Yeesh, okay it is out and I can stop telling you I am going to write it.

https://screwlisp.small-web.org/complex/lisp-feedforward-deep-learning-example/

#deepLearning #symbolic #ffnn pure ansi #commonLisp

The example deliberately causes a hallucination programmed into the training data via simple #roc reasoning. To my knowledge the ROC interpretation of ffnn inference is original to me, here.

The hallucination is in both lockstep and single #neuralnet activations.

Be the first to complain my #DL works on subregions of jagged lists! #programming

Avi Chawla (@_avichawla)

한 연구자가 제안한 뉴럴넷 최적화 트릭으로 CPU에서 GPU로의 데이터 전송 속도가 약 4배 향상됩니다. 이미지 분류 작업에서 네트워크 정의와 데이터 전송 과정의 병목을 줄이는 방법으로, 모델 학습의 효율성을 대폭 개선할 수 있는 기술적 접근법입니다.

https://x.com/_avichawla/status/2024733274392911885

#neuralnet #optimization #gpu #training #performance

Avi Chawla (@_avichawla) on X

Here's a neural net optimization trick that leads to ~4x faster CPU to GPU transfers. Imagine an image classification task. - We define the network, load the data and transform it. - In the training loop, we transfer the data to the GPU and train. Here's the problem with this:

X (formerly Twitter)

Build neural networks in Rust using matmul, gradients, and backpropagation.

#rust #neuralnet #backprop

Getting 6DOF with older 3DOF XR glasses

Video: How to get 6DOF with older 3DOF XR glasses using Breezy and OpenTrack

Breezy can now turn a 3DOF (degree of freedom) device into a 6DOF device by augmenting the missing positional data from a webcam. Spoiler! It is not the cam strapped to my face – this is just for the demo you can watch here, on PeerTube or YouTube.

The cam, that I used for this task, is sitting on my monitor. How this works? Well not with magic! This requires a somewhat decent webcam – really anything from the last decade should suffice – and OpenTrack, of course.

OpenTrack is a head-tracking application with multiple tracker plugins. One of it’s plugins is the Neuralnet Tracker, an AI powered extension that comes with a bunch of different head pose models to choose from. With a webcam connected this can now locally run the detection model with very low latency – so it’s usually blazing fast on most systems!

This alone is already 6DOF and is used a lot for gaming already – so what does Breezy do with this? Simple! It reads the forwarded data via an UDP listener, a very quick way to transmit data on a local network or system [and complements it’s own rotational data with the missing positional data].

With this a Breezy user still gets the rotational data from the XR’s very sensitive IMU, that is short for Inertial Measurement Unit btw, and the not so important positional data sent from OpenTrack.

This works of course only while the webcam can still see the user. So sadly no walking around while using this.

And the best thing? It can also send the data back! This means that the very same combined values can be forwarded – e.g. to a computer game – benefiting from the best available data sources for rotation and position.

That’s not the main use case, of course, and only of importance for some nerds like myself. This is mostly relevant for the productivity features of Breezy, because sometimes a text may be too small to read with the glasses on. We do no longer have to increase the font size – we can now simply lean in! That is a feature that is usually only available with glasses, that come with little cameras of their own, so they can have native 6DOF support. And when I say native I mean that such glasses usually also outsource exactly this calculation to the connected computer. It’s my understanding that this seems to require a lot of computation power, which is something many XR users with the more modern devices complain about.

Well not so much with OpenTrack and the Neuralnet tracker, that utilizes the ONNX runtime under the hood. That’s a high-performance, cross-platform engine to power exactly such models locally. The runtime automatically makes use of the best available hardware acceleration, if there is any.

Overall I’m rather hyped about this feature – especially because I’m using the OpenTrack output option of Breezy for quite some time now, to get a VR like experience with stereoscopic 3D rendering in Side-By-Side mode. I can now keep using my older XR glasses and still enjoy this more modern 6DOF feature. This is rather expensive hardware after all.

And all that on Linux PC!

Breezy xr_driver: https://github.com/wheaney/breezy-desktop by https://www.youtube.com/@WayneHeaney

Official Announcement XR desktop with 6DoF + multiple displays: https://www.youtube.com/watch?v=eFLmjpjF-rA

Music “Life’s Worth Dying For” CC BY-SA 3.0 “LostDrone”. Licensed to the public under https://creativecommons.org/licenses/by-sa/3.0/ Verify at https://soundcloud.com/lostdrone/rock-lostdrone-lifes-worth-dying-for-free-download-and-creative-commons-license

This content is licensed under a Creative Commons Attribution 4.0 International license.

https://beko.famkos.net/2026/02/06/geting-6dof-with-older-3dof-xr-glasses/

#3DOF #6DoF #AR #Breezy #gaming #Neuralnet #opentrack #Viture #ViturePro #VR #XR

NeuralNet – trợ lý giao tiếp thông minh: chat tự nhiên, dịch đa ngôn ngữ, tìm kiếm internet tự động, chạy offline trên PC hoặc di động (API Engrok). Hoạt động local trong LM Studio, bảo mật cao, liên tục học từ người dùng. Khởi động ngay để trải nghiệm AI đa năng! #AI #NeuralNet #TrợLýAI #DịchNgônNgữ #CôngNghệ #Tech #Vietnam

https://i.redd.it/rgjsh4820leg1.png

NeuralNet – trợ lý giao tiếp thông minh: chat tự nhiên, dịch đa ngôn ngữ, tìm kiếm internet tự động, chạy offline trên PC hoặc di động (API Engrok). Hoạt động local trong LM Studio, bảo mật cao, liên tục học từ người dùng. Khởi động ngay để trải nghiệm AI đa năng! #AI #NeuralNet #TrợLýAI #DịchNgônNgữ #CôngNghệ #Tech #Vietnam

https://i.redd.it/43lapgdkykeg1.png

Un progetto che sviluppai negli anni '90, quando nessuno parlava di IA.

Nella simulazione una Rete Neurale Artificiale apprende le regole della tabella di payoff del Dilemma del Prigioniero Iterato (IPD) battendo tutti gli altri agenti artificiali, compreso il famoso "Tit-for-Tat" del politologo Robert Axelrod, il primo a lanciare tornei di gioco al IPD dove chiunque poteva caricare agenti non adattivi con strategie di gioco specifiche.

#GameTherory #ai #NeuralNet

---[music: Jimi Hendrix]

Dự án LLaMA 3.2 3B phát triển công cụ phân tích mới giúp quan sát các lớp transformer và token mức độ chi tiết, hỗ trợ điều chỉnh kích hoạt kết nối theo thời gian thực. Tính năng đáng chú ý: nhãn chiều nổi bật khi hoạt động mạnh, giao diện điều hướng token linh hoạt. Cơ hội học hỏi đáng ngạc nhiên từ người không phải phát triển Godot.
#AI #MachineLearning #LLaMA32 #FMRI #PhânTíchToken #NeuralNet

https://www.reddit.com/r/LocalLLaMA/comments/1pnhi27/llama_32_3b_fmri/

@ret @farewell_ladmin well, with #TETRA, the advantage is that it's basically #ETSI trying to make a "secure" #GSM fork and thus TETRA Radios constantly transmit pings like a GSM phone.

  • Espechally the vehicle- and on-persin worn radios as they also act as safety devices, sending distress beacon signals when flipped over (which is why it's explicitly being instructed to always put the radios on a desk in an upright/vertical position and never lay them down when switched on!)…

They are intended to allow personnel to discreetly call for backup and/or be found if they go radio silent or seem incapacitated. They don't transmit vitals but I think we all can agree that a patrol presumably laying on the ground whilst on duty is a clear enough sign for dispatch to send an ambulance and reinforcements.

  • Don't ask me how I know…

So yeah, one can to an extend detect them and it's propably a matter of building some #NeuralNet and train it with passing-by data as well as integrating that with #GNUradio.

  • OFC the results will be best in open countryside at a vantage point than at strert level in a big city.

And whilst it may warn against a regular patrol car, it won't help against #SWAT-esque units like #GSG9 that operate basically exclusively in #DMO mode aka. "direct #PTT" use without registering into a TETRA network, so instead they only get their signal "digipeated" (at most) by local TETRA cells based off the upload frequency and thus only can be located when they transmit.

  • Plus they tend to sometimes use civilian / commercially rented transports for a more stealthy approach, putting an entire platoon on a "boxtruck" van...

This #NeuralNet2025 conference seems interesting:
https://neuralnet2025.sciencesconf.org/resource/page/id/1

where: Bordeaux Neurocampus (France)
when: Wednesday 26 noon to Friday 28 noon
preceded by: minischool on spike sorting (Tuesday 25 - Wed.26).

#NeuralNet #NeuroConf #Neuroscience #SpikeSorting

14th meeting of Neuralnet (CNRS GDR) - Sciencesconf.org