Sascha Kirch
@sascha_kirch
🎓 PhD Student in Deep Learning @ UNED 🚙 Expert Deep Learning @ Bosch 🤖 Collaborating Researcher @ Volograms ⚡️President Elect @ IEEE Eta Kappa Nu Nu Alpha
ID: 1479556576260276224
https://sascha-kirch.github.io/ 07-01-2022 20:52:44
89 Tweet
47 Followers
42 Following
"On our journey towards Mamba selective state space models and their recent achievements in research, understanding the state space model is crucial." Sascha Kirch explores the Mamba state space model in a new, well-illustrated deep dive. buff.ly/4dwUE9b
The Mamba model architecture has generated a lot of buzz as a potential replacement for the powerful Transformer. Sascha Kirch shares the first in a series of articles, aiming to look closely at its inner workings and potential use cases.. buff.ly/3SL95Ov
In part three of his top-notch series on Mamba state space models, Sascha Kirch turns to use cases focused on images, videos, and time series. buff.ly/3Z6C0R4
"The Structured SSM approximates the context using the HiPPO matrix resulting in some compression, while it can be trained more efficiently as the RNN because of its convolutional representation." Sascha Kirch's deep dive explores Mamba state space models for images, videos,
Dive into the future of image processing with Vision Mamba! Unlike Transformer models, Vision Mamba’s sub quadratic scaling is a game-changer for dense-prediction tasks on high-res images. Read Sascha Kirch's full article now. towardsdatascience.com/vision-mamba-l… #DataScience
Ever wondered how Vision Mamba outperforms Transformers in handling long sequences and high-resolution images? It’s all about state representation! Discover the innovative design choices making waves in vision tech, written by Sascha Kirch. #DataScience #MachineLearning
The Mamba model architecture has generated a lot of buzz as a potential replacement for the powerful Transformer. Sascha Kirch recently shared the first in a series of articles, aiming to look closely at its inner workings and potential use cases. towardsdatascience.com/towards-mamba-…
Being equipped with the Mamba selective state space model, we are now able to let history repeat itself and transfer the success of SSMs from sequence data to non-sequence data: Images. 🖊️ by Sascha Kirch | #DataScience #Programming towardsdatascience.com/vision-mamba-l…
You may have heard the phrase: “Attention scales poorly with the sequence length N, specifically with O(N²).” In this article, Sascha Kirch explores why attention is so slow & resource-intensive on modern GPUs, and how FlashAttention addresses the issue. ai.gopubby.com/5a9f2407d739