Kenichiro Goto

Writing

These pieces are written by a human, not AI. Some are written in English as part of my own English learning.

  1. VLM notes 2: training objectivesThe losses behind VLMs: contrastive pretraining for the vision encoder and next-token prediction for the full model.
  2. VLM notes 1: from pixels to tokensHow a vision-language model turns an image into a sequence of tokens that a language model can read.
  3. English learning notesNotes collected while studying English and improving day-to-day practice.