Dave Dumps

#SideProject Have you ever been bored and just started organizing your desktop like it's a tiny room, all neat and perfectly arranged? Well, I was going through exactly that today. As I was doing this, I had the idea to automate it using AI/ML. If it's possible, how would a system like this work? Before I dove deep into the project, I already had a vague idea of using t-SNE—a dimensionality reduction algorithm that maps high-dimensional vectors to a lower-dimensional representation while conserving the local similarity between close points in space. Great! That cleared up a lot of ambiguity from the start. So, where to next? The features to apply the t-SNE algorithm to. This wasn't straightforward, as there isn’t anything stored by these icons that contains high-level features to compare with other icons—or is there? *Insert VSauce Audio Here* We humans use two specific properties of a desktop icon to decide if it's the thing we're looking for: the name and the icon, or, in the case of a file in a folder, just the name. That already gives us what we need to tell one icon apart from another, but… how do we turn the meaning it holds into a vector of numbers (features)? Embeddings. Word embedding, or word vector, is an approach used to represent documents and words. It is defined as a numeric vector input that allows words with similar meanings to have similar representations. It can approximate meaning and represent a word in a higher-level, lower-dimensional space, as opposed to the endless and sparse one-hot encoded form. We can use these embedding vectors to extract semantic features from an icon. However, most of the time, icon names are so vague or mean something completely different without further context. For example, Blender, the 3D software, might be confused with an actual blender, a home appliance. So, this invalidates the idea of using the icon name’s embedding as a feature. We need to ensure more context is fed to the embedding model. We can use an LLM by instructing it to describe what the desktop icon is by giving it the name, type (folder/file), and full path. It will be able to understand exactly what we're talking about to the fullest extent. By bundling the LLM explanation with the icon name and then embedding it, we finally have a way to accurately represent these icons with a vector in a latent space, where similar icons are located close to each other. Now that we have the feature vectors of the icons, how do we determine their desired location on the desktop? It’s not as simple as you might think. These embedding vectors consist of over 780 floating-point numbers—how could we possibly map this to a 2D point on the desktop? We use a dimensionality reduction algorithm. In my case, I chose t-SNE. t-SNE finds a way to remap the 780+ dimension space into just two, while still preserving the structure of the higher-dimensional latent space. Once that’s done, we can finally start working on the low-level desktop management tasks—which, believe it or not, was the hardest part of this whole project. The Windows Desktop API is the least documented thing I’ve ever encountered in all of Win32API. I thought it would be a simple P/Invoke or something, but I was wrong. Luckily, I found a premade C# library called DesktopIconsManipulator that did exactly what I needed. And just like that, a simple shower thought turned into a complete project. I’ll be posting a demo, and most importantly, the repo will be made open-source for anyone interested in the inner workings. #SideProject #Latent #Utility @TheKerVerse

እርግጥ በአማርኛ እንላለን.... 😂

"ESHI MALET ESHI NEW"

🤣🤣🤣🤣

I hope you know I'm rooting for you

this is free and open access alternative to cell press or any other paid journals out there and the do have some good stuff https://www.biorxiv.org/

While its not fully open like arixv, some articles are free , they do have some cutting edge stuff https://www.cell.com/current-biology/home

lets get quarky

Exciting times ahead! We're building a Quantum Computing Team and looking for curious minds with some dev experience and a passion for learning✨ If you're ready to grow and explore cutting-edge tech, fill out this quick form and join us on this journey! https://forms.gle/koGpfzEqu4zU3XfG9

after i posted this elpa edged me 3 times ahun demo metual eski

before moving forward I just want to give my gratitude FUCK YOU ELPA reminding me Christmas is on edge