Job Description
職務内容 (Responsibilities)
- Training large-scale image, video, and multimodal generation models, ensuring advanced performance and scalability.
- Translating cutting-edge research into practical solutions by investigating downstream applications that enhance the model’s business value, and delivering actionable insights to the product team.
- Presenting research findings at premier international conferences, building connections with global research communities.
- 大規模な画像・動画・マルチモーダル生成モデルの開発。
- 市場の動向を調査し、プロダクトチームと連携し研究成果を実用的なプロダクトに繋げる。
- トップカンファレンスでの研究成果の発表、グローバルな研究者コミュニティとのコネクション構築。
応募資格（必須） (Required Qualifications)
- (Research Scientist) First-author publications on visual or multimodal generation (or related topics), in top conferences and journals (e.g., ICCV, CVPR, NeurIPS, ICML, ICLR, ECCV, ACL, TPAMI, IJCV, or equivalent).
- (Research Engineer) Hands-on experience in large-scale generative model training or fine-tuning.
- Master’s or Ph.D. in Computer Science, Data Science, or a related field, with a strong foundational understanding.
- Proven research (for scientist) or industrial (for engineer) experience in visual or multimodal generation.
- Willingness to work collaboratively within a team.
- Proficiency in PyTorch or similar deep learning frameworks.
- Solid understanding and experience with recent visual generative models (diffusion, flow matching, AR).
- Self-motivated with a passion for continuous learning and staying current with new technologies.
- Willingness to pursue industrial research aimed at practical real-world solutions, not just academic publication.
応募資格（歓迎） (Preferred Qualifications)
- (Research Engineer) Experience with large-scale distributed training (FSDP/DeepSpeed/Megatron-LM), low precision training, and model parallelism.
- Contributions to open-source AI projects or significant GitHub projects related to visual/multimodal generation.
- Knowledge of machine learning techniques for data collection, cleaning, and pre-processing.
- Experience handling non-visual modalities, such as audio, speech, or language.
ミッション (Mission)
- Research Scientist: Design, implement, and evaluate novel algorithms, models, and techniques for visual and multimodal generation.
- Research Engineer: Focus on implementation and optimization, translating research concepts into practical multimodal AI model development, and building/maintaining robust data pipelines.
求める人物像 (Ideal Candidate Profile)
- Alignment with the mission and willingness to take on new challenges.
仕事の魅力 (Benefits of the Role)
- Opportunity to participate in large multimodal generation model training and development projects.
- Apply research to real-world applications with measurable business impact.
- Work with a diverse, international team based in Tokyo.
- Access to the largest computation resources in Japan.
- Competitive compensation package.
雇用形態 (Employment Type)
- Full-time employee (正社員)
備考 (Notes)
- Interviews are primarily conducted online.
- Specific interview details will be provided individually.
勤務時間 (Working Hours)
- Flexible working hours available.
- Standard hours: 9:00 AM – 5:45 PM (1 hour break).
- Overtime: Possible.
賃金・賞与 (Salary & Bonus)
- Monthly salary: ¥541,667 – ¥1,500,000
- Estimated annual salary: ¥6,500,000 – ¥18,000,000 (no strict upper limit)
- Salary includes base + fixed overtime (35 hours).
- Additional incentives possible.
- Overtime pay applicable beyond fixed hours (for non-managerial positions).
諸手当 (Allowances)
- Commuting allowance (up to ¥150,000/month)
- Overtime, late-night, holiday work allowances (general roles)
- Management allowance, late-night, commute allowance (manager roles)
休日・休暇 (Holidays/Vacation)
- Full two-day weekends (Sat/Sun), national holidays
- Year-end/New Year holiday (Dec 29–Jan 3)
- Paid annual leave (6–21 days, depending on hire month)
福利厚生 (Benefits)
- Health, employment, workers’ accident insurance, employee pension
- Additional benefit programs (Benefit One, premium services)
- Savings & welfare insurance, group insurance schemes
- Defined contribution pension (401K)
出社頻度 (Work Location/Frequency)
- Hybrid: Best mix of remote and on-site work (no fixed requirement)
- Average: 1–2 days per week on-site
- Remote is possible for those residing far from the office (pre-approval required)
- Travel costs for on-site work are covered (with limits)