نبذة عني
Lead Data Scientist with over 8 years of experience taking data projects from concept to…
الخبرة
-
-
-
-
-
-
-
-
-
-
-
التعليم
التراخيص والشهادات
ملفات شخصية أخرى مشابهة
استكشاف مزيد من المنشورات
-
Mohammed TAYEBI
Sampling is a crucial concept in inferential statistics, as it allows us to draw conclusions about an entire population based on a subset of data. Excited to have completed the 'Sampling in Python' course, which deepened my understanding of how to effectively use sampling techniques in data analysis!
-
Abed Khooli
Benchmarking Arabic Semantic Similarity Models Following the release of a few Arabic semantic similarity models (based on Sentence Transformers and ColBERT: Omartificial-Intelligence-Space/Arabert-all-nli-triplet-Matryoshka, AbderrahmanSkiredj1/Arabic_text_embedding_for_sts and akhooli/Arabic-ColBERT-100k), it is great to have robust benchmarking to compare performance. This involves selecting the right datasets (no leakage from training to evaluation) and running the scripts (and maybe publishing a leaderboard like the English MTEB for pure Arabic). This is probably a good idea for a researcher and a paper (I am not into either) but will indicate a relevant repo and some datasets to filter/clean/edit. I noticed that triplets used in the 100K ColBERT are poorly translated and focus on Western content, so there is room to work on better training and evaluation data for Arabic, but for a starter, even a decent ColBERT model is showing promising results (see previous post for code).
٨ تعليق -
Mostafa Zafer
Recently, a Gartner study estimated that 30% of Generative AI projects will be abandoned after proof of concept by the end of 2025. In the study, Gartner highlights 4 key possible causes for the 30% of projects that will be dropped out to be either: 1- Poor data quality 2- Inadequate risk controls 3- Escalating costs 4- Ambiguous business value As #generativeAI starts to rate lower in the emerging tech “Hype Cycle” – an expected and common phenomenon in the tech world, users and vendors are collectively reaching a realistic stage of the technology adoption lifecycle, where the excitement about potential takes a backseat, and realities of cost, difficulties of implementation and the pressure of achieving business value become more prominent. In my view, this percentage is a realistic estimate given where Generative AI is at the moment. Continuous experimentation and the search for innovation and exciting use cases should continue, but on the same footing, organizations need to realistically assess Gen AI proposed projects to improve the chances that these projects will make it to wide scale adoption as opposed to being dropped out post POC. Focusing on the cost challenge above, one of the reasons I see many Gen AI projects fail is the high cost of deploying Large Language Models (LLMs). As exciting as #LLMs are, the costs associated with running them is high including computational resources needed to run them in terms of storage, data processing capabilities and other operational costs. In many cases, Small Language Models (#SLMs) can be a more effective choice for Gen AI projects due to their efficient use of computing resources, the ability to scale them faster and cheaper as well as the lower cost of training these models as they focus on smaller sets of parameters and data sources.
٤ تعليق -
Tharun Sanjay
🚀 Exploring the Future of AI with Diffusion Models 🧠 In the world of AI and machine learning, Diffusion Models are making waves! 🌊 These cutting-edge models are revolutionizing how we approach data generation and image synthesis by offering new ways to solve complex problems. ✨ But what exactly are they? 🔍 #Diffusion_Models are probabilistic models that iteratively generate data by "diffusing" it from noise to a clear signal. Think of it like a process that starts with random noise and gradually refines it into a meaningful output, like an image, text, or even audio! 🎨📝🎧 Here are a few reasons why Diffusion Models are a game-changer: 🔑 #Superior_Performance: They can create high-quality images and other complex data that rival other generative models like GANs. 🧩 #Stable_Training: Diffusion Models offer better stability during training compared to alternatives, reducing some common challenges. 💡 #Flexible_Applications: From artificial creativity to biomedical research, the potential uses are limitless. 🖼️🔬 🌟 If you're curious about how Diffusion Models are shaping the future of AI innovation, now’s the time to dive in and explore their potential! 📚 Let’s connect and discuss how Diffusion Models could power the next generation of AI! 🔥 #Innovation #DiffusionModels #AIRevolution #DeepLearning #GenAI
١ تعليق واحد -
Kaushik Deb
2️⃣ 0️⃣ 2️⃣ 5️⃣ : A Year That’s Mathematically Magical 🪄 🧝♀️ 💥 🎉 🧨 ✨ As we are already in 2025, let’s take a moment to appreciate how this isn’t just another year—it’s a masterpiece of mathematics. A year so unique that it deserves a standing ovation from anyone who loves patterns, precision, or just really cool facts. Here’s what makes 2025 extraordinary: 1️⃣ A Perfect Square 2025 = 45² In a world where perfection feels rare, 2025 is literally a perfect square. It’s a reminder that with the right effort, you too can square your challenges and rise to the occasion. What’s next on the perfect square list? 2116 = 46² 2️⃣ Sum of the First 9 Cubes 2025 = 1³ + 2³ + … + 9³ Every cube from 1 to 9 adds up to 2025. Think of it as life’s way of saying that small, consistent steps lead to big results—whether in your career, personal growth, or even your gym routine (remember those resolutions?). What’s the next in this sequence? 3025, the sum of the first 10 cubes. 3️⃣ A Product of Two Squares 2025 = 9² × 5² This property shows how 2025 juggles multiple roles with elegance. It’s both a perfect square and a product of two smaller squares. A true multitasker, just like you managing deadlines, meetings, and last-minute client requests. Who’s next in line? 2116 = 2² × 23² 4️⃣ The Triple-Square Combo 2025 = 40² + 20² + 5² 2025 isn’t just a perfect square—it’s the sum of three squares too. It’s a triple win, much like when you balance strategy, execution, and team leadership all at once. The next in this special triple-square family? 2916 = 48² + 24² + 6². 5️⃣ A Harshad (Niven) Number What’s a Harshad number, you ask? It’s a number divisible by the sum of its digits. For 2025, 2+0+2+5=9, and 2025÷9=225. In simple terms, it plays well with others—something we all strive for at work. The next Harshad champ? 2028. Why Should You Care? Because numbers like 2025 don’t come along often—and neither do opportunities to reflect on the magic of patterns in life. This year is a perfect reminder that consistency, balance, and a little bit of math can create something truly extraordinary. So as we step into 2025, let’s embrace its energy: wrap up old cycles, take meaningful steps forward, and aim to be as impactful as this once-in-a-lifetime year. Here’s to a year of innovation, inspiration, and maybe a little more love for the numbers that shape our world. #2025 #PerfectSquare #Numerology #MathMagic #SumOfCubes #HarshadNumber #LinkedInInspiration
-
Ahmed Soliman
Happy for my dear students for their progress and their project. Starting with them from the basics of machine learning and mathematics until deep learning and how they can build a deep learning model then fine tuning LLMs. Thanks to the Ministry of Communications and Information Technology (MCIT), Egypt and Digital Egypt Pioneers Initiative - DEPI for letting me have the chance to teach this course and lead the students. Also, thanks to Hugging Face the platform that I really appreciate and go for many tasks and documentation for these lessons. #HuggingFace #DEPI #MachineLearning #DeepLearning
١ تعليق واحد -
Venugopal Adep
🔍 𝐃𝐞𝐞𝐩 𝐃𝐢𝐯𝐞 𝐢𝐧𝐭𝐨 𝐀𝐈 𝐈𝐧𝐧𝐨𝐯𝐚𝐭𝐢𝐨𝐧: 𝐋𝐥𝐚𝐦𝐚 3 𝐯𝐬. 𝐆𝐏𝐓-3.5 🔍 🦙 𝐔𝐧𝐫𝐚𝐯𝐞𝐥𝐢𝐧𝐠 𝐋𝐥𝐚𝐦𝐚 3: 🆕 Freshly launched by Meta, Llama 3 showcases significant upgrades in performance and efficiency. 🚀 Enhanced with a tokenizer of 128K tokens and Grouped Query Attention, enabling better data handling and longer context management. 🌐 Versatile applications in coding, reasoning, and multilingual tasks, supported by a diverse dataset of 15 trillion tokens. 🔒 Focus on robust privacy and security features, coupled with strong bias reduction strategies. 🧠 Fine-tuned with innovative techniques like supervised fine-tuning and preference optimization for nuanced response generation (AI Meta) (Unite.AI). 🤖 𝐆𝐏𝐓-3.5 𝐈𝐧𝐬𝐢𝐠𝐡𝐭𝐬: 🚀 Developed by OpenAI, known for powerful performance in language understanding. 🧠 Equipped with advanced fine-tuning capabilities, allowing high adaptability across varied tasks. 🛠️ Capable of generating text that closely mimics human-like discourse, making it ideal for creative and conversational applications. 📊 Tests across extensive real-world scenarios ensure reliability and practicality in everyday applications【 (AI Meta) (Unite.AI) 🔄 Leverage the potential of these groundbreaking AI models in your technological toolkit! 🛠️ 🔗 𝐅𝐨𝐥𝐥𝐨𝐰 𝐟𝐨𝐫 𝐦𝐨𝐫𝐞 𝐮𝐩𝐝𝐚𝐭𝐞𝐬 𝐨𝐧 𝐭𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐯𝐞 𝐀𝐈 𝐭𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐢𝐞𝐬! 🔗
-
Aisha Alabdullatif
🚀 Just wrapped up an insightful day at the Google Cloud Summit Saudi Arabia’s Innovators Hive! The hands-on labs and real-time challenges were exactly the kind of experience that keeps me motivated and engaged. From exploring GCP’s functionality to competing in labs on multimodal AI and BigQuery, each session sharpened my skills and expanded my problem-solving toolkit. It’s refreshing to see a summit that prioritizes real, actionable knowledge and empowers a whole community of tech enthusiasts. The thrill of diving into new tools, collaborating, and finding solutions on the spot is exactly why I love these events. A big thank you to Google Cloud for putting together a truly value-driven experience. Looking forward to more opportunities to learn, connect, and contribute! 🌐 #GoogleCloudSummit #InnovatorsHive #GCP #AI #BigQuery
١ تعليق واحد -
Karine Boucher
➡️ Saudi Arabia's $71B AI revolution begins with a Google-powered leap into the future ⬅️ Saudi Arabia’s Public Investment Fund (PIF) has partnered with Google Cloud to create an advanced AI hub, projected to generate thousands of jobs and contribute $71 billion to the country's GDP over the next eight years. This agreement, signed at the Future Investment Initiative in Riyadh, aligns with Saudi Arabia's goal of becoming a regional tech hub by 2030. The partnership will focus on providing AI training for millions of Saudi students and professionals, aiming to expand the information and communication technology sector by 50%. It includes joint research on Arabic language models and AI applications specific to Saudi Arabia. Key benefits of this collaboration include enhanced access to AI applications across various industries such as healthcare, retail, and financial services. Additionally, PIF and Google Cloud plan to improve the Arabic-language capabilities of Google's Gemini AI model by integrating more Arabic datasets, facilitating the development of AI solutions tailored for Arabic-speaking users. https://lnkd.in/ghPQudJn #SaudiAI #GoogleCloud #TechInnovation #SaudiVision2030 #MiddleEastTech #AIinvestment #DigitalTransformation #ArabicAI #PIF #TechHub #SaudiTech #MENA #ArtificialIntelligence #SaudiEconomy
١ تعليق واحد -
Luis De Mendoza
#AI is everywhere so it should come as no surprise that the US military is using #artificialintelligence . DeepLearning.ai states that the US military has 800 AI projects currently in development. One of those projects is named #Maven and it was recently used in the Middle East to identify targets in Yemen and the Red Sea. Unfortunately the accuracy of Maven seems to be around 60% which is less than human analysts which were accurate about 84% of the time. What do you all think? Should we require a human in the loop to make decisions about when to shoot missiles? Should we even be using AI in war?
٢ تعليق -
Vandana R.
Neuroscientist says holding a thought for just 17 seconds, will start to manifest !! #How_it_works: When you hold a thought for 17 seconds, you emit a signal that attracts similar energy. Extend that focus to 68 seconds, and you generate a momentum that can influence your emotions, behaviors, and outcomes. #Example: • Positive Focus: When you hold thoughts of gratitude for 17 seconds, more positive and uplifting thoughts naturally follow. This elevates your mood and attracts opportunities aligned with your energy. • Negative Focus: On the other hand, dwelling on frustration or anger for just 17 seconds can invite more situations that reinforce those emotions, perpetuating a negative cycle. #Try_it_today: Your thoughts hold immense power. Intentionally focus on what brings you joy, excitement, or gratitude, for just 17 seconds, you can shift your energy and start manifesting positive experiences. #Mindset #Manifestation #PositiveEnergy #Neuroscience #Positivethought #Negativethought
-
Abdou Kane, PhD
Bonjour 👋, Data scientist and data engineer fellows, have you heard about #GenAI? #LLMs? #Transformers? It seems to be the next big thing in the data space and I've been receiving more and more inquiries about it. Unfortunately you won’t be able to train such models unless your company has $20M to spent on it. The good news is you don’t have to, to create powerful applications such as #RAG based ones. As a humble start, I will be sharing a couple of tips and sources to get started with this space and tools to get practical ideas and pathways for upskilling. Starting with the article from Google released in 2017, that opened all the great things we're seeing in this space. #OpenAI, #HuggingFace, #MistralAI, #Anthropic... I don’t wanna diverge here by talking about #Langchain (a must have library) My very first recommendation is to spend a month reading this article - "Attention is all you need." Make sure you deeply understand what a Transformer (the T in GPT) is and what the attention mechanism is. #generativeai #neuralnetwork #transformer #datascience
٤ تعليق -
Mohamad Zeinni
🌟 Excited to Share My Latest Project on Enhancing CNN Model Performance! 🌟 Over the past few months, I've been working on a fascinating project aimed at enhancing the performance of AI models for grocery classification by using the Convolutional Neural Networks (CNNs) . 🚀 Optimization Strategies: -Baseline Convolution Model with Dropout Layer: Implemented dropout layers to prevent overfitting. -VGG16 Transfer Model with Dropout Layer: Leveraged transfer learning for enhanced accuracy. -VGG16 Transfer Model without Dropout Layer: Explored the impact of dropout layers on performance. -VGG16 Transfer Model with Adjusted Learning Rate: Tuned learning rates for optimal training. -ResNet50 Transfer Model: This model emerged as the best performer with an accuracy of 98-99%. 🔍 Conclusion: The ResNet50 model demonstrated superior performance in terms of both accuracy and loss, proving to be highly effective for high-level grocery classification tasks. The innovative use of multiple convolutional neural networks, including the ResNet50's network-in-network architecture, has been pivotal. The global average pooling feature reduced the model size, enhancing efficiency and performance. I'm thrilled with the progress and can't wait to see the further advancements in this field. A big shoutout to Walid Semaan, @Ali H.Zayter who beleived with the sturdent of cohort 1 LAU - Academy of Continuing Education (ACE) Feel free to reach out if you’re interested in learning more about this project or if you have any insights to share. Let’s connect and innovate together! #AI #MachineLearning #DataScience #DeepLearning #ArtificialIntelligence #Innovation #GroceryClassification #ResNet50 #VGG16 #ModelOptimization
٥ تعليق -
Ahmed Hisham
It was one of the best short courses ever to enhance the LLMs memory by using the concept of MemGPT to keep memorizing the context and chat history , especially using the agentic memory and exploring different memory types such as the archival and context memory #GenAI #LLMs #AI #Agents #Datascience #MemGPT
-
Iana Iatsun
Trees in #AI 🌳 In machine learning we love represent decision-making algorithms as a tree. I summarize the basics of the most used ones below : 🍀 Expert system is a decision-making tool that utilizes human expertise and input data. An expert system relies on having a good knowledge base. If we organize the knowledge base in a tree, experts define the nodes, its outputs and associated thresholds. Expert system is explainable and trustworthy. 🍀 Decision trees is a machine learning algorithm. They seek to find the best split to divide the data. Decision tree learning employs a divide and conquer strategy by conducting a greedy search to identify the optimal split points within a tree. This process of splitting is then repeated in a top-down, recursive manner until all, have been classified. Each node helps an individual to arrive at a final decision, which would be denoted by the leaf node. Data is used to determine the nodes. The decision trees have preference for small trees, but they are prone for overfitting and costly due to the greedy search. 🍀 Random forest is an uncorrelated forest of decision trees using bagging and feature randomness. Three main hyperparameters to set before training: include node size, the number of trees, and the number of features sampled. About what other things would you like to know ?