David ‘s Machine Learning Hub

Hello! I am currently navigating through the second year of my graduate studies in the Master’s Programs in Informatics, nestled within the Graduate School of Comprehensive Human Sciences at the University of Tsukuba, Japan. My academic quests are primarily anchored in the realms of Natural Language Processing and Information Retrieval, under the nurturing guidance of Professor Haitao Yu, in the innovative environment of the Information Intelligence Lab.

When it comes to research, my curiosity veers towards Emotion Analysis, Hierarchical Classification, and the broad expanse of Machine Learning. Every discovery and understanding in these fields not only fuels my academic voyage but also finds a mention on this page.

This platform serves as a repository of my learning and explorations. Through this page, I am keen on documenting and disseminating the knowledge I accrue over time, fostering a space of learning and collaborative intellectual growth.

Please feel free to share your opinions and make progress together.


论文Publications論文

全部 →All →すべて →
Emotional Earth Mover's Distance for Fine-Grained Hierarchical Emotion Analysis International Conference on Advanced Data Mining and Applications (ADMA 2025) · 2025

最新技术情报Latest tech intel最新の技術インテル

全部 →All →すべて →
  1. Goldman Sachs 把 Claude Code 从交易员和银行家手里收回,HN 669👍 · 阿里 Co-Design Bench:GPT-6 Astra 在真实用户请求上只完整解决 26.4%,旧基准同模型 99%+ · Karpathy 用 Claude Opus 5.5 一个周末把 4300 万条 HN 评论筛成 6,454 个视频,称效果超过 YouTube 推荐
  2. Claude Code 的 "Don't Ask Again" 把密码写进 1,122 个公开 GitHub 文件 · Anthropic 顶着科技股抛售推进感恩节前 IPO,估值降到 1.25–1.75 万亿美元,募资 80–100 亿 · Vals AI 基准:开源与闭源模型的网络攻击能力差距在缩小,DeepSeek V4.1 Flash 以 1/10 成本达到 72.8% 攻击目标完成率(GPT-6 Sol Daybreak 80.8%)
  3. Ripple-65B Preview:Apache 2.0 全开放权重,TerminalBench 2.1 上美国开源模型第一,研发总成本 3200 万美元 · Chainalysis 承认全部客户的凭证和 token 被窃,攻击含 AI agent 与 on-path 数据截取 · Anthropic IPO 买方估值低至 1.1 万亿美元,同时 Ramp 数据显示 AI 支出正在停滞

博客文章Blog postsブログ記事

全部 →All →すべて →
Jev,一个不说话的 AI:关于“System One 模型”的笔记Jev, the AI that does not talk: notes on a "System One" model話さない AI、Jev:「System One モデル」についてのメモ 2026-09-20 Agent = 模型 + Harness:Pi、OpenClaw、DeepSeek Harness 和 Claude Code 横向对比Agent = model + harness: Pi, OpenClaw, DeepSeek Harness and Claude Code side by sideエージェント = モデル + ハーネス:Pi・OpenClaw・DeepSeek Harness・Claude Code の横並び比較 2026-09-12 Claw 家族:每一个分支都在修 OpenClaw 的一种失败方式The Claw family: every fork fixes one way OpenClaw failsClaw ファミリー:どのフォークも OpenClaw の失敗モードを一つずつ直している 2026-09-04