Dust: Pretraining Transformers Without Backpropagation
Technical research and proposals regarding a new Transformer pre-training method called 'Dust' that does not require backpropagation.
Related articles(Research & Papers)
See moregihyo.jp
Sakana AI、「The AI Scientist」の第3世代を公開 ——複数エージェントで仮説検証から論文執筆まで、図生成ツール「AutoIllustrate」もオープンソース化
Sakana AI announced the release of the third generation of The AI Scientist, named 'The Agentic Scientist', on October 15, 2026.
ITmedia AI+
Microsoft、社内エンジニア数万人を対象に「Claude Code」「Copilot CLI」導入の影響を分析 生産性どう変わる
Microsoft researchers published a paper analyzing the usage and productivity impact of command-line AI coding tools among tens of thousands of internal engineers.
Hacker News
Turn your regular JavaScript code into a neural net
Explains a technique to convert regular JavaScript code into a neural network without training, making it fully differentiable and runnable in a browser without a GPU.
Cloud Watch
リコー、高い日本語リーズニング性能と高速・省リソースな運用を両立するMoE型マルチモーダルLLMを開発
Ricoh developed the multimodal LMM 'Ricoh LMM_35B-A3B' based on Alibaba Cloud's Qwen3.6, enhancing Japanese reasoning performance via reinforcement learning.
Hacker News
Neural Networks Are Already Clean
A working hypothesis regarding the ecology of neurons, generalization, and circuits is presented, examining neural network characteristics from an ecological perspective.