Mohammad Abir Abbas

"This is a collection of simple PyTorch implementations of neural networks and related algorithms. These implementations are documented with explanations, and the website renders these as side-by-side formatted notes. We believe these would help you understand these algorithms better." Theory and code for many leading ML papers -- that's all you can ask for! Paper implementation with code link: https://nn.labml.ai/

৯

Daily Professional Journaling 📕 Day 86- April 16, 2024 Upskilling Tasks: 1. Solve one Schrodinger case by GGI 2. Solve SQL tasks on Hackerrank Personal Tasks: 1. Read the fourth chapter of the book- Thinking Fast & Slow by Daniel Kahneman CAT Journey- Tasks of the Day: 1. Concept building on QA- Triangles cont. 2. Practice questions on QA- P&C 3. Practice questions on VARC- Odd Sentences 4. Practice solving DILR sets 5. Read one Aeon essay Yesterday's Update: Tasks Accomplished: 1. Solved one Schrodinger case by GGI 2. Solved SQL tasks on Hackerrank 3. Read the third chapter of the book- Thinking Fast & Slow by Daniel Kahneman 4. Concept building on QA- Triangles 5. Practiced questions on QA- P&C 6. Practiced questions on VARC- Para Jumbles 7. Practiced solving DILR sets 8. Read one Aeon essay (PS: A little initiative towards staying consistent on this upskilling journey.) #dontwaitstarttoday #dailyprofessionaljournaling #upskillingjourney

৫

Daily Professional Journaling 📕 Day 88- April 18, 2024 Upskilling Tasks: 1. Solve one Schrodinger case by GGI 2. Solve SQL tasks on Hackerrank Personal Tasks: 1. Read the sixth chapter of the book- Thinking Fast & Slow by Daniel Kahneman CAT Journey- Tasks of the Day: 1. Concept building on QA- Triangles cont. 2. Practice questions on QA- P&C 3. Practice questions on VARC- Odd Sentences 4. Practice solving DILR sets 5. Read one Aeon essay Yesterday's Update: Tasks Accomplished: 1. Solved one Schrodinger case by GGI 2. Read the fifth chapter of the book- Thinking Fast & Slow by Daniel Kahneman 3. Concept building on QA- Triangles 4. Practiced solving DILR sets 5. Read one Aeon essay (PS: A little initiative towards staying consistent on this upskilling journey.) #dontwaitstarttoday #dailyprofessionaljournaling #upskillingjourney

৭

Let's go multimodal with Prompt Engineering. There are many techniques that have been developed to get the best possible output from LLMs. They are applicable to MM-LLMs too. But, there are certain techniques that are specific to multimodal LLMs. Day 61 of #100DaysofAI - Multimodal Chain-of-Thought prompting. The first technique we see is Multimodal-CoT(Chain-of-Thought) prompting. For LLMs, as simple as saying "Let's think step by step" helps reach correct answers. In spite of this, some times there could be hallucinations in the reasoning process which leads to wrong answers. In Multimodal CoT, there are two steps. 1️⃣ Rationale Generation - Prompt the model with both the question and the image to generate a reasoning(rationale) as shown in the picture. 2️⃣ Answer Inference - After getting the rationale, the model is prompted with the question, the rationale and the image to get the final answer. This works much better than only prompting the model with the question and reasoning. This is because even if there is a hallucination in the reasoning, the image acts as additional reference during inference. Learn more here: https://lnkd.in/e9nr5edT On Day 62 of #100DaysofAI, we will discuss Visual Referring prompting. Follow me for more such updates on AI. Would you like these short bit sized learnings of AI topics in your email? Let me know in the comments below. #ai #tech #generativeai #datascience #deeplearning #machinelearning #data #computervision #multimodal #promptengineering

১৫

Large Language Models (LLMs) Explained for Beginners: A Friendly Guide 🤔 What's a Large Language Model? Let's Use a Real-Life Analogy! Imagine LLMs like an incredibly well-read librarian who has: - Read millions of books - Memorized vast amounts of information - Can understand and explain almost anything - Can help you with diverse tasks like writing, answering questions, and solving problems Language Model vs Large Language Model: The Difference Traditional Language Model (Small Library) - Like a local library with limited books - Can help with basic questions - Limited understanding - Needs specific instructions for each task Large Language Model (Massive Mega-Library) - Like a global super-library with millions of books - Can understand context and nuance - Learns and adapts quickly - Handles complex tasks with minimal guidance 🧠 How Do LLMs Actually Work? Think of an LLM like a super-smart prediction machine: 1. Learning Phase   - Imagine reading millions of books, articles, and websites   - Absorbing patterns, language structure, and knowledge   - Learning how words and ideas connect 2. Prediction Magic   - When you ask a question, it predicts the most likely, intelligent response   - Uses its vast "reading experience" to craft answers   - Similar to how you might complete a sentence based on context Real-World Example - You type: "What's the capital of..." - LLM quickly predicts and completes: "...France? It's Paris!" - It doesn't just search, it understands and generates 🚀 Three Ways to Use LLMs 1. Simple Prompting - Like asking a question to a smart friend - No technical skills needed - Example: "Help me write an email" 2. Model Fine-Tuning - Customize the LLM for specific needs - Like training a general chef to specialize in Italian cuisine - Requires some technical knowledge - Example: Training for medical or legal language 3. Building Custom LLMs - Creating your own language model from scratch - Highest complexity - Needs deep machine learning expertise - Like building an entire library instead of borrowing books 🌐 Key Players in LLM World OpenAI - Think: Apple of AI world - Closed, commercial models - Created famous ChatGPT - Known for cutting-edge technology Hugging Face - Think: GitHub for AI models - Open-source platform - Community-driven - Allows sharing and collaborating on AI models Who Should Care? - Students learning technology - Professionals in tech, writing, research - Entrepreneurs exploring AI solutions - Anyone curious about future technology **Pro Tip**: Learning about LLMs is like learning a new language - start simple, be patient, and enjoy the journey! #AIForBeginners #MachineLearning #LLMExplained #ArtificialIntelligence #TechLearning #AITechnology #CareerInTech #InnovationSkills #GenerativeAI #TechEducation

৬

Hallucinations in #LLMs is very common and dangerous process. There are few ways to deal with hallucinations in LLMs. One powerful technique to avoid hallucinations is called NLI (Natural Language Inference). Read ⬇ In simple words #NLI uses #premise (given condition) and #hypothesis (generated condition or condition before premise) pair of sentences to find the fact and avoid hallucinations. Consider below conversation between user and LLM generated statement. User: All dogs are animals To this reply LLM might try to generate one of following output, these are inferences (hypothesis'), let us mark them with numbers for easier understanding. LLM -- [1- {Some dogs are animals}, 2- {No dogs are animals} ] It is passed on to hallucinations detector to filter out wrong statements. The detector uses NLI to detect problems in generated statements. NLI uses techniques like entailment, contradiction to select correct statement. In Entailment - The hypothesis logically follows from the premise. If the premise is true, the hypothesis must also be true. With this logic statement 1({Some dogs are animals}) is true. In contradiction - The hypothesis logically contradicts the premise. If the premise is true, the hypothesis must be false. So by this logic second statement is false because premise is true. With this technique, the system then select statement one for final output. It is important to have hallucinations detector as first component after llm output so you pass on only correct data to your system from LLMs. I hope this helps.

১২

It's Day 29 and today, I'll be posting about Overview of Seq2Seq model in Natural Language Processing. Seq2Seq, short for Sequence-to-Sequence, is a type of model in Natural Language Processing (NLP) used for transforming sequences from one domain to sequences in another domain. This architecture is particularly powerful and widely used in applications such as machine translation, text summarization, and chatbot responses. Here's a detailed breakdown of the Seq2Seq model: Core Components of Seq2Seq: 1. Encoder:  - The encoder processes the input sequence and compresses the information into a fixed-size context vector. It is typically implemented using Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), or Gated Recurrent Units (GRUs). - The encoder reads the input sequence one token at a time and updates its hidden state. The final hidden state of the encoder represents the entire input sequence. 2. Decoder: - The decoder generates the output sequence token by token. It is also usually an RNN, LSTM, or GRU. - At each time step, the decoder takes the previous token generated and the current hidden state as inputs to produce the next token. 3. Context Vector: - The context vector (also known as the thought vector) is the final hidden state of the encoder. It captures the summary of the input sequence and is passed to the decoder to generate the output sequence.    Sequence of Operations: 1. Encoding Phase: - The input sequence is fed into the encoder. - The encoder processes the sequence and produces the context vector. 2. Decoding Phase: - The decoder takes the context vector as its initial hidden state. - The decoder generates the output sequence one token at a time until a stopping criterion is met (e.g., generating an end-of-sequence token). Applications 1. Translating sentences from one language to another. 2. Condensing a long document into a shorter version while preserving the main information. 3. Generating responses in a conversation based on the previous dialogue history.  4. Converting spoken language into text. For Deeper Understanding: https://lnkd.in/gamAsNeY ​Image Source: Google Search Cultivate your knowledge journey with Siddhant Srivastava for an enriching stream of related content. Follow for insights that resonate and elevate your understanding.

৩০

Must read for LLM users and newcomers A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications What you will learn - 1. Core Concept: Prompt engineering is used to guide LLM outputs for specific tasks without modifying model parameters, making models more versatile and efficient. This approach facilitates using pre-trained models on various tasks by designing effective prompts. Prompt engineering has evolved from basic zero-shot and few-shot techniques to advanced strategies like chain-of-thought (CoT) prompting and automatic prompt generation. 2. Categories and Techniques The paper systematically classifies 29 prompt engineering techniques into application-specific categories, such as task complexity, reasoning, user interaction, hallucination reduction, code generation, and optimization. New Tasks Without Extensive Training: Zero-shot Prompting: Allows models to perform new tasks without training data by providing task descriptions alone. Few-shot Prompting: Offers models a few examples, enhancing accuracy but requiring extra token space. Reasoning and Logic Enhancements: Chain-of-Thought (CoT) Prompting: Improves reasoning by breaking tasks into sequential logical steps. Automatic Chain-of-Thought (Auto-CoT): Automates the CoT process by generating reasoning chains, reducing manual intervention. Self-Consistency: Increases accuracy by generating multiple reasoning chains and selecting the most consistent response. Tree-of-Thought (ToT) and Graph-of-Thought (GoT): Extend reasoning through a tree or graph structure, allowing exploration and backtracking for complex tasks. Reducing Hallucinations: Retrieval-Augmented Generation (RAG): Enhances factual accuracy by integrating retrieved information into responses. ReAct Prompting: Combines reasoning and action generation for dynamic response adjustments. Chain-of-Verification (CoVe): Employs a verification step to check the validity of model outputs, reducing errors. Improving Consistency and Coherence: Contrastive CoT Prompting (CCoT): Includes both valid and invalid examples to help the model learn more robustly by contrasting solutions. Specialized Applications: Emotion Prompting: Introduces emotional cues to enhance model understanding of psychological factors, boosting response relatability. Scratchpad and Program of Thoughts (PoT) Prompting: Enhances code generation by structuring outputs as computational steps. Optimization and Efficiency: Optimization by Prompting (OPRO): Uses prompts for optimization tasks like linear regression, illustrating LLM potential in numerical applications. 3. Future Directions and Challenges Research Challenges: Addressing biases, mitigating hallucinations, and enhancing interpretability remain significant challenges. There is a need for robust prompt designs that minimize bias and ensure reliable responses.

৪৯

১ মন্তব্য

👉Ask ChatGPT<<ML ed>>: List a summary of the NLP field - page 1 of 2 ChatGPT 🤖 ### Natural Language Processing (NLP) Summary **Natural Language Processing (NLP)** is a subfield of artificial intelligence (AI) that focuses on the interaction between computers and humans through natural language. The goal of NLP is to enable computers to understand, interpret, generate, and respond to human language in a way that is both meaningful and useful. NLP is used in a wide range of applications, from chatbots and virtual assistants to translation services and sentiment analysis. #### Key Areas in NLP 1. **Text Processing and Understanding:**   - **Tokenization:** Breaking down text into smaller units like words or sentences.   - **Part-of-Speech Tagging:** Identifying the grammatical category (noun, verb, etc.) of each word.   - **Named Entity Recognition (NER):** Detecting and classifying entities (like names, dates, or locations) in the text. 2. **Language Modeling:**   - Predicts the likelihood of a sequence of words and is crucial for tasks like text generation, speech recognition, and machine translation.   - **Transformers:** Advanced deep learning architectures like GPT (Generative Pre-trained Transformer) and BERT (Bidirectional Encoder Representations from Transformers) are widely used. 3. **Sentiment Analysis:**   - Determines the sentiment or emotional tone behind a piece of text (positive, negative, neutral).   - Commonly used in customer feedback analysis, social media monitoring, and market research. 4. **Machine Translation:**   - Automatically translates text or speech from one language to another.   - Utilizes models such as neural machine translation (NMT), which has significantly improved the quality of translations. 5. **Speech Recognition and Synthesis:**   - **Speech Recognition:** Converts spoken language into text (e.g., voice typing, virtual assistants).   - **Speech Synthesis (Text-to-Speech):** Converts text into spoken language, enabling human-like responses. 6. **Question Answering and Information Retrieval:**   - Extracts relevant information from large datasets or documents in response to user queries.   - Used in search engines, digital assistants, and customer service bots. 7. **Chatbots and Conversational AI:**   - Programs that simulate human conversation through text or voice.   - Used for customer support, e-commerce, and interactive applications. ---continues---

১

১ মন্তব্য

🚨 Blog alert: DSPy 🚨 The current framework of working with LLMs don't promote best practices. DSPy helps you with modularity and evaluation, the corner stones of great machine learning outputs. I have previously released a video about this as well, you can find it in there as well as a link to the GitHub repo with all of the code.

১১

৬ মন্তব্যসমূহ

👉Ask ChatGPT<<ML ed>>: List a summary of the NLP field - page 2 of 2 ChatGPT 🤖 ### Natural Language Processing (NLP) Summary **Natural Language Processing (NLP)** is a subfield of artificial intelligence (AI) that focuses on the interaction between computers and humans through natural language. The goal of NLP is to enable computers to understand, interpret, generate, and respond to human language in a way that is both meaningful and useful. NLP is used in a wide range of applications, from chatbots and virtual assistants to translation services and sentiment analysis. ---continued--- #### Key Techniques in NLP😍😍 - **Machine Learning:** Supervised and unsupervised learning algorithms like 👉logistic regression, 👉support vector machines (SVMs), and 👉neural networks. - **Deep Learning:** Techniques using architectures like recurrent neural networks (RNNs), 👉long short-term memory networks (LSTMs), and transformers. - **Linguistic Rules:** Handcrafted rules based on 👉syntax and semantics, often combined with machine learning for improved results. #### 👉👉Key Skills for NLP Jobs 1. **Linguistics:** Understanding of syntax, semantics, and phonetics. 2. **Programming:** Proficiency in Python, Java, or similar languages, and experience with 👉NLP libraries like NLTK, spaCy, or Hugging Face Transformers. 3. **Data Handling:** Skills in data preprocessing, cleaning, and transformation. 4. **Algorithm Knowledge:** Familiarity with 👉 machine learning and deep learning algorithms. 5. **Domain Expertise:** Ability to tailor NLP models to 👉specific industries or use cases (e.g., healthcare, finance).

👨‍💻 For GenAI Engineers: Key Points to Keep in Mind 👩‍💻 #Day10: Common Challenges with LLM & Future of LLMs. Welcome back to the #GenerativeAI series! Today, we will tackle two important topics: the common challenges faced when working with Large Language Models (LLMs) and the exciting future of this technology. Let’s dive in! 🚀 🤔 𝐂𝐨𝐦𝐦𝐨𝐧 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞𝐬 𝐰𝐢𝐭𝐡 𝐋𝐋𝐌𝐬 Q1: What are some common challenges with LLMs? 📍 Working with LLMs can be powerful but challenging. Key issues include: 🔵 High Computational Costs: Training and fine-tuning LLMs require significant computational resources, making them expensive to run. 🔵 Bias and Fairness: LLMs can inherit biases from training data, leading to biased or unfair outputs. 🔵 Data Privacy: Handling large datasets involves sensitive information, raising privacy concerns. 🔵 Interpretability: LLMs are often seen as "black boxes," making it difficult to understand their decision-making process. 🌟 𝐇𝐨𝐰 𝐭𝐨 𝐀𝐝𝐝𝐫𝐞𝐬𝐬 𝐓𝐡𝐞𝐬𝐞 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞𝐬 Q2: How can we address these challenges? 🔵 Optimizing Resources: Use model distillation to create smaller, efficient versions of large models. 🔵 Bias Mitigation: Implement techniques like adversarial debiasing and use diverse training data. 🔵 Data Privacy Measures: Employ differential privacy techniques to ensure data remains anonymous and secure. 🔵 Improving Interpretability: Use explainable AI techniques to make LLMs more transparent. 🤔 𝐓𝐡𝐞 𝐅𝐮𝐭𝐮𝐫𝐞 𝐨𝐟 𝐋𝐋𝐌𝐬 Q3: Where is the technology headed? 📍 The future of LLMs is promising with several key trends: 🔵 Smaller, More Efficient Models: Development of models that maintain performance while being less resource-intensive. 🔵 Multimodal Models: Combining text, image, and audio processing into a single model for richer applications. 🔵 Enhanced Collaboration: Improved collaboration between humans and AI, assisting in complex decision-making and creativity. 🔵 Ethical and Responsible AI: Greater focus on developing ethical AI practices and ensuring responsible use of LLMs. 🗣️ Join the Conversation! Q4: What challenges have you faced with LLMs, and how do you see their future? 📌 Share your thoughts in the comments! 📅 Stay Tuned Stay tuned for more learning and growing together! 🌟 #MachineLearning #DeepLearning #AI #Tech #LLM #PythonDev #Challenges #Future #Python

৭

🚀 Exciting News! 🚀 I am thrilled to share that I have successfully implemented and trained Generative Adversarial Network (CCGAN - Context Conditional GAN) as part of my latest implementation for learning purpose! 🎉 Throughout this project, I leveraged a suite of powerful tools and technologies to ensure robust model development and deployment. Here’s a breakdown of what I used: 🔧 **Tools & Technologies**: - **DVC**: For efficient data version control, ensuring all data and model checkpoints are tracked and reproducible. - **MLflow**: To track and log experiments, providing a comprehensive view of the model training process. - **MongoDB Atlas**: As the database solution for storing and managing the dataset. - **Docker**: To containerize the application, ensuring consistency across different environments. - **Python**: For implementing the CCGAN model and running training scripts. One of the key milestones in this project was effectively monitoring the training process using various metrics, including the training loss, model registry as well. Visualizing these metrics was crucial for understanding the model's performance and making necessary adjustments. A huge shoutout to the amazing tools and frameworks that made this possible, and to the continuous learning and support from the community. Looking forward to exploring more in the field of generative models and AI! 💡 link of the code: https://lnkd.in/esetQi66 #MachineLearning #DeepLearning #AI #GenerativeAdversarialNetwork #GAN #DataScience #ArtificialIntelligence #ModelTraining #TechInnovation #DVC #MLflow #MongoDB #Docker #Python

২

Introducing AppAgent 📱—a multimodal agent framework using Large Language Models (LLMs) to interact with smartphone applications through standard touch actions like tapping and swiping, without requiring system-level access. Key Features: 1. Human-like Interactions 👤: Navigates apps with actions just like a regular user. 2. Learning Mechanisms 📚: Builds a knowledge base by exploring or observing user interactions. 3. Versatile Use 🔄: Handles diverse tasks across apps, from social media to image editing. AppAgent’s approach, combining exploration and deployment, represents a structured advancement in AI-driven mobile applications. Learn more here: https://lnkd.in/dCqDWSiz #AI #AppAgent #MachineLearning #LLM #MobileTechnology

১০

৩ মন্তব্যসমূহ

5 of the most popular NLP Applications (part1): Natural Language Processing (NLP) is a field within Artificial Intelligence (AI) that focuses on enabling machines to understand, interpret, and generate human language. From analyzing social media sentiments to powering virtual assistants like Siri and Alexa, NLP plays a crucial role in how we interact with technology. You can think about NLP as a bridge between computers and human language! Some of the most popular applications in the NLP field are: 1. Text Classification: Categorizing Information It involves assigning predefined categories to text. Examples of Text Classification are: Email spam detection and Topic classification (tagging articles as sports, politics, or entertainment). These Applications are done using models like Logistic Regression (the traditional way), or by using Transformers models like BERT. 2. Named Entity Recognition (NER): Finding Key Information It identifies and labels specific entities within a text, such as names, dates, organizations, and locations. Examples of NER are: Extracting customers names and complaint details from support tickets, and Identifying medical terms in healthcare documents. To achive such tasks, we can use CRFs (Conditional Random Fields) for simpler tasks, or Fine-tuned transformers like spaCy's NER pipeline and Hugging Face models for more complicated tasks! 3. Sentiment Analysis: Understanding Emotions in Text This application focuses on determining the tone or emotion behind a piece of text. Examples of this applications is analyzing customer feedback to improve services, and maybe monitoring social media comments and hide offensive comments. Models used here are: LSTMs (Long Short-Term Memory networks) for sequence-based analysis, and Transformers models for deeper contextual understanding. 4. Question Answering (QA): Giving Precise Answers As the name suggests, QA systems extract or generate answers to user queries based on a dataset or context. Examples are simple chatbots that are used in customer service. Models for this application can be as smaller as BERT, or as bigger as Large Language Models (such as GPTs!) 5. Machine Translation: Breaking Language Barriers This application enables automatic translation between languages. Examples are for sure Google Translate, and even translating websites for global audiences. You can do so using Sequence-to-sequence models with attention mechanisms (e.g., Transformer), and Multilingual LLMs like mBERT or M2M-100. follow Saad Shahrour and Raizer If you want to learn more about AI and NLP!

১৮

৫ মন্তব্যসমূহ

🚨 New video alert! DSPy: machine learning attitude towards prompting🚨 LLMs have fatigued me, it's so easy to create a good demo and so hard for evaluation and productanization. That's a poisonous mix for a data-scientist. DSPy can help you elevate this pain, by treating the generative stack as if it was a neural network, with its own parameters. Composing different patterns one on top of another, evaluating, optimizing. We don't have to reinvent the wheel each time, rephrasing again and again CoT pattern, Karpathy prayer's, few shot examples and so forth. In the video you can watch a tutorial that is not too simple and not too complex that can simulate how to use this framework, by implementing a customer service LLM based intent classifier. Personal note - This is the first video that wasn't born out of my streaming experience. Meaning it is more polished than my previous videos. Technical note - DSPy is great

১৮

১ মন্তব্য

📖 My reading/studying list for July... 📱🧠 "Building LLM Powered Applications" by Valentina Alto A comprehensive guide to developing applications with Large Language Models (LLMs). The book covers theoretical foundations, hands-on implementation, and practical applications of LLMs. It explores popular LLM architectures, training methods, and real-world use cases. Additionally, it addresses emerging trends and responsible AI practices to mitigate risks associated with AI tools. 📊 🦙 "Building Data-Driven Applications with LlamaIndex" by Andrei Gheorghiu Is a practical guide for using Retrieval-Augmented Generation (RAG) to enhance Large Language Model (LLM) applications. The book covers foundational concepts of generative AI and LlamaIndex, practical examples of data ingestion, indexing, and querying, and building end-user applications. It also delves into customizing and deploying LlamaIndex projects, focusing on overcoming LLM limitations and best practices in prompt engineering. 📊 🔗 "Graph Data Modeling in Python" by Gary Hutson and Matt Jackson A practical guide for using graph data structures in Python. It covers the fundamentals of graph theory, comparing relational databases with graph databases, and real-world applications in various industries. The book introduces popular graph packages like NetworkX and igraph, demonstrating how to curate, analyze, and model data using graphs. 📊 🔗 "Building Knowledge Graphs: A Practitioner’s Guide" by Jesús Barrasa and Jim Webber An essential resource for developers and data scientists aiming to master knowledge graph technologies. This guide, a free e-book edition by Neo4j and O'Reilly, covers foundational concepts to advanced techniques in building, managing, and utilizing knowledge graphs. It provides practical insights and technical skills necessary for leveraging this transformative technology in real-world projects.

৩৭

৪ মন্তব্যসমূহ