy setting.

They define priorities considering user needs and business impact, and contribute to product growth from an AI perspective with a long-term vision.

Some of the work achievements are also published externally as papers or open-source code. When creating ML models for product use, existing research is often insufficient. To fill these gaps, the results of the research are collaboratively organized by all project participants, and if possible, released with code. As a result, we have achieved about 20 external research achievements to date, including the following:

• 2024: CUPID: Real-time session-based reciprocal recommendation system for 1:1 social discovery platforms presented at ICDM Workshop
• 2023: TiDAL: Active learning technique based on model behavior during efficient training published at ICCV 2023
• 2023: Research on setting thresholds to satisfy multiple classification criteria simultaneously in a moderation environment published at WSDM 2023
• 2022: Research on increasing semantic diversity in dialogue generation published at EMNLP 2022
• 2022: Effective learning method in environments with severe label noise published at ECCV 2022
• 2022: Chatbot research mimicking a target character using only a few utterances from the target character published at NAACL 2022
• 2022: Research on improving performance by utilizing examples in dialogue generation models presented at ACL 2022 Workshop
• 2022: Research on distillation technology for audio classification in mobile environments published at ICASSP

To solve business problems with AI, a proper infrastructure for deep learning training must also be in place. At Hyperconnect, we have built and utilize our own deep learning research cluster to allow ML Research Engineers to sufficiently develop models and conduct experiments.

Various on-premise equipment, including a total of 160 A100 GPUs and multiple H100 GPUs, can be used for research and development. Additionally, we have built and operate our own data pipeline, including data collection and preprocessing, using cloud services.

Contents Understanding Role Introduction

In the contents understanding work at AI Lab, we focus on extracting useful information for business by taking unstructured data consisting of video, images, audio, and natural language as input, with the main goal of contributing to Trust & Safety operations through moderation (Interview). We contribute to understanding content generated across Hyperconnect and Match Group brands and perform various tasks in collaboration with Match Group to meet global Trust & Safety standards. To this end, we are very interested in addressing the following AI problems:

• Lightweight model design and optimization techniques that can achieve high accuracy while maintaining short latency and low power consumption in mobile and web environments.
• Active learning, core-set selection, semi-/self-supervised learning methods to track and manage label quality in noisy and imbalanced data, and secure performance with minimal labeling.
• Multi-task or multi-label classification optimization within a limited parameter budget, and modeling techniques that integrate multi-modal information such as text, images, and video.
• Domain adaptation to overcome distribution differences between domains, and meta-learning techniques for service scalability.
• Learning methods to ensure Fairness and Privacy to meet international AI standards.
• Streaming-based modeling that utilizes user behavior logs and content analysis information to detect or predict abnormal behavior such as spam and fake accounts in real-time.
• LLM utilization methods to innovate ML production processes.

Requirements

• Understanding of the entire AI/ML domain and in-depth knowledge of at least one specific domain.
• Strong interest in productizing AI technology.
• Problem-solving skills using AI and project management skills for this purpose.
• Sufficient Python development skills, including development capabilities based on open-source frameworks such as Tensorflow, PyTorch, CatBoost, JAX.
• Ability to take strong ownership and complete projects from A-Z.
• Strong communication skills to collaborate with stakeholders from various job functions.
• Engineering capabilities to understand the software development structure and content of ML systems and plan features.
• Ability to discover statistical characteristics and patterns in data and apply them to AI problem-solving.
• Understanding of A/B testing and SQL-based data analysis skills.
• Regardless of academic background or nationality, fluent communication in Korean is required.

Preferred Qualifications

• Publication record in top-tier machine learning conferences and journals (NeurIPS, ICLR, ICML, CVPR, ICCV/ECCV, KDD, ...) or awards in AI-related competitions.
• Extensive knowledge across the entire AI/ML domain.
• Experience integrating AI technology into actual services and significantly improving key metrics.
• Experience as a PO/PM or equivalent.
• Fluent communication in English.
• Deep understanding of causal analysis (DID, RCT, Causal Inference, etc.), multivariate testing, and Sequential Testing.

Employment Type/Recruitment Process

• Employment Type: Full-time
• Recruitment Process: Document Screening > Coding Test/Assignment > 1st Interview > Recruiter Call > 2nd Interview > Final Offer (* The process may be added or changed if necessary.)
• For document screening, only successful candidates will be notified individually.
• Application Documents: Free-form detailed English resume based on career (PDF)
• This position allows for transfer/switch to military service exemption for R&D personnel (active duty or supplementary service). For military service special exemption personnel, service management will proceed according to military service special exemption related laws.