Alex Bodner

open source @roboflow | AI engineering at @UdeSA🇦🇷
prev founder @SatsOnFire

[Alex Bodner @AlexBodner_]

I’m thrilled to present the KAN Convolutional Layer, a promising neural architecture for image processing. Last week KANs came out, as an alternative to the MLP. We extended this idea to Convolutional Layers, creating the KAN Convolution. Join me in this thread to more about it🧵

[mirrash.eth 🦇🔊 @mirrash7]

Full breakdown of how to build a 3D soccer offside (VAR) system
https://t.co/x7iRwND9K0

[Alex Bodner @AlexBodner_]

@mirrash7 Nice one !

[SkalskiP @skalskip92]

supervision just hit 45k github stars; 5k in 3 weeks

link: https://t.co/xXMRaS3Guk

[Alex Bodner @AlexBodner_]

@tokufxug Thanks for sharing our work!

[Alex Bodner @AlexBodner_]

dont know what IoU variants are?
find the detailed explanation in the docs
https://t.co/O4VcsvRXn3

[Alex Bodner @AlexBodner_]

⚽Look at that ball tracking!
We have just released Roboflow trackers v2.5.0, making the tracker's association step customizable with different IoU variants, inspired by the CBIoU tracker.
💡 What this means is that, when associating existing tracks with new detections, the tracker can now leverage richer signals beyond standard Intersection over Union, including box distance (DIoU) and aspect ratio (CIoU).
🌐 This opens a whole new world of possibilities for improving object tracking in challenging scenarios, such as small and fast moving objects or irregular motion.
🫂 Excited to see what the community builds with these new features!
Congrats to everyone involved in this release, and much more to come from this incredible team! 🚀
See comparison in ball tracking by incorporating IoU variant CIoU.

[Alex Bodner @AlexBodner_]

release notes: https://t.co/9Su2oXRwHT
docs: https://t.co/w2tKwGgdtQ

[Alex Bodner @AlexBodner_]

@econar_data Al ser suma de muchas variables aleatorias (cada gol y asistencia), por TCL no debería tender a normal?

[Alex Bodner @AlexBodner_]

What the

[Surya Ganguli @SuryaGanguli]

Our new paper lead by @vedanglad w/@AToliasLab "Letting the neural code speak..." https://t.co/B6H1Vvcmso
We show how to get *monkey* visual neurons to TELL us in *human* language what images make them fire. We do this is an automated verifiable way at scale! How?

1) Build a digital twin of monkey visual areas that can accurately map visual inputs to neural activity.

2) Perform in-silico experiments on this twin to find many complex images that make a model neuron fire.

3) Use a vision-language model to describe these complex images.

4) Verification: use a language-conditioned diffusion model to generate new images, and check they make the monkey digital twin neurons fire a lot.

To our knowledge, for the first time, we have a way to convert monkey vision to human language, and from *human* language to sample infinitely many images that make any given *monkey* visual neuron fire, all in an algorithmic fashion.

For more exciting details, see @vedanglad's excellent thread!

[SkalskiP @skalskip92]

RF-DETR keypoints is finally out

preview release: real-time transformer keypoint detection

Apache 2.0

71.8 AP on COCO, 9.7ms on T4. outperforms YOLO11-pose and YOLO26-pose at similar latency

[SkalskiP @skalskip92]

basketball + computer vision

- RF-DETR - player, number, ball, basket detection
- SAM3 - player tracking
- SigLIP2, UMAP, K-Means - team clustering
- GLM-OCR - number recognition

[Alex Bodner @AlexBodner_]

@skalskip92 Did we just ratio’ed DeepMind?

[Alex Bodner @AlexBodner_]

@SamJWasserman @skalskip92 Thank you!

[Alex Bodner @AlexBodner_]

@Juanchodatt @skalskip92 Todavía no estamos en su nivel (todavía)😉

[Alex Bodner @AlexBodner_]

Coming soon👀

[SkalskiP @skalskip92]

looks like we are working on the same thing haha; more soon

[Alex Bodner @AlexBodner_]

@jxmnop Agents for iterating QuickType and AI assisted programming for complete certainess

[Alex Bodner @AlexBodner_]

Yolo before NMS

[Martin Shkreli @MartinShkreli]

isn't it beautiful? what does it remind you of?

this is a KEGG pathway chart.

[Alex Bodner @AlexBodner_]

What’s the best way you’ve found to use LLMs as a second brain to keep up with research?

[Alex Bodner @AlexBodner_]

whattt @cursor_ai

[Alex Bodner @AlexBodner_]

Buenas Maxi!
todo se basa en estos papers de Anthropic:
- https://t.co/w48TMBqzx0
- https://t.co/K6M65FzR89
uno de los investigadores tambien proveyó el modelo de NLA entrenado para varias versiones de Qwen, y nosotros deployeamos en una GPU alquilada la version de 7B, que implica desplegar 2 modelos, uno que es el que responde las consultas y otro que es el que "decodifica los pensamientos".
El repo de nuestro proyecto de la Hackathon es:
https://t.co/1Vn3tlkwM9
Es un tema super interesante, cualquier duda charlemos!

[Alex Bodner @AlexBodner_]

@AnthropicAI published a way to read LLM’s inner thoughts last week (Natural Language Autoencoders).
2 days later, we spent 36h straight hours at @platan_ventures Hack Buenos Aires, were we built an open-source system that uses it to detect deception and steer models back into alignment.
Here’s what we found: