Encoder-based言語モデルによるテキスト生成：RoBERTa Diffusionの実装と検証
RoBERTaのMasked Language Modelingを用いたテキスト拡散モデルの実装と、per-batch・per-sampleマスク戦略の比較。
Blog
Written in Mandarin, English, or Japanese.
2024年の振り返り
2024年の出来事と、東京で社会人として過ごした一年の振り返り。
gzip is all you need (gzip + kNN outperforms Transformers)
Notes on parameter-free text classification with gzip compression and k-nearest neighbors.
Exploiting Cloze Questions for Few Shot Text Classification and Natural Language Inference
A reading note on Pattern-Exploiting Training (PET), a semi-supervised prompt-based method for few-shot text classification.
Breadth-First Search (using Python)
An introduction to breadth-first search, with a Python implementation, complexity analysis, and traversal example.