Robustness Verification for Transformers
Development of sound and scalable verification methods for Transformer models, with particular emphasis on attention operations, nonlinear bound construction, and robustness certification.
Development of sound and scalable verification methods for Transformer models, with particular emphasis on attention operations, nonlinear bound construction, and robustness certification.
Research on low-resource neural machine translation and case-based methods conducted during the master’s program at Waseda University.
Development and evaluation of document-understanding pipelines that combine OCR, large language models, structured extraction, and domain-specific reference data.
Development of pipelines that transform PDF, Markdown, and spreadsheet documents into structured training data.
A retrieval and question-answering workflow for technical manuals using document parsing, vector search, and language models.