Asher Fredman

🎥 Luma Labs unveils Dream Machine! #Luma_Labs has just launched their revolutionary new #AI model, #Dream_Machine, which can generate high-quality, realistic 5-second video clips from text and image prompts — and it's already available to the public. 🔍💡 Key Highlights of Dream Machine: 🔹 Smooth, Consistent Videos: The model generates 5-second clips with impressive smoothness, consistency, and camera movements, creating coherent outputs that mimic real-world interactions and physics. 🔹 Advanced Transformer Model: Dream Machine is trained on video content, allowing for more realistic video generations that better capture the essence of the prompts provided. 🔹 Quick Production: Capable of producing 120 frames of video in 120 seconds, though current wait times might be longer due to high demand at launch. 🔹 Accessible Plans: Launched with a free plan offering 30 video generations, with a paid tier extending to 2,000 outputs per month, making advanced video generation accessible to everyone. 🌟 Why It Matters: As the public eagerly awaits the launch of #OpenAI’s #Sora, models like #KLING and Dream Machine are already offering powerful new capabilities, pushing the boundaries of AI video generation. Each new competitor raises the bar, setting higher standards for future models. 🔗 Check out some of the best generations from Luma’s launch here: https://lnkd.in/gc94-d4d 🔗 To Know More About Dream Machine: https://lnkd.in/g9Xs5kjq 📢 Join the Conversation! What are your thoughts on these exciting new capabilities from Luma Labs? 🤝 Connect with us at ShellKode for more updates and discussions on the future of AI technology. #LumaLabs #DreamMachine #VideoGeneration #TechInnovation #AIRevolution #GenerativeAI #TechUpdates #InnovationInTech #AICommunity

6

The world's most advanced processors, including Intel(Intel 7), Apple(M1, M2, M3, M4 etc chips) and Nvidia(Blackwell) were researched and developed in Israel. The list goes on. Apple's 2nd largest R&D, Nvidia's 2nd largest R&D, Intel's 2nd largest R&D all in Israel. If there was no Israel today and the people inside, the advancement of software and hardware as it is today(GPUs, AI, CPUs, humanoid robots etc) would cease to exist. Not sure what it's going to take for the world to recognize Israel as a force for good, an ally, and get behind backing Israel and Israelis. Let these people get back to work, their government defend themselves and there will be a brighter future for all of us. Enough is enough.

12

Excited to share my recent interview on "הכל כלכלה" discussing Nvidia's acquisition of Run:ai, an Israeli software company, for a whopping $700 million! This move, along with potential future acquisitions, demonstrates Nvidia's strong belief in Israeli AI innovation. Run:AI's software expertise will enhance Nvidia's hardware offerings, making AI more accessible and efficient. As the global AI race intensifies, Israel's talent and technology are crucial. With strategic partnerships and continued investment, we can ensure Israel remains a leader in shaping the future of AI.

28

2 תגובות

Interesting panel discussion at Google and Reichman Tech School about #ai and how to enter the field. Barr Morgenstein Michael(Mike) Erlihson Dr. Barak Or Hila Noga Evgeny Artyomov #deeplearning #artificialintelligence #machinelearning #llm #generativeai #genai

65

3 תגובות

🚀 𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝗟𝗹𝗮𝗺𝗮 𝟯.𝟮 𝗳𝗿𝗼𝗺 𝗠𝗲𝘁𝗮: 𝗧𝗵𝗲 𝗙𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗲𝗱 𝗔𝗜 𝗼𝗻 𝗘𝗱𝗴𝗲 𝗗𝗲𝘃𝗶𝗰𝗲𝘀! 🌐 Meta's 𝗟𝗹𝗮𝗺𝗮 𝟯.𝟮 represents the next leap in AI technology—a new generation of lightweight vision and text models designed to fit on edge devices, offering more 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗲𝗱 𝗮𝗻𝗱 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗔𝗜 𝗲𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲𝘀. With 𝘀𝗺𝗮𝗹𝗹 𝘁𝗼 𝗺𝗲𝗱𝗶𝘂𝗺-𝘀𝗶𝘇𝗲𝗱 𝘃𝗶𝘀𝗶𝗼𝗻 𝗟𝗟𝗠𝘀 (𝟭𝟭𝗕 𝗮𝗻𝗱 𝟵𝟬𝗕) and 𝗹𝗶𝗴𝗵𝘁𝘄𝗲𝗶𝗴𝗵𝘁 𝘁𝗲𝘅𝘁-𝗼𝗻𝗹𝘆 𝗺𝗼𝗱𝗲𝗹𝘀 (𝟭𝗕 𝗮𝗻𝗱 𝟯𝗕), Llama 3.2 excels at on-device processing with a focus on 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗹𝗲 𝗶𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻 and 𝘀𝘆𝘀𝘁𝗲𝗺-𝗹𝗲𝘃𝗲𝗹 𝘀𝗮𝗳𝗲𝘁𝘆. 🔑 𝗞𝗲𝘆 𝗕𝗲𝗻𝗲𝗳𝗶𝘁𝘀: 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗢𝗻-𝗗𝗲𝘃𝗶𝗰𝗲 𝗣𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴: Llama 3.2 reduces latency and improves performance, making it ideal for various applications across industries. 𝟭𝟮𝟴𝗞 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗟𝗲𝗻𝗴𝘁𝗵: Captures nuanced relationships in data for deeper insights. 𝗠𝘂𝗹𝘁𝗶𝗹𝗶𝗻𝗴𝘂𝗮𝗹 𝗖𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀: Supports 𝗲𝗶𝗴𝗵𝘁 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲𝘀, including English, Hindi, Spanish, and more. 𝗧𝗿𝗮𝗶𝗻𝗲𝗱 𝗼𝗻 𝟭𝟱 𝗧𝗿𝗶𝗹𝗹𝗶𝗼𝗻 𝗧𝗼𝗸𝗲𝗻𝘀: Built to understand complex language intricacies from public data. 𝗔𝗺𝗮𝘇𝗼𝗻 𝗕𝗲𝗱𝗿𝗼𝗰𝗸 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻: Access Llama's power via Amazon Bedrock's serverless API for seamless integration into your AI applications without worrying about infrastructure. 🌟 Use Cases: Llama models are designed for a wide array of tasks: 𝗜𝗺𝗮𝗴𝗲 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 and 𝗩𝗶𝘀𝘂𝗮𝗹 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗧𝗲𝘅𝘁 𝗦𝘂𝗺𝗺𝗮𝗿𝗶𝘇𝗮𝘁𝗶𝗼𝗻, 𝗦𝗲𝗻𝘁𝗶𝗺𝗲𝗻𝘁 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀, and 𝗗𝗼𝗰𝘂𝗺𝗲𝗻𝘁 𝗩𝗶𝘀𝘂𝗮𝗹 𝗤&𝗔 𝗗𝗶𝗮𝗹𝗼𝗴𝘂𝗲 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻, 𝗧𝗿𝗮𝗻𝘀𝗹𝗮𝘁𝗶𝗼𝗻, and 𝗖𝗼𝗱𝗲 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗩𝗶𝘀𝘂𝗮𝗹 𝗚𝗿𝗼𝘂𝗻𝗱𝗶𝗻𝗴, 𝗜𝗺𝗮𝗴𝗲-𝗧𝗲𝘅𝘁 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹, and much more! With Meta’s 𝗟𝗹𝗮𝗺𝗮 𝟯.𝟮 models (1B, 3B, 11B, 90B), you can handle multi-step, complex AI tasks while ensuring high performance and scalability. 🌍 Whether you're working in AI-powered 𝘃𝗶𝘀𝘂𝗮𝗹 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴, 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗺𝗼𝗱𝗲𝗹𝗶𝗻𝗴, or 𝗺𝘂𝗹𝘁𝗶-𝗹𝗶𝗻𝗴𝘂𝗮𝗹 𝘁𝗮𝘀𝗸𝘀, Llama 3.2 empowers your next-gen AI solutions with unmatched efficiency and capability. Let’s revolutionize the world of AI together! 🤖💡 #ai #genai #llama3.2 #meta #amazonbedrock #llm #personalizedai

11

1 תגובה

𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐭𝐨 𝐃𝐞𝐞𝐩 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 (𝐌𝐈𝐓, 2024 𝐂𝐨𝐮𝐫𝐬𝐞 𝐋𝐞𝐜𝐭𝐮𝐫𝐞 𝐕𝐢𝐝𝐞𝐨𝐬) **𝐂𝐨𝐮𝐫𝐬𝐞 𝐂𝐨𝐧𝐭𝐞𝐧𝐭𝐬** - Intro to Deep Learning - Deep Sequence Modeling - Deep Computer Vision - Deep Generative Modeling - Deep Reinforcement Learning - New Frontiers - Generative AI for Media - Stories from Models in the Wild **𝐒𝐨𝐟𝐭𝐰𝐚𝐫𝐞 𝐋𝐚𝐛𝐬** - Intro to TensorFlow; Music Generation - Facial Detection Systems - Large Language Models Course playlist details (refer to the comments) #deeplearning #beginners #nlproc #computervision #generativeai

90

2 תגובות

🚀 𝑩𝒓𝒆𝒂𝒌𝒊𝒏𝒈 𝑵𝒆𝒘 𝑮𝒓𝒐𝒖𝒏𝒅: 𝑴𝒆𝒎𝒐𝒓𝒚 𝑨𝒄𝒄𝒆𝒔𝒔 𝑴𝒐𝒅𝒖𝒍𝒆𝒔 𝒇𝒓𝒐𝒎 𝑺𝒄𝒓𝒂𝒕𝒄𝒉!🚀 I’m thrilled to announce the next step in my "𝓝𝓣𝓜 𝓯𝓻𝓸𝓶 𝓢𝓬𝓻𝓪𝓽𝓬𝓱" series: 𝑴𝒆𝒎𝒐𝒓𝒚 𝑨𝒄𝒄𝒆𝒔𝒔 𝑴𝒐𝒅𝒖𝒍𝒆𝒔!  In this project, I’ve implemented the core mechanism that allows neural networks to read from and write to external memory in a differentiable way. This is a crucial milestone on the journey to a complete Neural Turing Machine, bridging the gap between computation and memory access. 🔗 Check it out on GitHub: [From Scratch by Abhishek Pandey](https://lnkd.in/gXN9hN5Q)   Next up: putting it all together for a 𝘉𝘢𝘴𝘪𝘤 𝘕𝘦𝘶𝘳𝘢𝘭 𝘛𝘶𝘳𝘪𝘯𝘨 𝘔𝘢𝘤𝘩𝘪𝘯𝘦 𝘪𝘮𝘱𝘭𝘦𝘮𝘦𝘯𝘵𝘢𝘵𝘪𝘰𝘯. Stay tuned! #MachineLearning #NeuralTuringMachine #DeepLearning #NeuralNetworks #FromScratch #AI

5

بسم الله الرحمن الرحيم . 🚀 Exciting Project Update: Image Classification with HOG Features and GUI! 📊 I’m thrilled to share a recent project where I combined Histogram of Oriented Gradients (HOG) features with a Stochastic Gradient Descent (SGD) Classifier for image classification. The model achieved remarkable results: ✅ Accuracy: 86% ✅ AUC (Area Under the Curve): 0.93 Highlights include: Advanced Feature Extraction: Leveraged HOG for detailed image features. Robust Classification: Applied SGDClassifier with logistic loss and regularization. Comprehensive Evaluation: Utilized accuracy, ROC curve, confusion matrix, and loss curves to evaluate model performance. User-Friendly GUI: Developed a graphical user interface that allows users to easily upload images and receive predictions. The project demonstrates how combining sophisticated feature extraction with powerful classifiers and a user-friendly interface can create an effective image classification solution.  Explore the full details and code in my GitHub repository If I noticed a change in accuracy on Kaggle, where it's now 87% instead of 86%, and this is because I cropped some images on my personal device for testing the GUI ^^ Dataset : https://lnkd.in/dPMKsSF3 Github : https://lnkd.in/dhda3vSk Kaggle : https://lnkd.in/d2jU9b83

32

13 תגובות

🧠 🚨 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐞𝐫-𝐛𝐚𝐬𝐞𝐝 𝐃𝐨𝐠 𝐁𝐞𝐡𝐚𝐯𝐢𝐨𝐫 𝐂𝐥𝐚𝐬𝐬𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐰𝐢𝐭𝐡 𝐌𝐨𝐭𝐢𝐨𝐧 𝐒𝐞𝐧𝐬𝐨𝐫𝐬 ⚡ 🔥 - a new solo-authored paper by Dr. Barak Or 💥 💥 With a great pleasure I am sharing a new and a very cool paper of my dear friend Dr. Barak Or. ✅ This paper deals with classifying dog behavior using motion sensors, leveraging a transformer-based #deepneuralnetwork model. ✅ Understanding dog behavior is essential for fostering positive relationships between dogs and humans and ensuring their well-being. ✅ Traditional methods often fall short i capturing temporal dependencies and efficiently processing high-dimensional sensor data. ✅ Our proposed architecture, inspired by its success in #nlp, utilizes the self-attention mechanism of the transformer to effectively identify relevant features across various time scales, making it ideal for real-time applications. 😎 This research is done by MetaOr Artificial Intelligence's aiming at pushing AI to a new level, where it can be anywhere, even in pet care! #pets #transformers #ai

51

4 תגובות

I recently gave a 1.5-hour tech talk at NLP IL - Natural Language Processing Israel on several exciting papers related to Speculative Decoding. Here's a brief summary of each paper that I believe you'll find valuable: 1️⃣ Fast Inference from Transformers via Speculative Decoding: This paper presents a method to speed up and reduce the cost of LLM inference, called speculative decoding. Instead of generating tokens with the target (and slow) LLM, a faster draft model generates them. These tokens are then verified with a single forward pass through the target LLM. This ensures mathematically equivalent generations to those you'd get using the target LLM, at a significantly faster rate. 2️⃣ Draft & Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding: This approach eliminates the need for a separate draft model by using a subset of the target model's layers. The optimal subset for minimizing inference time is found through Bayesian optimization. Keep reading about the other methods in the comments 👇

49

4 תגובות

Our own Harpreet Sahota 🥑 discusses the key to building better models, the power of community, and his #1 piece of advice for newcomers in data science + AI. Give it a watch to catch some inspiration and practical tips! 💡 #DataScience #MachineLearning #AI #DataQuality

7

It was so simple, they couldn’t believe no one figured this out before. Prof. Moran Bercovici and Dr. Valeri Frumkin from the Technion - Israel Institute of Technology developed a new way to create optical lenses using cheap materials and a straightforward process. Here’s how it works: they inject a liquid polymer into a mold that’s submerged in an immersion liquid. By carefully adjusting the immersion liquid's density to match that of the polymer, they eliminate the effects of gravity, allowing the polymer to retain a perfectly smooth and precise shape as it hardens under UV light. The result is a lens with an exceptionally smooth surface—smoother than what traditional polishing can achieve. This method isn’t limited to simple lenses. It can also be used to create complex freeform optics, which are surfaces designed to bend and focus light in precise ways for applications like multifocal glasses, advanced projectors, and even virtual reality systems. These optical components are usually difficult to manufacture, but with this technique, they can be produced much more affordably. #NASA has recognized the potential of this innovation for space exploration. In microgravity, this method could be used to produce large, high-quality optics for telescopes directly in orbit, bypassing the limitations of launching bulky equipment from Earth. This simple yet ingenious Israeli invention opens up new possibilities both on Earth, by providing affordable glasses to those who need them, and in space, by enabling the next generation of telescopes. #IsraeliInnovation #Optics #Space

53

8 תגובות

🚀 Research Paper Highlights: "Diffusion Models Are Real-Time Game Engines" by Dani Valevski, Yaniv Leviathan, Moab Arar, Shlomi Fruchter. 🔍 Technical Approach: GameNGen is a groundbreaking neural model enabling real-time gameplay at 20 frames per second. This innovative method transforms interactive software, like computer games, into neural models. Inference Setup: Employs DDIM sampling and Classifier-Free Guidance (CFG) to boost generation quality, focusing on strong conditioning from past observations while avoiding action artifacts. 📊 Methods and Techniques: Denoiser and Auto-Encoder- Both run during inference to optimize high-quality image generation with minimal steps, efficiently operating on TPU-v5 hardware. Model Distillation- Explored for higher FPS, but the 4-step method is retained for optimal simulation quality. 📈 Evaluation: Performance Metrics: Image Quality-Achieves a PSNR of 29.43, similar to lossy JPEG compression, with an LPIPS of 0.249. Human Evaluation- High visual fidelity, with human raters finding it challenging to distinguish between simulated and actual gameplay. 🏆 Results and Comparisons: Simulation quality is high fidelity, especially in short trajectories, with minor divergences over longer simulations. ⚠️ Limitations: Requires significant computational resources, particularly for parallel sample generation, limiting broader application. 🌟 Responsible Development: Future research should focus on optimizing for diverse game environments, improving stability, and developing more efficient training methods. 🔗 Further Reading: https://lnkd.in/dkdTRvkN 🌟 Stay tuned for more updates on innovative methods and advancements in AI! #GameEngines #Innovation #AIResearch

8

𝐓𝐨𝐰𝐚𝐫𝐝𝐬 𝐓𝐡𝐞 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐄𝐜𝐨𝐧𝐨𝐦𝐲: Modern AI agents, driven by advances in large foundation models, promise to enhance our productivity and transform our lives by augmenting our knowledge and capabilities. To achieve this vision, 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭𝐬 𝐦𝐮𝐬𝐭 𝐞𝐟𝐟𝐞𝐜𝐭𝐢𝐯𝐞𝐥𝐲 𝐩𝐥𝐚𝐧, 𝐩𝐞𝐫𝐟𝐨𝐫𝐦 𝐦𝐮𝐥𝐭𝐢-𝐬𝐭𝐞𝐩 𝐫𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠 𝐚𝐧𝐝 𝐚𝐜𝐭𝐢𝐨𝐧𝐬, 𝐫𝐞𝐬𝐩𝐨𝐧𝐝 𝐭𝐨 𝐧𝐨𝐯𝐞𝐥 𝐨𝐛𝐬𝐞𝐫𝐯𝐚𝐭𝐢𝐨𝐧𝐬, 𝐚𝐧𝐝 𝐫𝐞𝐜𝐨𝐯𝐞𝐫 𝐟𝐫𝐨𝐦 𝐞𝐫𝐫𝐨𝐫𝐬, 𝐭𝐨 𝐬𝐮𝐜𝐜𝐞𝐬𝐬𝐟𝐮𝐥𝐥𝐲 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞 𝐜𝐨𝐦𝐩𝐥𝐞𝐱 𝐭𝐚𝐬𝐤𝐬 𝐚𝐜𝐫𝐨𝐬𝐬 𝐚 𝐰𝐢𝐝𝐞 𝐫𝐚𝐧𝐠𝐞 𝐨𝐟 𝐬𝐜𝐞𝐧𝐚𝐫𝐢𝐨𝐬. 𝐈𝐧 𝐭𝐡𝐢𝐬 𝐰𝐨𝐫𝐤, 𝐭𝐡𝐞 𝐚𝐮𝐭𝐡𝐨𝐫𝐬 𝐢𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐞 𝐌𝐚𝐠𝐞𝐧𝐭𝐢𝐜-𝐎𝐧𝐞, 𝐚 𝐡𝐢𝐠𝐡-𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐢𝐧𝐠 𝐨𝐩𝐞𝐧-𝐬𝐨𝐮𝐫𝐜𝐞 𝐚𝐠𝐞𝐧𝐭𝐢𝐜 𝐬𝐲𝐬𝐭𝐞𝐦 𝐟𝐨𝐫 𝐬𝐨𝐥𝐯𝐢𝐧𝐠 𝐬𝐮𝐜𝐡 𝐭𝐚𝐬𝐤𝐬. Magentic-One uses a multi-agent architecture where a lead agent, the Orchestrator, plans, tracks progress, and re-plans to recover from errors. Throughout task execution, the Orchestrator directs other specialized agents to perform tasks as needed, such as operating a web browser, navigating local files, or writing and executing Python code. 𝐓𝐡𝐞 𝐚𝐮𝐭𝐡𝐨𝐫𝐬 𝐬𝐡𝐨𝐰 𝐭𝐡𝐚𝐭 𝐌𝐚𝐠𝐞𝐧𝐭𝐢𝐜-𝐎𝐧𝐞 𝐚𝐜𝐡𝐢𝐞𝐯𝐞𝐬 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥𝐥𝐲 𝐜𝐨𝐦𝐩𝐞𝐭𝐢𝐭𝐢𝐯𝐞 𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐭𝐨 𝐭𝐡𝐞 𝐬𝐭𝐚𝐭𝐞-𝐨𝐟-𝐭𝐡𝐞-𝐚𝐫𝐭 𝐨𝐧 𝐭𝐡𝐫𝐞𝐞 𝐝𝐢𝐯𝐞𝐫𝐬𝐞 𝐚𝐧𝐝 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐢𝐧𝐠 𝐚𝐠𝐞𝐧𝐭𝐢𝐜 𝐛𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤𝐬: 𝐆𝐀𝐈𝐀, 𝐀𝐬𝐬𝐢𝐬𝐭𝐚𝐧𝐭𝐁𝐞𝐧𝐜𝐡, 𝐚𝐧𝐝 𝐖𝐞𝐛𝐀𝐫𝐞𝐧𝐚. Magentic-One achieves these results without modification to core agent capabilities or to how they collaborate, 𝐝𝐞𝐦𝐨𝐧𝐬𝐭𝐫𝐚𝐭𝐢𝐧𝐠 𝐩𝐫𝐨𝐠𝐫𝐞𝐬𝐬 𝐭𝐨𝐰𝐚𝐫𝐝𝐬 𝐠𝐞𝐧𝐞𝐫𝐚𝐥𝐢𝐬𝐭 𝐚𝐠𝐞𝐧𝐭𝐢𝐜 𝐬𝐲𝐬𝐭𝐞𝐦𝐬. Moreover, 𝐌𝐚𝐠𝐞𝐧𝐭𝐢𝐜-𝐎𝐧𝐞'𝐬 𝐦𝐨𝐝𝐮𝐥𝐚𝐫 𝐝𝐞𝐬𝐢𝐠𝐧 𝐚𝐥𝐥𝐨𝐰𝐬 𝐚𝐠𝐞𝐧𝐭𝐬 𝐭𝐨 𝐛𝐞 𝐚𝐝𝐝𝐞𝐝 𝐨𝐫 𝐫𝐞𝐦𝐨𝐯𝐞𝐝 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐭𝐞𝐚𝐦 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐚𝐝𝐝𝐢𝐭𝐢𝐨𝐧𝐚𝐥 𝐩𝐫𝐨𝐦𝐩𝐭 𝐭𝐮𝐧𝐢𝐧𝐠 𝐨𝐫 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠, 𝐞𝐚𝐬𝐢𝐧𝐠 𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐦𝐚𝐤𝐢𝐧𝐠 𝐢𝐭 𝐞𝐱𝐭𝐞𝐧𝐬𝐢𝐛𝐥𝐞 𝐭𝐨 𝐟𝐮𝐭𝐮𝐫𝐞 𝐬𝐜𝐞𝐧𝐚𝐫𝐢𝐨𝐬. The authors provide an open-source implementation of Magentic-One, and they include AutoGenBench, a standalone tool for agentic evaluation. AutoGenBench provides built-in controls for repetition and isolation to run agentic benchmarks in a rigorous and contained manner -- which is important when agents' actions have side-effects. SOURCE: https://lnkd.in/ga33RY-r CODE: https://lnkd.in/gSgmT36s

42

1 תגובה

✅ DIY Code Completion with #Mamba 🐍 ➡️ https://nvda.ws/3SLBCDq ✨ ⚡ #TensorRT #LLM for speed 🤖 NIM for ease and ☁️ Deploy anywhere

2

Why is the Kernel Trick called a "trick" 🧩? Many ML algorithms use kernels for robust modeling, like SVM, KernelPCA, etc. The core objective of a kernel function is to compute dot products in some other feature space (mostly high-dimensional) without projecting the vectors to that space. But how does that even happen? I covered this in detail here: https://lnkd.in/gFvW6Yty. Consider the image below. Let’s assume the following polynomial kernel function: - k(X, Y) = (1+XᵀY)². Also, for simplicity, let’s say both X and Y are two-dimensional vectors: - X = (x1, x2) - Y = (y1, y2) As shown in the image below, simplifying the kernel expression produces a dot product between the two 6-dimensional vectors. This shows that the kernel function we chose earlier computes the dot product in a 6-dimensional space without explicitly visiting that space. And that is the primary reason why we also call it the “kernel trick.” More specifically, it’s framed as a “trick” since it allows us to operate in high-dimensional spaces without explicitly computing the coordinates of the data in that space. -- 👉 Join 79k+ data scientists and get a Free data science PDF (550+ pages) with 320+ posts by subscribing to my daily newsletter: https://lnkd.in/gzfJWHmu -- 👉 Over to you: Can you tell a major pain point of the kernel trick algorithms?

223

8 תגובות