Alex Bodner
@AlexBodner_
Followers
1K
Following
18K
Media
296
Statuses
4K
AI engineering at @UdeSA๐ฆ๐ท | intern @roboflow Posting on AI progress and my own projects, check them out: https://t.co/RDtxC6yWqE
Joined May 2016
Iโm 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 it๐งต
32
152
1K
Remember this day
GPT-5 just casually did new mathematics. Sebastien Bubeck gave it an open problem from convex optimization, something humans had only partially solved. GPT-5-Pro sat down, reasoned for 17 minutes, and produced a correct proof improving the known bound from 1/L all the way to
0
0
3
donโt tell me magic doesnโt exist in real life because we literally took sand and made it predict the next token such that it can answer almost any question you throw at it
30
50
842
๐ค Simposio Cientifico de Inteligencia Artificial y Aplicaciones ๐ค Abiertas las pre-inscripciones al SCIAA-@UdeSA Fecha limite de inscripcion y poster-abstracts: 8 Agosto. + info -> https://t.co/mIZFg1r1zI
1
6
17
we plugging ViTPose into Basketball AI according to @NBA rules, a player is considered to be in the paint only if both feet are inside the paint notebook: https://t.co/qiNamenCo2
ViTPose++ is crazy good! look at the interaction between pink and green player. getting that right is really impressive. we are plugging it into basketball AI.
9
59
510
Just released with @skalskip92: Detect the 3 second violation in NBA videos with AI. Featuring the @roboflow blog and open source code๐ฅ Learn how we did it in the blog: https://t.co/AZr3mTYWa3
blog.roboflow.com
Learn how to build an AI system for 3 second violations using player detection, tracking, and dynamic zone monitoring.
first part of basketball AI project is out: detect @NBA 3 second violations together with @AlexBodner_ we published blogpost and jupyter notebook; links below links to datasets and fine-tuned models here: https://t.co/VK0RQFWud1
1
7
44
@skalskip92 @AlexBodner_ with the detailed guide and colab
blog.roboflow.com
Learn how to build an AI system for 3 second violations using player detection, tracking, and dynamic zone monitoring.
0
2
5
first part of basketball AI project is out: detect @NBA 3 second violations together with @AlexBodner_ we published blogpost and jupyter notebook; links below links to datasets and fine-tuned models here: https://t.co/VK0RQFWud1
21
42
363
One of the many cool things we are working at @roboflow ๐ฅ
weโre experimenting with SAM2 for the automatic detection of 3-second violations in NBA games it's pretty tricky, but @AlexBodner_ is working hard on the algo! supervision: https://t.co/xXMRaS3Guk
2
0
24
๐ Just published my first blog post with @roboflow! ๐๏ธ How do computers "follow" things in a video?๐๏ธ Itโs called ๐ผ๐ฏ๐ท๐ฒ๐ฐ๐ ๐๐ฟ๐ฎ๐ฐ๐ธ๐ถ๐ป๐ด and in this blog post I explain the ๐ฆ๐ถ๐บ๐ฝ๐น๐ฒ ๐ข๐ป๐น๐ถ๐ป๐ฒ ๐ฎ๐ป๐ฑ ๐ฅ๐ฒ๐ฎ๐น๐๐ถ๐บ๐ฒ ๐ง๐ฟ๐ฎ๐ฐ๐ธ๐ถ๐ป๐ด (๐ฆ๐ข๐ฅ๐ง) algorithm in depth,
3
4
68
Basketball AI is taking form๐ฅ
I can finally map @NBA player's position from the camera perspective onto the court map it's still a bit shaky... I'll smooth it out later it's time to detect shooting motions and mark the shot location! some of the code has already been migrated to: https://t.co/VK0RQFWud1
0
0
9
๐ฅToday I found out that my first paper surprassed the 100 cites๐ฅฒ Iโd like to thank everyone who supported me in the process This is just the beggining๐
0
1
15
We're missing (at least one) major paradigm for LLM learning. Not sure what to call it, possibly it has a name - system prompt learning? Pretraining is for knowledge. Finetuning (SL/RL) is for habitual behavior. Both of these involve a change in parameters but a lot of human
726
1K
10K
Life update: ๐ Super excited to share that Iโve joined the Open Source team at @roboflow as an intern! During this intership, I will be working on trackers โ a unified, from-scratch implementation of state-of-the-art object tracking algorithms. Taking part of this project
7
10
158
trackers v2.0.0 is out combo object detectors from top model libraries with multi-object tracker of your choice for now we support SORT and DeepSORT; more trackers coming soon link: https://t.co/9Fam5U1zuC
32
249
2K
If you liked this compression perspective on why Unsupervised Learning works, I highly highly recommend watching this incredible talk from Ilya It is an absolute classic and you get to hear directly from the man himself about his intuitions
9
74
1K
I finally got tired of not fully understanding ROPE (Rotary Positional Embeddings), so I went back to the original paper. In short: ROPE replaces the classic positional vector by rotating embeddings using two sine waves. When people say "ROPE of 2k", "500k", or "8M", they're
7
55
359