Ghadi Al HajjPaper Summary: “Self-Supervised Disentanglement by Leveraging Structure in Data Augmentations”This paper proposes a data augmentation strategy that helps the model disentangle the input's "style" components rather than discard them…Aug 14Aug 14
Ghadi Al HajjPaper Summary: “Mixture of A Million Experts”This paper proposes a way to scale an MoE architecture to, as the name suggests, millions of experts. But first, what is an MoE, and why…Aug 6Aug 6
Ghadi Al HajjPaper Summary: xLSTMWarning: this paper has many details, so I’ll skip some for brevity and focus on the main points :)Aug 3Aug 3
Ghadi Al HajjPaper Summary: “Slow and Steady Wins the Race Maintaining Plasticity with Hare and Tortoise…This paper is the first in a series where I write summaries of the interesting papers I read.Aug 2Aug 2
Ghadi Al HajjThe Hyena Operator: Say Goodbye to Self-Attention?Hyena arrives with prowess, self-attention pales, Convolution’s might in stride, efficiency prevails. (by ChatGPT)May 25, 20231May 25, 20231
Ghadi Al Hajj“Segment Anything” Foundation Model for CV from Meta — An OverviewIf language foundation models are not enough, here’s the Segment Anything, an foundation model for Computer vision.Apr 9, 2023Apr 9, 2023