Josep Ferrer’s Post

🗓️ Day 21 of 100 Days of Machine Learning🚀 Today, I took a deep dive into the world of K-Nearest Neighbors (KNN) algorithms! 🧠🔍 KNN is a powerful and intuitive classification technique that works by identifying the closest data points (neighbors) to make predictions. The simplicity and effectiveness of KNN make it a go-to method for many real-world applications. 🌍💡 Excited to continue exploring and implementing this algorithm further! Onward and upward in this learning journey! 💪📈 #MachineLearning #100DaysOfML #DataScience #KNN #LearningJourney

good quality 10K dataset is all that you need to near GPT4 benchmarks as claimed by the YI model paper "Quality is All You Need Our finetuning dataset consists of less than 10K multi-turn instructionresponse dialog pairs, with each and every one of the entry constructed and polished over multiple iterations and from user feedback. We take this approach because in our preliminary experiments, we observe that compared to the open-source data of several hundred thousand entries, the results from a smaller, manually annotated dataset are superior. These observations align with those reported in Gemini Team et al. [23], Touvron et al. [77], Zhou et al. [94]." #llmfinetuning

🚀 Mastering Retrieval-Augmented Generation (RAG): Beyond Basics 🚀 Are you pushing the boundaries of your RAG pipelines and looking to enhance retrieval efficiency? I’ve just published a new blog article that dives deep into advanced chunking strategies, Qdrant, and "reranking" methodologies to boost retrieval performance all powered by LlamaIndex. https://lnkd.in/gRFcwrGe #GenAI, #LLM

Two great blog posts by Julien Heiduk on how to use ZenML with MLflow ➡ https://lnkd.in/dhaV8FYx ➡ https://lnkd.in/dyW34QuY In the words of Julien: "By adding MLflow to your ZenML pipeline, you can effectively track your machine learning experiments, making it easier to monitor, analyze, and reproduce your work. The integration shown in this tutorial allows you to log important parameters and metrics in a structured way." Love the community content <3!

Just trying to construct an alternative dataset for Polyphonic-TrOMR (https://lnkd.in/ebbfqPCe), aiming to transition it from "open source" - a machine learning model lacking a public training set - to a truly open-source project. Let's see how that goes. With an estimated training time of approximately 79 hours, it's a compelling reason to tidy things up and go into the weekend.

🚀 𝗝𝗼𝗶𝗻 𝗠𝘆 𝟱-𝗗𝗮𝘆 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗥𝗲𝘃𝗶𝘀𝗶𝗼𝗻 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲! 🚀 Sharpen your ML skills with this 5-day challenge! Each day covers a key algorithm, its principles, and a hands-on project. Let's elevate our understanding and application of Machine Learning together! 𝗗𝗮𝘆 𝟯: 𝗦𝘂𝗽𝗽𝗼𝗿𝘁 𝗩𝗲𝗰𝘁𝗼𝗿 𝗠𝗮𝗰𝗵𝗶𝗻𝗲𝘀 (𝗦𝗩𝗠) 🔍 Algorithm: Finds the optimal hyperplane that separates classes by maximizing the margin. 🛠️ Project: Spam Email Classification 📊 Tasks: • Load and preprocess the dataset. • Convert text to numerical features using TF-IDF. • Implement and train an SVM classifier. • Evaluate performance with precision, recall, and F1-score.

K-Nearest Neighbors (KNN) is a widely-used machine-learning algorithm, and for good reason. Its simplicity, versatility, and adaptability make it a favorite among data scientists. But what makes KNN so popular and effective? In this article, we peel back the layers of KNN to reveal how it works and why it remains relevant in a field driven by the latest technologies. We'll dive into its applications, explore the math behind it, and even build it from scratch. If you're interested in machine learning, you won't want to miss this deep dive into KNN.

🚀 Day 2 of My MLOps Journey 🚀 Today i Cover these Topics deeply : 1. Supervised Learning 2. Regression 3. Algorithms of Regression 4. Classification 5. Algorithms of Classification Today Notes 👇

🚀 Exploring CNN Architectures for Dog Breed Recognition! 🐶📊 I recently delved into three powerful CNN architectures—AlexNet, GoogleNet, and ResNet-50—to tackle the fascinating challenge of dog breed recognition. Here’s a quick overview of what I’ve learned: • AlexNet: A pioneering model that set the stage for deep learning in computer vision. With its simple yet effective architecture, it uses deep convolutional layers and ReLU activation to handle complex image recognition tasks. • GoogleNet (Inception V1): This architecture introduced the Inception module, which efficiently combines multiple convolutional filters and pooling operations. It achieves a remarkable balance between depth and computational efficiency. • ResNet-50: Known for its innovative use of residual connections, ResNet-50 allows the training of very deep networks by bypassing layers, addressing the vanishing gradient problem, and enabling exceptional performance in image classification. For a detailed walkthrough and hands-on implementation of these models for dog breed recognition, check out my Kaggle notebook: https://lnkd.in/gvQx38gv #MachineLearning #DeepLearning #CNN #AlexNet #GoogleNet #ResNet #DogBreedRecognition #DataScience #AI

🌟 Day 9/20: October Daily Learning Challenge 🌟 Today, I tackled the Sliding Window Technique in Data Structures & Algorithms (DSA)! 🔍 This powerful approach helps solve problems involving subarrays, substrings, and more by reducing the time complexity from brute force methods to something much more efficient. 💡 Key takeaways: 🔸Sliding Window allows for dynamic range processing with minimal overhead. 🔸Great for problems like finding the max/min sum in a subarray, longest substring without repeating characters, and more! Excited to apply this technique to real-world problems! 🚀 Masai #dailyLearning #DSA #SlidingWindow #Algorithms #ProblemSolving #EfficientCode #LearningJourney #OctoberChallenge