Markos Gavalas

Calling all #statisticians, #programmers, #dataanalysts and #datascientists - you don't want to miss this conference! Explore new horizons with Clivoffective at The Effective Statistician conference in fall 2024. Dr. Alexander Schacht, Evelyn Hund and André Hund as well as many other speakers will provide you with valuable tools for your daily business through presentations, inputs and workshops. Take a look at the program now and be sure to register today. You can find the link in the comments👇🏼

26

2 Kommentare

Data Scientists, This is BIG! Snowflake's Arctic LLM is Here – and Open-Source! 🤯 Get your hands on it now: Interactive Demo: Explore Arctic's capabilities with this Streamlit app → https://lnkd.in/gJ7y38Ng Replicate: Ready to experiment? Find it on Replicate → https://lnkd.in/gyDXiMju Why This Matters: Arctic lets you quickly build demos of your models. This is a game-changer for showcasing your work and making your models more accessible! Huge thanks to Snowflake for making this happen! Are you excited to try Arctic? Let me know in the comments! 👇 #LLM #OpenSource #DataScience #Streamlit #Snowflake

8

Excited to share a guide that helps you choose the right regression algorithm based on data understanding and objectives. Before diving deep into notebooks and implementing multiple algorithms with low accuracy, this guide will help you make informed decisions. 🔍 What's Inside? Detailed explanations of top 15 regression algorithms Real-world examples Practical Python code snippets 🔗 Read the Full Guide Here https://lnkd.in/dpZKRBzW Let's improve our predictive modeling skills and achieve higher accuracy together! #MachineLearning #DataScience #AI #RegressionAlgorithms #Learning #ProfessionalGrowth

3

#buildinginpublic #pccloner Post 18/N After 20+ hours of work distributed over two and a half weeks, I finally built the first complete version of this architecture from DeepMind. Of course, I had about 12 points of where I'm not sure what the authors did, but I know exactly where those points are and I can make changes easily in most of them. I still have to train this model, but up to today it is working well with expected data from miniwob++ benchmark. This model has more moving parts than I thought. Paper: https://lnkd.in/di6hH29m

1

FlexAttention: Simplifying Attention in PyTorch _________________________________________________________________ FlexAttention is a new PyTorch API that combines the flexibility of PyTorch with the performance of optimized attention methods. It allows researchers to implement various attention mechanisms easily without needing to write custom kernels. Key Points: - Flexibility with Performance: -- Traditional attention methods offer high performance but limited flexibility. -- FlexAttention provides a flexible API that supports many attention variants with minimal effort. - How It Works: -- Standard Attention: Calculates attention scores using queries, keys, and values, normalizes them, and computes the weighted sum of values. -- FlexAttention: Allows for modifications to the attention scores through a flexible scoring function, enabling customization for different types of attention mechanisms. - Examples of Attention Variants: -- Relative Position Encoding: Adds positional information to attention scores based on relative positions. -- ALiBi Bias: Applies a bias to attention scores to account for position differences, enhancing model performance. -- Causal Mask: Masks future positions in the sequence to ensure the model only attends to previous positions, useful for autoregressive models. Handling Sparsity: -- BlockMask: Efficiently manages sparse attention masks, such as those used in causal attention, by creating masks that prevent unnecessary computations. Learn More: https://lnkd.in/g58pxB4A

8

Believe it or not, there are 4 use cases for `else` in Python. https://lnkd.in/eBhZ32en

3

Amazing work.👏🏼 This interesting paper introduces SheetCompressor, a new way to compress spreadsheets for large language models. It has three parts: structural-anchor-based compression, inverse index translation, and data-format-aware aggregation. #llms #chatgpt #machinelearning #excel #ai #nlp https://lnkd.in/ddfrYWjB

2

Anomaly detection is about finding rare events very different from most data. It's used in many areas, like making things and healthcare https://lnkd.in/emyTe7Dh

1

When writing code, you should make sure it's readable & works as it's designed to do. That's where testing comes into play. This course covers class-based tests, fixtures, mocking, & more as you learn how to test Python code using the Pytest framework.

431

1 Kommentar

📘 Just published a new blog post diving into the math behind self-attention in #DeepLearning! 🤓 I'm a visual learner - until I get a visual intuition behind a concept it's very hard for me to understand it properly. This blog is such an attempt to understand self-attention and how it is so effective. Thanks to Luis Serrano for publishing an excellent video on this topic. A big thanks to Sebastian Raschka, PhD for this amazing book-Build a large language model from scratch. 🔗 https://lnkd.in/gUnWPdJ6 #nlp #attention #MachineLearning #Datascience

365

2 Kommentare

Kolmogorov-Arnold Networks (KANs)[1] are the new RAGs in terms of the latest buzz in the AI landscape. Non-parametric regression using cubic B-Splines is a well known technique in the classical non-parametric statistics, which seems to be the basis (no pun) of KANs. In fact, I implemented a Bayesian cubic B-splines to scale and fuse datasets in the latent space, more than a decade and half ago. In the modern deep learning world, this is known as (non-parametric) layer normalisation :). I admit, it was notoriously hard to implement and get it right. Apparently, the connections between KAN and two layer neural networks appeared much before (as early as 1985 or even earlier perhaps)[2]. KANs have scaled them depth-wise and come with some tricks. So, we see this template at play again Stack up all classical non-parametric regression techniques like Kernels, Splines, Polynomials, Wavelets and learn the layer-specific regression parameters via gradient descent. Deep Learning, as a scientific computing techology, does the rest of the magic. You potentially have a factory of new deep learning architectures. [1]KAN: Kolmogorov-Arnold Networks https://lnkd.in/gcrVMbPn [2]Kolmogorov's Mapping Neural Network Existence Theorem  https://lnkd.in/g9C26TCg

236

11 Kommentare

I really liked reading this article as it provides an insightful explanation of quantization, a technique commonly used in machine learning to simplify models. It covers key concepts such as scalar and vector quantization, and illustrates their practical applications (Quantization-Aware Training (QAT), Post-Training Quantization (PTQ)) with clear visuals. https://lnkd.in/gYKw3PDu

23

Quantized the Open LLM Romanian model to 4 bits. It runs now in 1 sec/answer (vs the original version which is taking 40 seconds on GPU T4 x2 - see excelent post by Gabriel Preda ), it uses only 7GB of Vram and you can try it on gradio (hhttps://lnkd.in/dxZQTRFA) or build it youself in 3 easy steps as explain on our page - https://lnkd.in/dXE9mWdu This 4bit version is based on the Open LLM from ilds.ro You can download the 4bit model from HuggingFace: https://lnkd.in/dvZiyW6D

53

5 Kommentare

Yes, we can now perform function calling on images using Claude! The Claude 3 family of models by Anthropic now includes enhanced vision capabilities to read and process images. This development allows efficient extraction of data from tables, charts, graphs, and text within images or document images. For a detailed explanation of how these capabilities can be leveraged, especially through the Claudetools package, please view the detailed blog post linked below. ⛓️‍💥 https://lnkd.in/gf8QcKJy

20

1 Kommentar

Here is my blog of this week: "Implementing simple linear regression from scratch using Julia Programming Language". In this blog, we will look at the Math behind the implementation of the simple Linear Regression, then subsequently implement this Linear Regression from scratch in Julia Programming Language for the California Housing dataset and finally, compare the results of this implementation with the results obtained using scikit-learn in Python for the same the same dataset and I found that the my implementation of Linear regression closely matches the implementation of Linear Regression in scikit-learn. Here is the blog: https://lnkd.in/gjtCmkwm. Implementing regularization will be a topic for another blog I learnt two things from this blog: 1. Julia Programming Language, a new programming language for me 2. Embedding latex like Math equations in HTML 😍. Your thoughts and feedback are welcome #machinelearning #deeplearning #AI #linearregression #regression #mathematics #artificialintelligence

2

1 Kommentar

LLMs do approximate retrieval of stored information on a computer, and that's great for many tasks, but not directly comparable with human smarts

4

1 Kommentar