耳の中の音から目の動きを推定する研究、凄いなこれ。
AirPodsをつけてるだけでどこを見ているかわかる未来が近い。胸につけるカメラ付きウェアラブルデバイス（HumaneのAi Pinのような）との相性も良さそう。
自分がどこを見ているかがAIに伝わる https://twitter.com/AiBreakfast/status/1733987436974518488…
Retweeted by Shigekazu Ishihara
retweeted at 13:40:32
| Stats | Twitter歴 5,335日（2009/05/04より） |
ツイート数 180,453（33.8件/日） |
表示するツイート :
2023年12月11日(月)1 tweetsource
2023年12月10日(日)63 tweetssource
Image to Text ROS2 https://pic.twitter.com/mAagCDfG0M
Retweeted by Shigekazu Ishihara
retweeted at 16:16:58
@Rainmaker1973 Rafael Araujo on mathematical illustrations of nature
The artist found inspiration from the Golden Ratio.
Rafael Araujo, is an architect and illustrator born in Caracas, Venezuela in 1957
Inspired by mathematics, Araujo uses geometry, trigonometry, and dot sequences to create his… https://twitter.com/i/web/status/1733476996368454054… https://pic.twitter.com/uTMGJZsfw3
Retweeted by Shigekazu Ishihara
retweeted at 16:08:43
Rafael Araujo draws structures which could perfectly be described as illustrated equations: he does it by hand, with no help of computing and by using classical architectural drawing tools to go deeper in the visualization of tridimensional geometry
https://rafael-araujo.com https://pic.twitter.com/VIA9KmeCMf
Retweeted by Shigekazu Ishihara
retweeted at 16:08:22
The Strange Order of Things by Antonio Damasio review – why feelings are the unstoppable force
https://www.theguardian.com/books/2018/feb/02/strange-order-of-things-antonio-damasio-review…
Retweeted by Shigekazu Ishihara
retweeted at 16:06:09
The Matern kernel interpolates between Laplacian and Gaussian. The most popular kernels in ML, allows to represent functions with a given smoothness nu (number of derivative) for regression and classification using RKHS. https://en.wikipedia.org/wiki/Mat%C3%A9rn_covariance_function… https://pic.twitter.com/DaNQkoFQxI
Retweeted by Shigekazu Ishihara
retweeted at 16:03:12
visionOS, LLM: On Device Chat with mlc-llm🌛
- model: rinna-3.6b (Ja)
- quantization: q4f16
mlc-llm generates an iOS project. Since it cannot run in the simulator, I used My Mac(Designed for iPad) to emulate iPad. 2.4GB memory was consumed.
I think this can run in visionOS. :) https://pic.twitter.com/pk7USqkGw8
Retweeted by Shigekazu Ishihara
retweeted at 14:35:19
https://twitter.com/leafs_s/status/1469094905620631552…
Retweeted by Shigekazu Ishihara
retweeted at 14:28:55
Nature Reviews Neuroscience
Attractor and integrator networks in the brain
https://www.nature.com/articles/s41583-022-00642-0…
Retweeted by Shigekazu Ishihara
retweeted at 14:28:36
https://x.com/miyamotty/status/1733498916111548899?s=20…
Retweeted by Shigekazu Ishihara
retweeted at 10:17:39
https://x.com/z_zabaglione/status/1733468181522731257?s=20…
Retweeted by Shigekazu Ishihara
retweeted at 10:17:33
@KBingxin https://x.com/kiyoshi_shin/status/1733310925904736636?s=20…
Retweeted by Shigekazu Ishihara
retweeted at 10:17:26
@KBingxin https://x.com/KenjiASABA/status/1733031212757983733?s=20…
Retweeted by Shigekazu Ishihara
retweeted at 10:17:19
Introducing Marigold 🌼 - a universal monocular depth estimator, delivering incredibly sharp predictions in the wild! Based on Stable Diffusion, it is trained with synthetic depth data only and excels in zero-shot adaptation to real-world imagery. Check it out:
🌐 Website:… https://twitter.com/i/web/status/1732946419663667464… https://pic.twitter.com/UeEHp3fMKQ
Retweeted by Shigekazu Ishihara
retweeted at 10:14:05
Since Marigold depth launch yesterday, we've seen amazing uses: 3D printing, holographic images, UE5 assets, ControlNet, and optical illusion disentanglement. Share your creations with us! Use our free tool and follow @KBingxin for updates! https://huggingface.co/spaces/toshas/marigold… https://twitter.com/i/web/status/1733603655733985368…
Retweeted by Shigekazu Ishihara
retweeted at 10:13:47
書籍では2グループのみの潜在変数モデルですが、
赤石先生は、僕たち学習者のために、「3グループ版」もWeb上で公開してくださっています😆👍
- ディリクレ分布
- カテゴリカル分布
どちらも、色んな場面でお目に掛かる重要な要素ですから、必見と思います😆👍 https://twitter.com/makaishi2/status/1733640456968560899…
Retweeted by Shigekazu Ishihara
retweeted at 10:03:11
低周波(0.1Hz未満)のresting state globalBOLDは潜在的にglymphatic機能に関連しDMNでのAβ/tau蓄積に関連
MRIにてgBOLDとCSF流入域から抽出したCSF信号は相関
DMNにおける上記相関係数はAβおよびtauの2年後蓄積量と負の相関
Open Access
#refMRI
#glymphatics
#papers
https://doi.org/10.1038/s41467-023-43627-y… https://pic.twitter.com/Qos1alnY3g
Retweeted by Shigekazu Ishihara
retweeted at 10:02:23
深度推定のMarigoldでAdobe Firefly生成のドラゴン推定させた例。感心するのがMiDaSやZoeで破綻しやすかった、ボケた無限遠っぽい背景をかなり意図に近く推定してくれるところ。透過や反射も。
アニメやイラストにはやや弱そうだけど応用は幅広い。 https://twitter.com/antonobukhov1/status/1733647102595486140…
Retweeted by Shigekazu Ishihara
retweeted at 10:00:10
Outputしつつ、勉強もすべし…。 https://pic.twitter.com/JmkkAjDCws
Retweeted by Shigekazu Ishihara
retweeted at 09:53:12
深津 貴之 / THE GUILD / note@fladdict
3DCGファッションブランド作って欲しい。シェイプを固定して、テクスチャーで柄もの作るようにしつつ、生成AIで様々なテクスチャバリエーション作ったらすごい面白い気がするよ。 https://twitter.com/nemchan_nel/status/1733404575015866863…
Retweeted by Shigekazu Ishihara
retweeted at 09:51:49
Neuroscience News@NeuroscienceNew
“Benefits of Wasabi Supplements with 6-MSITC (6-Methylsulfinyl Hexyl Isothiocyanate) on Memory Functioning in Healthy Adults Aged 60 Years and Older: Evidence from a Double-Blinded Randomized Controlled Trial” by Rui Nouchi et al. Nutrients
https://www.mdpi.com/2072-6643/15/21/4608…
2/2
Retweeted by Shigekazu Ishihara
retweeted at 09:50:25
Neuroscience News@NeuroscienceNew
Wasabi Compound Enhances Memory
A new study shows that wasabi's compound, 6-MSITC, boosts memory in older adults. Could this spicy addition be a key to better cognitive health in our golden years?
https://neurosciencenews.com/wasabi-aging-memory-25342/…
1/2 https://pic.twitter.com/uaPEIwfxoX
Retweeted by Shigekazu Ishihara
retweeted at 09:50:21
共著論文が早期公開されました
https://www.jstage.jst.go.jp/article/jjppp/advpub/0/advpub_2314si/_article/-char/ja/… https://twitter.com/d_iimura274/status/1727204316170408059…
Retweeted by Shigekazu Ishihara
retweeted at 09:39:17
🤩We are excited to share with you perhaps the largest open-access #fNIRS data set! This #multimodal research includes data collected with the #NIRSport2 as well as #EEG, #EKG, #GSR, eye tracking, and more! Applications for #HCI, #ToM, and more 👉 https://openreview.net/pdf?id=ZJWQfgXQb6… https://pic.twitter.com/VuT4WDKKp9
Retweeted by Shigekazu Ishihara
retweeted at 09:38:59
Amazonに売ってるモノだけで6脚ロボットを完全自作してみた (3Dモデル＋ソースコード付き) - ABEJA Tech Blog https://tech-blog.abeja.asia/entry/advent-2023-day08…
Retweeted by Shigekazu Ishihara
retweeted at 09:38:12
アドベントカレンダー9日目,1300sさんの「自転車のデータ測定」!
https://cycle1300ss.blogspot.com/2023/12/bluetooth2023.html…
市販のサイコンでも同じ事は出来るがSimulinkでトライされる心意気が良いですね。Bluetooth連携は接続までは割と出来るがその後がやや面倒というのは私も経験ありますね…。
Retweeted by Shigekazu Ishihara
retweeted at 09:37:41
アドベントカレンダー9日目その2,kentaPtさんの「3次元点群ファイル(E57形式)の解説」!
https://kentapt.hatenablog.com/entry/2023/12/03/223445…
3次元点群データってものすごいデータの塊という印象があったけど想像以上だった。スキャンしたときのカメラの姿勢を可視化するMATLAB関数plotCameraが便利すぎる…。
Retweeted by Shigekazu Ishihara
retweeted at 09:37:31
アドベントカレンダー8日目, @carter_control さんの「ダイヤル調整で理解する2 次遅れ系パラメータ」(超意訳)!
https://qiita.com/Carter/items/d8b3961e367b76d75626…
制御工学の大事な要素である2次遅れ系パラメータの意味をインタラクティブに学習する環境を構築。各ブロックの細かい設定まで記載された親切な記事です。
Retweeted by Shigekazu Ishihara
retweeted at 09:37:18
そぞら@Raspberry Pi 電子工作@sozoraemon
フライトレーダーの画面とレーダーの動画素材を合成したら、めちゃくちゃそれっぽくなった！
簡単なプログラムを組んで、小型PCのRaspberry Piで動かしています。 https://pic.twitter.com/8MDEDKZTNk
Retweeted by Shigekazu Ishihara
retweeted at 09:33:54
『Pythonでスラスラわかるベイズ推論「超」入門』著者@makaishi2
@kenken26679105 @kenken26679105 さん、
『Pythonでスラスラわかる ベイズ推論「超」入門 』著者の赤石です。
昨日の階層ベイズに引き続き、潜在変数モデルについても執筆時のこだわりポイントを丁寧に1つ1つ取り上げていただき、うれしい限りです。… https://twitter.com/i/web/status/1733640456968560899…
Retweeted by Shigekazu Ishihara
retweeted at 09:30:53
NEW SOTA - #CausalLM 72B-preview rank #1 among chat / instruct finetune models ~70B on Open LLM Leaderboard! https://huggingface.co/CausalLM/72B-preview…
Uncensored, 100% LLaMA Compatible, work with GGUF https://pic.twitter.com/usi6YU5ntd
Retweeted by Shigekazu Ishihara
retweeted at 09:17:07
室井まさね『少女の檻』1〜3巻発売中‼︎@Dimetrodon_
@NazologyInfo テトリスの適度な難易度が「フロー状態」を生みやすいらしい🤔
テトリスに抗不安薬効果。プレイすることで不安を緩和させる効果があることが判明 https://karapaia.com/archives/52267222.html… @karapaiaより
Retweeted by Shigekazu Ishihara
retweeted at 07:51:04
「テトリス」にはトラウマを癒す効果がある!?
https://nazology.net/archives/140002
実はテトリスを使った心理研究は多く、その中には精神医学上非常に有用な効果を報告したものがあります。スウェーデンKIはテトリスの視覚的負荷がトラウマ体験の記憶形成を阻害しフラッシュバックを低下させると報告しています https://pic.twitter.com/DsmPwJ5o3x
Retweeted by Shigekazu Ishihara
retweeted at 07:50:55
投稿しました。
（解説動画は）初投稿です！
【#ニコニコダム祭り】久山田貯水池堰堤と上島諸島の水事情のお話
https://www.nicovideo.jp/watch/sm43121957
#ニコニコ動画 #sm43121957
Retweeted by Shigekazu Ishihara
retweeted at 07:49:48
"La société face aux défis de l'intelligence artificielle"
Un débat public organisé par la revue européenne Le Grand Continent hébergé à l'ENS.
Participants: Anne Bouverot, Marc Mézard, Martin Tisné, Jamal Atif, et moi-même.
Ce débat s'inscrit dans la continuité de l'entretien… https://twitter.com/i/web/status/1733615209485353140…
Retweeted by Shigekazu Ishihara
retweeted at 07:43:04
しばぱぱ三上昌史Gugenka CEO@Mikami_Gugenka
Luma AIのデータをLooking Glassに映すやり方をnoteに残しました～
https://note.com/gugenkamikami/n/nc349a4764cef… https://pic.twitter.com/qBsV6cXpgj
Retweeted by Shigekazu Ishihara
retweeted at 07:37:45
［改訂第9版］LaTeX美文書作成入門 https://gihyo.jp/book/2023/978-4-297-13889-9… いよいよ本日発売。電子版はGihyo Digital PublishingのPDFがお薦めです
Retweeted by Shigekazu Ishihara
retweeted at 07:37:12
Happy to OSS gpt-fast, a fast and hackable implementation of transformer inference in <1000 lines of native PyTorch with support for quantization, speculative decoding, TP, Nvidia/AMD support, and more!
Code: https://github.com/pytorch-labs/gpt-fast…
Blog: https://pytorch.org/blog/accelerating-generative-ai-2/…
(1/12) https://pic.twitter.com/BTa10O3KGJ
Retweeted by Shigekazu Ishihara
retweeted at 06:34:14
The EU AI Act negotiations ended.
One contentious issue was the regulation of foundation models, particularly open source ones.
Kudos to the French, German, and Italian governments for not giving up on open source models.
Juicy part:
"The legislation ultimately included… https://twitter.com/i/web/status/1733481002234679685…
Retweeted by Shigekazu Ishihara
retweeted at 06:28:55
MistralのMoEモデル、VRAM24GBで動くの？マジ？エキスパート全員ロードしておく必要はないって事？ →RT
Retweeted by Shigekazu Ishihara
retweeted at 06:28:45
Moore’s Law for Everything@Evangelize_K
7Bの32 Expertが、僅か24GBで動作！？
Multi-Modalに続いて、Multi-Agentの時代も一気に到来してしまうのでしょうか？
「我々はMistral MoE (7Bx32experts) を 2 か月間使用しており、それは24GBで動作しています。」 https://twitter.com/far__el/status/1733297696536969478…
Retweeted by Shigekazu Ishihara
retweeted at 06:28:43
新清士@(生成AI)インディゲーム開発者@kiyoshi_shin
できた。 #LookingGlassPortrait 2枚表示。予想以上に違和感ない。ちゃんと2枚合わさって立体に見える。こんなに迫力が出るとは。やり方は単純。2台のPCで別々に繋いで表示して合わせているだけ。 #LookingGlassGo のマルチ対応に期待。 https://pic.twitter.com/1lQEaOlol2
Retweeted by Shigekazu Ishihara
retweeted at 06:23:25
stable Diffusion WebUI(1111版)をpythonから1クリックでインストールし起動までできるインストーラのプロトタイプがほぼ完成しました。
同類のStabilityMatrixもWebui内設定やコマンドライン、モデル、VAE、LoRAのダウンロードまで一括ではないのでこれらもすべて自動化をしています。
Retweeted by Shigekazu Ishihara
retweeted at 06:03:20
MistralAIが急にtorrentのマグネットリンクだけ貼って何じゃこりゃという感じだけど、これはオープンソースのMoEのLLMモデルで、中身は7Bのエキスパートが8人いて、合計42Bパラメータ。（8*7=56Bじゃねえの？）最近はモデルをtorrentで公開すんの？GPT-4の構造をコンパクトにした感じの作りだという。… https://twitter.com/i/web/status/1733518394518032412…
Retweeted by Shigekazu Ishihara
retweeted at 06:03:11
magnet:?xt=urn:btih:5546272da9065eddeb6fcd7ffddeef5b75be79a7&dn=mixtral-8x7b-32kseqlen&tr=udp%3A%2F%http://2Fopentracker.i2p.rocks%3A6969%2Fannounce&tr=http%3A%2F%http://2Ftracker.openbittorrent.com%3A80%2Fannounce
RELEASE a6bbd9affe0c2725c1b7410d66833e24
Retweeted by Shigekazu Ishihara
retweeted at 06:03:08
Mistral AI releases first ever OpenSource MoE model
Fun facts:
- Released as 87 GB torrent
- Seems to be a scaled down version of GPT-4
- Released on X, no press release and refuse to elaborate
Google left the AI community in awe with the over rehearsed demo video that's now… https://twitter.com/i/web/status/1733364854973456561… https://pic.twitter.com/Pit1e7nrpr
Retweeted by Shigekazu Ishihara
retweeted at 06:03:02
cocktail peanut@cocktailpeanut
Another round of experiment with StyleAligned.
Q: "Is the image ACTUALLY working as the style source? Or is it the style description (for example "wes anderson style") that's doing the work?"
I took this sim city image, and instead of "sim city style" I just used "game style". https://twitter.com/i/web/status/1733238908375171211… https://pic.twitter.com/I98P9VBy3g
Retweeted by Shigekazu Ishihara
retweeted at 05:23:34
cocktail peanut@cocktailpeanut
@kusichan When you play with it, make sure to tweak the shared_score_shift, shared_score_scale, and guidance_scale to get what you want.
I find that these knobs and handles matter A LOT.
For example, try tweaking
- shared_score_shift between 1 and 2
- shared_score_scale between 0 and 1
Retweeted by Shigekazu Ishihara
retweeted at 05:20:26
cocktail peanut@cocktailpeanut
Here's the original announcement from the direct source, the man himself, @kusichan:
https://x.com/kusichan/status/1731908268740546985?s=20…
Retweeted by Shigekazu Ishihara
retweeted at 05:20:20
cocktail peanut@cocktailpeanut
Also, as always, if you're having trouble running stuff, please post on Discord, I miss a lot of the comments on this app and also it's a terrible place to help with something. https://discord.gg/TQdNwadtE4
Retweeted by Shigekazu Ishihara
retweeted at 05:20:15
cocktail peanut@cocktailpeanut
I posted on Reddit as well, if you're on Reddit, please upvote for visibility :)
https://www.reddit.com/r/StableDiffusion/comments/18dp687/ive_built_a_web_ui_for_googles_stylealigned/…
Retweeted by Shigekazu Ishihara
retweeted at 05:20:12
cocktail peanut@cocktailpeanut
Of course, this does not mean the end of LoRAs or Dreambooth. In fact, it already integrates with Dreambooth (and probably LoRA soon if it isn't already)
That said, it will DEFINITELY bring a huge change to the workflow of how people have been thinking about AI models. https://pic.twitter.com/dB3MbQBam0
Retweeted by Shigekazu Ishihara
retweeted at 05:19:58
cocktail peanut@cocktailpeanut
I can't emphasize enough how game changing this is.
Until now we've assumed we need to train custom models if we want to do any custom stuff.
But this obliterates that assumption. One image is all you need. https://twitter.com/i/web/status/1733145085003329693…
Retweeted by Shigekazu Ishihara
retweeted at 05:19:49
cocktail peanut@cocktailpeanut
This is what the web UI looks like. Here's how it works:
1. Provide a "reference image" you want to mimic.
2. Describe the style of the ref image in words.
3. Also describe what the ref image is about.
4. Then enter the prompts you want to generate, separated by new line. https://pic.twitter.com/JBXP0pc8uV
Retweeted by Shigekazu Ishihara
retweeted at 05:19:42
cocktail peanut@cocktailpeanut
When Google released StyleAligned a couple of days ago, the project didn't include a IMG-to-IMG code, which was pretty much the whole point of the project.
So I bugged them about it and they have released the code for img2img https://github.com/google/style-aligned/issues/6…
Retweeted by Shigekazu Ishihara
retweeted at 05:19:34
cocktail peanut@cocktailpeanut
And the obligatory 1 click installer.
It works on ALL platforms (Mac, Windows, Linux)
https://pinokio.computer/item?uri=https://github.com/cocktailpeanut/StyleAligned.pinokio…
Retweeted by Shigekazu Ishihara
retweeted at 05:19:29
cocktail peanut@cocktailpeanut
Here's the GitHub repository: https://github.com/cocktailpeanut/StyleAligned.pinokio…
Retweeted by Shigekazu Ishihara
retweeted at 05:19:25
cocktail peanut@cocktailpeanut
StyleAligned Img-to-Img Web UI
Google just released the code for IMG2IMG for StyleAligned
And I made a Gradio UI so you can play with it.
Generate imgs mimicing the style of a SINGLE img.
No LoRA/Dreambooth training needed.
One image is all you need. This changes everything. https://twitter.com/i/web/status/1733139983819554858… https://pic.twitter.com/zShFezH7EC
Retweeted by Shigekazu Ishihara
retweeted at 05:19:21
DiscoLM Mixtral 8x7b alpha is an experimental 8x7b MoE model based on Mistral AI´s Mixtral 8x7b. This model is based on experimental code converting the model weights to huggingface format and enabling Transformers-based inference. It was then finetuned on the Synthia,… https://twitter.com/i/web/status/1733580160576688279… https://pic.twitter.com/DCqwzORWMM
Retweeted by Shigekazu Ishihara
retweeted at 05:17:11
Brendan Bycroft@BrendanBycroft
Here's a technical guide on how I wrote & structured some of the LLM Visualization. A few people have been asking about it, so I thought I'd write it up. Lots of code screenshots, so not for everyone.
https://twitter.com/BrendanBycroft/status/1731042957149827140…
Retweeted by Shigekazu Ishihara
retweeted at 05:14:21
『Pythonでスラスラわかるベイズ推論「超」入門』著者@makaishi2
もう一つは、「人気ギフトランキング」1位になったこと。
私自身、どういうシチュエーションでこの本が贈り物にされるのか、よくわかっていませんが、自分の経験でも初めてのことかもしれず、ありがたい限りです。 https://pic.twitter.com/kDTU4J9zkm
Retweeted by Shigekazu Ishihara
retweeted at 04:40:54
『Pythonでスラスラわかるベイズ推論「超」入門』著者@makaishi2
『Pythonでスラスラわかる ベイズ推論「超」入門』著者です。本書に関わる good news を2つお伝えします。
1つは、長らくお待たせしていた Kindle版がようやく入手可能になったことです。
X… https://twitter.com/i/web/status/1733418199935590603…
Retweeted by Shigekazu Ishihara
retweeted at 04:40:50
GoogleはPixel8Pro上でどんなテクノロジー使ってGemini Nano動かしてるんやろ？と思ったら、普通にLlama. cppのggmlで動かしてるっぽい。なーんじゃ。それメタのLlamaから派生した技術なんだけど、オープンソースにフリーライドしてないでGeminiもオープンにしてどうぞ →RT
Retweeted by Shigekazu Ishihara
retweeted at 04:35:47
ggml will soon run on billion devices
@apple don't sleep on it 🙃 https://twitter.com/rgerganov/status/1733508095660237271…
Retweeted by Shigekazu Ishihara
retweeted at 04:27:58