Toyota Technological Institute
Just Landed
New Today
Postdoctoral Researcher in Knowledge Data Engineering for Diffusion and Flow-Based Language Models Toyota Technological Institute in Japan
Degree Level
Postdoc
Field of study
Computer Science
Funding
Fixed-term postdoctoral-equivalent position with annual salary about ¥6,048,000 to ¥6,648,000 in the first year (monthly ¥500,000 to ¥550,000, including a meal allowance), with a bonus of up to ¥500,000 per year from the second year depending on performance, bringing the maximum annual salary to ¥7,148,000. Commuting expenses or housing allowance are not included, but there is a meal allowance of ¥4,000 per month, staff housing is available, and travel support is provided for conference attendan
Deadline
Mar 31, 2027
Country
Japan
University
Toyota Technological Institute
How do I apply for this?
Sign in for free to reveal details, requirements, and source links.
Apply for this position
Keywords
Suggested positions
About this position
Toyota Technological Institute is recruiting a Postdoctoral Researcher in the Knowledge Data Engineering Laboratory for a NEDO project on multimodal platform models for AI robots and physical AI.
The position focuses on fundamental technologies for new language generation AI, especially diffusion language models and flow-based models. Two research directions are highlighted: (A) model analysis and learning algorithms and (B) learning/inference systems and acceleration. Topics include mechanistic analysis of generation and learning dynamics, parallel and non-autoregressive decoding, reinforcement-learning-based post-training, inference-time scaling, sampling strategy design, benchmark construction, distillation, large-scale distributed training, KV-cache optimization, quantization, CUDA/Triton kernels, and communication optimization.
The lab emphasizes publication at top international conferences such as ACL, EMNLP, COLM, NeurIPS, ICLR, and MLSys, and conference travel expenses are supported. Open-source release is also encouraged for system and acceleration results. The project provides access to substantial compute resources, including dedicated GPUs on ABCI and the possibility of training on dozens of nodes with 8× H200 GPUs.
Eligibility: applicants must hold a PhD by the start date (or have it in hand), and should have PyTorch deep learning development experience, basic knowledge of large language models, and the ability to communicate and write in Japanese or English. Prior diffusion/flow-model publications are not required; related experience in LLMs, NLP, deep learning, machine learning systems, HPC, or GPU optimization is welcomed.
Work location is Nagoya, Aichi Prefecture, Japan. The appointment is fixed-term for 1 year, renewable annually up to 3 years. Salary is approximately ¥6.048M–¥6.648M in the first year, with a possible bonus from the second year onward. Staff housing, meal allowance, commuting support, and some relocation support for overseas hires are mentioned.
Applications are accepted by email until 2027-03-31 (first deadline 2026-10-31, then rolling until filled). Required materials include a CV, publication list, about three representative papers, a research summary, a future research plan, and one recommendation letter. An informal online pre-interview is available before applying.
Funding details
Fixed-term postdoctoral-equivalent position with annual salary about ¥6,048,000 to ¥6,648,000 in the first year (monthly ¥500,000 to ¥550,000, including a meal allowance), with a bonus of up to ¥500,000 per year from the second year depending on performance, bringing the maximum annual salary to ¥7,148,000. Commuting expenses or housing allowance are not included, but there is a meal allowance of ¥4,000 per month, staff housing is available, and travel support is provided for conference attendan
What's required
Applicants must have a doctoral degree by the start date (or be expected to have one by then). Required skills include experience developing deep learning models with PyTorch, basic knowledge of large language models, and the ability to communicate and write documents in Japanese or English. Preferred qualifications include knowledge or theoretical understanding of diffusion models or flow-based models, analysis of model internals and interpretability, parallel/non-autoregressive decoding, reinforcement-learning-based post-training, inference-time scaling and sampling design, benchmark/evaluation design, large-scale training or distillation, experience with diffusion/flow models in images/audio/video, Hugging Face or similar libraries, multi-GPU/multi-node distributed training, inference acceleration and memory optimization, CUDA/Triton kernel implementation, communication optimization, and publication or presentation experience at top international conferences. OSS development and community contributions are also welcomed.
How to apply
Prepare the required documents as one PDF or ZIP file and email them to [email protected] with the subject line "PD（知識データ工学研究室）応募書類". You may first contact the lab for an informal online pre-interview and send only a CV if you want to confirm fit before applying. If the attachment is too large, it may be split into two emails after prior notice.
More information can be found here
Ask ApplyKite AI
How do I apply for this?
Sign in for free to reveal details, requirements, and source links.