寅太郎@技術書典21【お16】でオフライン参加 retweeted
【12月新刊】『新装版 位相幾何学:ホモロジー論』(中岡 稔 著)
kyoritsu-pub.co.jp/book/b102…
◇ 共立出版100周年記念 復刊リクエスト書籍(第6弾)◇
現代数学の多様な分野に広がるホモロジー論の理論と展開を体系的に詳述した本格的解説書。(12/23発売)
寅太郎@技術書典21【お16】でオフライン参加 retweeted
𝗞𝗩 𝗖𝗮𝗰𝗵𝗶𝗻𝗴 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗲𝗱
Ever wondered how LLMs remember thousands of tokens without processing everything from scratch every time?
That's where KV Caching comes in.
When an LLM generates a response, it doesn't need to recompute the Key and Value representations of all previous tokens at every step.
Instead, it stores them in a cache and reuses them while generating the next token.
Sounds simple, but this makes a huge difference in inference speed and computational efficiency.
Here's what happens:
- During prefill, the model processes the prompt and stores the Keys and Values.
- During decoding, it computes Q, K and V only for the new token.
- The new K and V are added to the existing cache.
- Previous KV values are reused instead of being recomputed.
But there's one important thing to understand: KV Caching doesn't eliminate attention over previous tokens. It eliminates the need to recompute their Keys and Values.
As context grows, KV cache memory can become a major bottleneck.
That's why techniques like:
- MQA (Multi-Query Attention)
- GQA (Grouped-Query Attention)
- PagedAttention
- Prefix Caching
play an important role in making LLM inference more efficient.
Understanding KV Caching is essential if you're learning how LLM inference actually works or building systems that serve LLMs at scale.
寅太郎@技術書典21【お16】でオフライン参加 retweeted
The EmbeddingGemma 2 tech report is officially live!
We are sharing the full architecture, training recipe, and ablations behind our 740M open multimodal embedding model.
Inside the paper:
🌐 Unified representations for text, code, images, audio, and video
📈 How it outperforms specialist and general models up to 12x larger
⚡ On-device efficiency: modular encoders, Matryoshka dims, and int4 quantization (down to 567MB)
Read the full report: arxiv.org/abs/2610.37364
寅太郎@技術書典21【お16】でオフライン参加 retweeted
LSTMやGRUのようなゲート機構を持った連想記憶ニューラルネットワーク(甘利ホップフィールド模型)を、統計物理の手法(動的平均場理論)で解析しました。
pnas.org/doi/10.1073/pnas.26…
寅太郎@技術書典21【お16】でオフライン参加 retweeted
【#pixiv_inside】
PHP 8.5.12の緊急リリースに
pixivはどう対応したか
inside.pixiv.blog/2026/10/22…
PHPにて緊急セキュリティリリースが予告され
pixivも対応しました。
セキュリティアップデート公開から
pixivメンテナンス明けまで約30分の対応を
ビルド・デプロイ基盤も踏まえながら説明します✍️
寅太郎@技術書典21【お16】でオフライン参加 retweeted
openai just released Decisions API, Jev-like "system 1" multi-modal model based on GPT6 Luna!
Each egg gets cropped and sent to OpenAI (~270ms latency) and returns P(clean, dirty, cracked).
One image crop consumes ~280 input tokens (~$0.03/1K imgs). @roboflow RF-DETR for detecting eggs
寅太郎@技術書典21【お16】でオフライン参加 retweeted
MicrosoftがローカルAI「MAI-Code-1.1-Flash」をリリース&ローカルAIとクラウドAIを組み合わせた「ハイブリッドインテリジェンス」も提唱
寅太郎@技術書典21【お16】でオフライン参加 retweeted
You can now train your own Decision model like Jev locally!
We increased Qwen3.5 0.8B’s aggregate accuracy from 20.7% to 74.3% across 3 decision benchmarks - on just 4GB VRAM.
Turn any LLM like Qwen3.8, Gemma 4 into decision models with our open-source Unsloth repo.
We fine-tuned with a Clef head using Unsloth and LoRA (r=64) for one epoch, increasing downstream accuracy from 30–37% to 78%.
GitHub: github.com/unslothai/unsloth
Guide and Notebooks: unsloth.ai/docs/basics/train…
【東京理科大学 硤合 憲三名誉教授がノーベル化学賞を受賞】
スウェーデン王立科学アカデミーは7日、2026年のノーベル化学賞を硤合 憲三 本学名誉教授に授与すると発表しました。「不斉有機合成における非線形効果と自己触媒反応の発見」に係る研究が高く評価されたものです。
tus.ac.jp/today/archive/2026…
寅太郎@技術書典21【お16】でオフライン参加 retweeted
【速報】このたび2026年ノーベル物理学賞にFrancis Halzen 氏が、ニュートリノ観測の国際共同研究であるIceCube実験の功績により受賞されました!
千葉大学ハドロン宇宙国際研究センターの吉田滋センター長、石原安野教授が所属する千葉大学チームは、同プロジェクト唯一の日本からの参加機関です。
BREAKING NEWS
The Royal Swedish Academy of Sciences has decided to award the 2026 #NobelPrize in Physics to Francis Halzen “for decisive contributions to the IceCube Neutrino Observatory and the discovery of high-energy neutrinos of astrophysical origin.”
寅太郎@技術書典21【お16】でオフライン参加 retweeted
Meet JEV-9B, a 9B text classifier built on Qwen3.5. It blends system one and system two thinking with blocks of experts. 305K downloads and counting. This is decision making AI done right.
寅太郎@技術書典21【お16】でオフライン参加 retweeted
2026年8月で「明治 エッセル スーパーカップ 抹茶 200ml」の製造は終了しました。
店頭在庫限りとなります。
sweetroad5.com/news/30713/
寅太郎@技術書典21【お16】でオフライン参加 retweeted
Google releases EmbeddingGemma 2, a new open model that runs locally on 0.5GB RAM.
The 740M parameter Apache 2.0 model combines a 270M text model with vision (170M) + audio (300M).
Run & train the model via Unsloth.
GGUF: huggingface.co/unsloth/embed…
Guide: unsloth.ai/docs/models/embed…
寅太郎@技術書典21【お16】でオフライン参加 retweeted
『サルゲッチュ』が約16年ぶりに復活か
ソニーが休眠シリーズ復活のプロジェクトを進行中とリーカーが言及
開発はTeam ASOBIではなく別スタジオが担当する模様
gazlog.jp/entry/saru-gechu-m…
寅太郎@技術書典21【お16】でオフライン参加 retweeted
無料でAWSをローカルでシミュレーションできるエミュレーター「MiniStack」、60以上のAWSサービスを単一ポートで利用可能でマルチアカウント&マルチリージョン対応、セルフホスト可能
寅太郎@技術書典21【お16】でオフライン参加 retweeted
👍Some(body+AI) made an artistic cover page for my article "An Elementary Introduction to Information Geometry"!
PDF: franknielsen.github.io/entro…
Thanks @burkov ! Check below!
teapawt.org/burkov/status/20977738…
A short textbook "An elementary introduction to information geometry" by Frank Nielsen (2020) is now in the @ChapterPal's collection.
The book provides a foundational entry into the differential geometric study of probability, statistics, and machine learning.
Designed for graduate students, researchers, and practitioners in information sciences, the book assumes a basic working knowledge of linear algebra, multivariable calculus, and classical probability and statistics.
While the text introduces the necessary differential geometry from first principles—covering tangent spaces, metric tensor fields, and affine connections—prior familiarity with basic differential notation helps reader comprehension.
The scope focuses primarily on finite-dimensional parametric families of probability distributions, presenting geometric frameworks to analyze decision-making, estimation, and optimization.
All books on ChapterPal are free to read with a free account,
Read the book with an AI tutor: chapterpal.com/ebook/adec358…
The table of contens:
1. Introduction
- 1.1. Overview of information geometry
- 1.2. Outline of the survey
- References
2. Prerequisite: Basics of differential geometry
- 2.1. Overview of differential geometry: Manifold $(M, g, \nabla)$
- 2.2. Metric tensor fields $g$
- 2.3. Affine connections $\nabla$
* 2.3.1 Covariant derivatives $\nabla_X Y$ of vector fields
* 2.3.2 Parallel transport $\prod^\nabla_c$ along a smooth curve $c$
* 2.3.3 $\nabla$-geodesics $\gamma_\nabla$: Autoparallel curves
* 2.3.4 Curvature and torsion of a manifold
- 2.4. The fundamental theorem of Riemannian geometry: The Levi-Civita metric connection
- 2.5. Preview: Information geometry versus Riemannian geometry
- References
3. Information manifolds
- 3.1. Overview
- 3.2. Conjugate connection manifolds: $(M, g, \nabla, \nabla^*)$
- 3.3. Statistical manifolds: $(M, g, C)$
- 3.4. A family $\{(M, g, \nabla^{-\alpha}, \nabla^{\alpha}=(\nabla^{-\alpha})^*)\}_{\alpha\in \mathbb{R}}$ of conjugate connection manifolds
- 3.5. The fundamental theorem of information geometry: $\nabla$ $\kappa$-curved $\Leftrightarrow$ $\nabla^*$ $\kappa$-curved
- 3.6. Conjugate connections from divergences: $(M, D)\equiv (M, {^{D}}g, {^{D}}\nabla, {^{D}}\nabla^*={^{D^*}}\nabla)$
- 3.7. Dually flat manifolds (Bregman geometry): $(M, F)\equiv (M, {^{B_F}}g, {^{B_F}}\nabla, {^{B_F}}\nabla^*={^{B_{F^*}}}\nabla)$
- 3.8. Hessian $\alpha$-geometry: $(M, F, \alpha)\equiv (M, {^{F}}g, {^{F}}\nabla^{-\alpha}, {^{F}}\nabla^\alpha$
- 3.9. Expected $\alpha$-manifolds of a family of parametric probability distributions: $(\mathcal{P}, {_{\mathcal{P}}}g, {_{\mathcal{P}}}\nabla^{-\alpha}, {_{\mathcal{P}}}\nabla^{\alpha})$
- 3.10. Criteria for statistical invariance
- 3.11. Fisher-Rao expected Riemannian manifolds: $(\mathcal{P}, {_{\mathcal{P}}}g)$
- 3.12. The monotone $\alpha$-embeddings and the metric gauge freedom
- 3.13. Dually flat spaces and canonical Bregman divergences
- References
4. Some applications of information geometry
- 4.1. Natural gradient in Riemannian space
* 4.1.1 The vanilla gradient descent method
* 4.1.2 Natural gradient and its connection with the Riemannian gradient
* 4.1.3 Natural gradient in dually flat spaces: Connections to Bregman mirror descent and ordinary gradient
* 4.1.4 An application of the natural gradient: Natural Evolution Strategies (NESs)
- 4.2. Some illustrating applications of dually flat manifolds
- 4.3. Hypothesis testing in the dually flat exponential family manifold $(\mathcal{E}, \mathrm{KL}^*)$
- 4.4. Clustering mixtures in the dually flat mixture family manifold $(\mathcal{M}, \mathrm{KL})$
- References
5. Conclusion: Summary, historical background, and perspectives
- 5.1. Summary
- 5.2. A brief historical review of information geometry
- 5.3. Perspectives
- References
A. Monte Carlo estimations of $F$-divergences
- References
B. The multivariate Gaussian family: An exponential family
- References
Notations
Lyria 3.5, our best-sounding music generation model, is now available in the @GeminiApp 🎵✨
With more expressive vocals and richer musical arrangements, it’s easier than ever to bring your idea to life:
🪄 Use our new templates to jumpstart your creativity
⌛ Choose to create short or longer tracks
✅ Select or describe your genre and choose between vocal or instrumental styles


































