---

I am a machine learning engineer and applied scientist with a Ph.D. in Computer Science (UC Irvine, 2022). My work spans vision-language models, multimodal representation learning, and foundation model development, with a strong foundation in scalable production systems built over 10+ years of industry experience.

Currently, I work independently as a freelance applied scientist and ML engineer, helping clients build and evaluate large-scale AI systems. I am particularly drawn to applications where AI can create genuine impact, most recently in geospatial intelligence and Earth observation, exploring how modern multimodal architectures can be adapted for satellite imagery and climate-relevant tasks.

---

Current Work

Earth Intelligence (2026–present) Evaluating state-of-the-art Earth Foundation Models (including Google DeepMind’s AlphaEarth Foundations) to inform internal model strategy. Analyzing architectures, training objectives, and data strategies; re-implementing core components for internal benchmarking.

Visual Search & Recommendation (2026–present) Improving multimodal image-text embedding pipelines using SigLIP2; designing LLM-based data cleaning workflows; leading proof-of-concept development for outfit recommendation using multimodal representations.

---

Experience

Freelance Machine Learning Engineer · Jan 2026–present

Applied Scientist, Amazon · Jun 2022–Jan 2026 Trained and fine-tuned state-of-the-art vision-language models for fashion item retrieval. Introduced a novel retrieval-augmented training methodology for video-based retrieval. Led large-scale dataset curation for product review and video data. Built visual search suggestion datasets for Japan and European marketplaces.

Graduate Student Researcher / Teaching Assistant, UC Irvine · Sep 2017–Jun 2022 Led research on generative world models for model-based reinforcement learning. TA for graduate and undergraduate ML and Python courses.

Developer Support Engineer, Amazon Web Services Japan · Apr 2013–Aug 2017 Supported enterprise BigData infrastructure migrations for clients including Toyota, Sony, and Canon. Resident expert on Hadoop ecosystem within AWS Japan.

Systems Engineer, Hewlett-Packard Japan · Apr 2010–Mar 2013 Engineered and maintained online banking infrastructure for Mitsubishi Tokyo UFJ Bank.

---

Education

Ph.D. Computer Science, UC Irvine, 2022 Specialization: Model-based reinforcement learning and generative models Advisor: Prof. Emre Neftci

M.S. Information Science, Tokyo University of Science, 2010 B.S. Information Science, Tokyo University of Science, 2008

---

Publications

Kim, D., Ranjan, V., Nagata, T., Dhua, A., & Kumar, A. KC. (2025). Rethinking Visual Information Processing in Multimodal LLMs. arXiv:2511.10301

Xing, J., Nagata, T., Zou, X., Neftci, E., & Krichmar, J.L. (2023). Achieving efficient interpretability of reinforcement learning via policy distillation and selective input gradient regularization. Neural Networks 161, 228–241.

Nagata, T., Xing, J., Kumazawa, T., & Neftci, E. (2022). Uncertainty Aware Model Integration on Reinforcement Learning. IJCNN’22.

Xing, J., Nagata, T., Chen, K., Zou, X., Neftci, E., & Krichmar, J.L. (2021). Domain Adaptation In Reinforcement Learning Via Latent Unified State Representation. AAAI’21.

Nagata, T., Takimoto, M., & Kambayashi, Y. (2013). Cooperatively Searching Objects Based on Mobile Agents. Trans. Comput. Collect. Intell., 11, 119–136.

Nagata, T., Takimoto, M., & Kambayashi, Y. (2009). Suppressing the Total Costs of Executing Tasks Using Mobile Agents. HICSS’09.

---

Skills

Machine Learning: Vision-Language Models, Multimodal Representation Learning, Foundation Model Development, Contrastive & Self-supervised Learning, Reinforcement Learning · PyTorch, HuggingFace

Programming: Python (primary), Java, SQL

Infrastructure: AWS, Hadoop, Spark/PySpark, MySQL, Redis, DynamoDB

Languages: Japanese (native), English (fluent)