Dharmendra Kumar Yadav

Fine tuning LLM model using PEFT technique. https://lnkd.in/gmp9eEet #finetuning #llm #genai #huggingface #peft #evaluation Hugging Face Medium #mlops #transformers

30

"The complexity of causal inference lies in the fact that we want to measure something that we will never actually observe. To measure a causal effect, you need a reference point: the counterfactual. The counterfactual is the world without your treatment or intervention. The causal effect is measured by comparing the observed situation with this reference (the counterfactual)."

2

❓ 🤔 Which Metric to Choose? - Accuracy or Precision or Recall ? 👓 🔎 🎯 How to decide which metric is the best one for evaluating the performance of a Machine Learning Model for Classification Task i.e that predicts a Class for a given set of Inputs. 👨‍💼 This is one of the most asked interview questions, and also its important to know on what basis a Model should be judged. 👨‍🏫 In this Video I have decoded and simplified how and why to choose a particular evaluation metric by explaining reasons and giving examples. 📋 This is a 1.5X speed version of original video, so that it can fit in linkedin. For original video : https://lnkd.in/gNiG97_X or select 0.75 speed over here!! 👨‍💼 Watch previous post where I explain the metrics: https://lnkd.in/gxQcPmE4 Consider these 3 ML use cases: 🙂 😀 😃 🕵‍♂️ Case 1 : Fraud Detection : What is the metric ? 👨‍⚕️ Case 2: Cancer Prediction : What is the metric ? 🏦 Case 3: Credit Reduction Prediction : What is the metric ? Post in comments 😀 Its really easy, at least after watching this video it should be easy. 😃 🚀 For more such fun and enjoyable learning, follow me at 👩‍🚀 https://lnkd.in/gdkvK9wc #machinelearning #datascience #AI #metrics #machinelearningmetrics #metricsforclassification #accuracy #precision #recall #f1score #evaluation #businessmetrics #kpis #technicalmetrics #naturallanguageprocessing #nlp #LLMs #dataanalysis #statistics #data #supervisedlearning #ml #mlinterviewquestions #datascienceinterview #datascienceinterviewquestions #machinelearninginterviewquestions #machinelearningtutorial #datasciencetutorial #machinelearningbasics #datasciencebasics

11

5 Comments

Unveiling Breakthrough: My Simple Algorithm Redefines Anomaly Detection Witness My Supreme Genius, IEEE JOURNAL OF MACHINE LEARNING RESEARCH INC, Machine Learning, Anomaly Detection/Analytics, Data Science I've explained it in simple words below. Imagine you have a group of points arranged in a circle (it's an ellipse but for simplicity I'm calling it a circle), and you're trying to find if a new point is an anomaly, meaning it's significantly different from the rest. You define the circle's size by its widest parts along the X and Y axes. Now, for a new point outside this circle, you measure how far it is from the circle's leftmost and rightmost points on the X axis (X1 and X2) and from the top and bottom points on the Y axis (Y1 and Y2). If any of these distances are greater than the circle's width or height (DX or DY), then the new point is considered an anomaly. This is an absolutely original algorithm. This entire work has been done by me, there are absolutely no references, no co-authors and no other external inspiration. I've submitted the preprint to Research Square , Research Square (Platform), Research Square and it's undergoing pre-screening. I don't really care whether It is published by IEEE or any great journal. But I'll make sure to make a lot of noise to ensure many people get to know about this. COMMENT FOR BETTER REACH. Krista King Nikolai Schuler Data Science Yash Diniz Kevin Antony Harpreet Sahota 🥑 Natassha Selvaraj Jose Marcial Portilla Danish Tamboli Daniel Bourke Kyle Simpson National Institute of Technology, Goa CSIR - National Institute of Oceanography CSIRO CSIR - National Institute of Oceanography Indian Institute of Science (IISc) Indian Institute of Technology, Bombay Harvard University Massachusetts Institute of Technology Stanford University Sal Khan Alex Freberg Andrew Ng Sanmitra Bharat Naik #DataScience #MachineLearning #ArtificialIntelligence #DataAnalytics #Mathematics #Technology #Algorithm #AnomalyDetection #Innovation #TechInnovation #AI #DeepLearning #DataMining #DataDriven #Research #Science #TechCommunity #TechLeadership #STEM #BigData #Hiring #DataScientist #Professor #Mathematics #Scientist #IEEE #Journal #Paper

7

1 Comment

🤖 Sarvam AI launches open-source foundational models in 10 Indian languages Sarvam AI has made a significant leap in AI with the launch of its open-source model supporting 10 Indian languages, including Hindi, Tamil, Telugu, and more. They’ve also introduced Sarvam 2B, a 2-billion-parameter Indic language model, and a suite of GenAI services for enterprises. Key Highlights: - Multilingual and Voice-Enabled: Revolutionizing AI accessibility across India’s diverse linguistic landscape. - Sarvam 2B: India’s first open-source, 2-billion-parameter Indic LLM. - Sarvam Agents: Multilingual business agents starting at ₹1 per minute. - Shuka 1.0: Open-source AudioLM for accurate voice-to-text conversion. - Sovereign AI: Focused on inclusive, efficient, and sovereign AI development. Why It Matters: Sarvam AI is setting a new standard in AI development in India, driving the next phase of AI adoption by making technology accessible to all, regardless of language or socio-economic status. What impact do you think Sarvam AI will have on India's AI landscape? #sarvamai #ai #innovation #technology #multilingualai #indicllm #indiaai

45

🎨 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴: 𝗧𝗵𝗲 𝗔𝗿𝘁 𝗼𝗳 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗶𝗻𝗴 𝗬𝗼𝘂𝗿 𝗗𝗮𝘁𝗮 🎨 Feature Engineering is a crucial step in preparing your data for machine learning models. It’s like giving your data a makeover, helping models extract better insights. Let’s explore how feature engineering can turn raw data into gold! 🔹 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴? In simple terms, feature engineering is about creating new features from the existing data or modifying them to better represent the information your model needs to make accurate predictions. It’s like giving your model a clearer view of the data. 🔸 𝗪𝗵𝘆 𝗗𝗼 𝗬𝗼𝘂 𝗡𝗲𝗲𝗱 𝗶𝘁? The quality of the features you input into a model can make or break its performance. Well-crafted features can help your model uncover patterns, leading to improved accuracy and better generalization. Sometimes, a simple model with excellent features can outperform a complex one. 🔵 𝗖𝗼𝗺𝗺𝗼𝗻 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗧𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲𝘀 🔵 🟤 1. Handling Date and Time Variables 📅 Date and time columns can provide valuable information beyond just the exact timestamp. You can extract meaningful features such as the day of the week, month, or hour. 𝑬𝒙𝒂𝒎𝒑𝒍𝒆: From a “transaction date” field, you might extract “day of the week” to understand if sales are higher on weekends. 🟠 2. Creating Interaction Features ✨ Interaction features are made by combining two or more existing features to create new ones. This can help the model capture relationships that are not obvious. 𝑬𝒙𝒂𝒎𝒑𝒍𝒆: In a dataset with house price data, multiplying "number of rooms" with "area" can create a new feature “room area” that might give better insights. 🔵 3. Binning Continuous Variables 🛠️ Binning is the process of converting continuous variables into discrete intervals or “bins.” This is helpful when you want to group continuous data into categories. 𝑬𝒙𝒂𝒎𝒑𝒍𝒆: Instead of using raw ages, you can bin them into categories like “young,” “middle-aged,” and “senior.” 🟤 4. Feature Scaling ⚖️ Feature scaling is essential for algorithms that rely on the distances between points (like k-NN or SVM). You can scale data using normalization or standardization. 𝑬𝒙𝒂𝒎𝒑𝒍𝒆: Normalizing the "income" and "age" columns ensures they are on a similar scale, so neither one dominates the learning process. 🟠 5. Encoding Categorical Variables 🏷️ Many algorithms require categorical variables to be converted into numerical form. Techniques like one-hot encoding or label encoding are commonly used. 𝑬𝒙𝒂𝒎𝒑𝒍𝒆: Converting “country” (USA, UK, Canada) into separate binary columns: [1,0,0] for USA, [0,1,0] for UK, etc. 🔹 𝗖𝗼𝗻𝗰𝗹𝘂𝘀𝗶𝗼𝗻 💡 Feature engineering is where your creativity and domain knowledge meet the data. #FeatureEngineering #DataScience #MachineLearning #AI #DataTransformation #cat

11

1 Comment

In Generative AI , its utmost important to know foundation models such as BERT, CLIP, DALL-E, and GPT-3 etc to proceed with any fine tuning . Transfer learning is the ML technique that enables the emergence of foundation models. Based on the accumulated knowledge gained at previous tasks, a model can learn new tasks through this technique. Foundation models need to be adapted because they serve as a base for new models, and there are numerous approaches to do this.

4

Is your data science game more number-crunching or code-munching? 🤔 Find out what's the bigger piece of the pie? . . . . . . #dataScience #datascientist #data #BigData #memepage #memesdaily #memes #statistics #coding #computerscience 

12

Image cropping is crucial for maintaining quality and aspect ratios in various applications. However, machine learning models for automatic cropping can exhibit biases and errors, leading to legal risks. A research team from Doshisha University developed new techniques to generate adversarial examples for image cropping, improving model accuracy. Their white-box and black-box approaches reduce perturbation sizes and target specific image regions effectively. This advancement in AI systems contributes to fairer AI governance and inspires further research in ML.

2

💡 Ensemble Learning: Bagging, Boosting, and Stacking Explained In machine learning, ensemble learning stands out as a powerful technique for combining multiple models to enhance predictive performance. 1️⃣ Bagging (Bootstrap Aggregating) 🦝: aims to reduce variance and prevent overfitting by training multiple models independently on different subsets of the data. Each subset is created through bootstrapping (data sampling with replacement). The final prediction is typically made by averaging the outputs (for regression) or majority voting (for classification). 2️⃣ Boosting 🚀: aims to reduce bias by sequentially training weak models, where each new model corrects errors made by the previous model by focusing on hard-to-classify instances. This iterative process emphasises complex cases, leading to a robust model. 3️⃣ Stacking🏋‍♂️: combines predictions from multiple diverse models (base models) by training a meta-model to learn the best way to integrate these predictions. Unlike bagging and boosting, which often use homogeneous models, stacking leverages the strengths of various algorithms. 😊 Key Differences ✔️ Bagging: Reduces variance by averaging predictions from multiple models trained on different data subsets. ✔️ Boosting: Reduces bias by focusing on errors of previous models, building a strong model sequentially. ✔️ Stacking: Combines diverse models by training a meta-model to optimise the final prediction. Understanding these techniques enables data scientists to build robust models that generalise well to unseen data. 👍 Like | 💬 Engage | 🔄 Repost | 🤝 Connect #DataScience #InterviewQuestionConceptSeries #MachineLearning #EnsembleLearning #Bagging #Boosting #Stacking

4

1 Comment

#geekstreak2024 #day11 #potd #python #gfg Edit Distance Given two strings str1 and str2. Return the minimum number of operations required to convert str1 to str2. The possible operations are permitted: 1. Insert a character at any position of the string. 2. Remove any character from the string. 3. Replace any character from the string with any other character. Input: str1 = "geek", srt2 = "gesek" Output: 1 Explanation: One operation is required, inserting 's' between two 'e'.

2

𝗘𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗼𝗳 𝗡atural 𝗟anguage 𝗣rocessing - 𝐏𝐚𝐫𝐭 2 Natural Language Processing (NLP) has revolutionized the way we interact with technology. From transformers , the journey of language models started in a new direction with attention mechanism and its achieved new height when it able to fine tune the model gpt-3 to get Instruct version or so called #chatGPT that complete strides in NLP have been nothing short of remarkable. Here's a glimpse into its transformative journey: 1️⃣ Transformers (2017) - The game-changer that set the foundation. 2️⃣ BERT (2018) - Enhanced contextual understanding. 3️⃣ ELMo (2018) - Word embeddings with deep context. 4️⃣ GPT (2018) - Generative pre-training begins. 5️⃣ GPT-2 (2019) - Scaling up language generation. 6️⃣ RoBERTa & ALBERT (2019) - Refining and optimizing BERT. 7️⃣ T5 & BART (2019) - Text-to-text transfer and improved summarization. 8️⃣ GPT-3 (2020) - Unleashing massive language models. 9️⃣ DeBERTa (2020) - Enhanced BERT with disentangled attention. 🔟 GPT-J (2021) - Open-source excellence. 1️⃣1️⃣ InstructGPT (2022) - Fine-tuning for better instructions. 1️⃣2️⃣ chatGPT (2022) - Conversational AI redefined. Each milestone brings us closer to more intuitive and human-like interactions with machines. Kudos to the brilliant minds driving this innovation forward! 🌟 #NLP #AI #MachineLearning #Innovation #Technology #Transformers #BERT #GPT #chatGPT #FutureOfAI 🏀 Keep following for more post on Data Science , NLP & Gen AI hashtag Feel free to share your thoughts on the evolution of NLP or your experiences with these technologies!

7

3 Comments

Decode the enchanting tales of machine learning from industries 🧠📚🏭 🌟LinkedIn's Budget Pacing for Targeted Ads youtu.be/R4EZ92VJvSI  🌟Netflix's Calibrated Recommendations youtu.be/DOWXNrBpO4w  🌟Netflix's Unified Recommendation ML Model: youtu.be/OKmv9sUrvk8 🌟Evolution of Recommendation Systems | Instagram Pinterest Twitter youtu.be/lgoyJn7MsH8 🌟Multi-Armed Bandit Strategies: youtu.be/2A5f3GrX0dA  🌟Overcoming Biases for a Better Recommender System | How Tech-Titans Combat Recommender System Biases youtu.be/oGb_mIdO0tA 🌟Building Scalable Query-Item Two-Tower Model based Retrieval System youtu.be/o-pZk5R0TZg  🌟Meituan's Game-Changing Two-Tower Retrieval Recsys Model | KDD'21 youtu.be/UhpbTSbi3lI  🌟 Innovative Linkedin's Deep-leaning based CTR Modeling: The Deep, Wide, and Shallow Towers Explained youtu.be/7l0HLYVFEuU 🌟 Lessons from Instagram’s Cutting-Edge Recommendation System youtu.be/Myna6rnmCG8 🌟 How Twitter recommends Tweets: Inside Twitter's Recommendation Algorithm 🐦 youtu.be/IhGq9jgcxFM 🌟 How eBay Built a Next-Generation Language Model to Revolutionise Online Shopping 🛒 youtu.be/h51nbWr7feo 🌟 Leveraging Pre-trained Bert to generate Entity Embeddings across domain 📚 youtu.be/v-0J7o-nDBE 🌟 Uplift Modeling to Detect Causal Effect | Uplift Modeling & Causal Inference to Improve Instagram ✨youtu.be/rKzG0Ct_ReA 🌟 Advertiser Recommendation Systems at Pinterest | Machine Learning & Algorithm-driven Ads Platform 📌youtu.be/KEiBJ0fal3Q 🌟 Dynamic Pricing Strategy to effectively fill slots | E-commerce | Travel | Hospitality Domains 💰youtu.be/a_CXpnsvPa0 🌟 Train your own Product Embeddings | Embeddings across Domains | Training on 1.02 Billion rows on GPU 💡youtu.be/DN4S96oHRhE 🌟 Model-Based Account Recommendations @ Twitter | Recommendation System | Retrieval & Ranking | RECSYS 📊 youtu.be/Xqo8fwgjxW4 🌟 Approximate Nearest Neighbour and a popular Library Annoy for k-Nearest Neighbour Search & Product Quantizer 🔍youtu.be/DSQOrBTqmYA & youtu.be/50PNumB7s3U 🌟 PID Controller to add a layer of diversity to the recommender systems 🎛️ youtu.be/laTxgnzjfR0 #datatrek #datascience #machinelearning #statistics #deeplearning #ai

13

A simple formula to calculate the GPU memory usage of hosting your LLMs 👇 #ai #ml #aiops #gpu #machinelearning #datascience #learning #learningeveryday #ram

494

17 Comments

A Step-by-Step Guide to Building a Machine Learning Model Building a machine learning model is more than just writing code – it’s an intricate, iterative process that requires technical skills, domain knowledge, and strategic thinking. Each step plays a crucial role in developing a model that performs well in the real world. Here’s the process broken down: - Business Understanding: Start by clearly defining the problem you're trying to solve and understanding the business context. Knowing your goals and objectives ensures that your model delivers actionable insights and value. - Data Collection: Gather your data from various sources, but remember – raw data is rarely perfect and often requires significant effort to get it in the right format. - Data Cleaning & Preprocessing: Handle missing values, outliers, normalize features, and encode categorical variables. This step is essential for ensuring clean, quality data that sets the foundation for reliable results. - Exploratory Data Analysis (EDA): Understand your data by visualizing distributions, correlations, and relationships. This helps uncover patterns and anomalies that can impact your model’s performance. - Feature Engineering & Selection: Create, combine, or transform features to capture important data patterns. This can significantly enhance your model's performance. - Model Selection & Training: Choose the right algorithm, train different models, and tune parameters. It's normal to revisit earlier steps – this is part of the iterative process. - Model Evaluation: Use metrics like accuracy, precision, recall, RMSE, etc. to measure performance. Be ready to tweak things to improve your results. - Deployment: Deploy your model into an application or make it accessible via an API. This is where your model starts delivering value. - Monitoring & Feedback: Continuously monitor how your model performs in the real world. Adjust for data drifts, user behavior changes, and gather feedback to keep it accurate and relevant. It’s important to note that this process is rarely linear. You’ll often find yourself going back and forth between these steps, refining and adjusting as new insights emerge or as your project goals evolve. Which part of this process do you find the most challenging? #PythonProgramming #MachineLearning #DataScience #AI #ModelDeployment #BusinessUnderstanding

12