Angie McAllister, Ph.D.’s Post

If you are like me, you are reeling 🤯 from all the amazing advancements in #generativeai the past two days. Not only did OpenAI unveil many amazing updates, but yesterday was another HUGE day in the #generativeAI world with all the amazing announcements at #googleIO. Many advancements are similar to the ones announced by OpenAI the day before- We will soon have TWO super helpful multimodal AI assistants, capable of emotional intonation, visual (image and video) assistance, huge context windows, and multiple ways to streamline work with context aware assistance. If you missed it, here is a pretty thorough summary of #GoogleIO by TechCrunch https://lnkd.in/epCSrf6n

In my last post, we discussed some of the text to video models and their capabilities, limitations. Now, Let's see what Sora is all about. OpenAI released Sora, a wonderful work which can not only create longer videos but also overcomes incoherencies. What makes OpenAI's #sora to generate the significant results in video generation?. let's see what OpenAI (ClosedAI 😉 ) says through their research blog (reference at comment section) In Simple words, Sora is not just diffusion, Its Diffusion + Transformers !! 1) Video compression network: - Raw video as input and outputs a latent representation that is compressed both temporally and spatially. - Sora is trained on and subsequently generates videos within this compressed latent space. Corresponding decoder model trained maps generated latents back to pixel space. 2) Spacetime latent patches: - Given a compressed input video, a sequence of spacetime patches are extracted which act as transformer tokens. - **Patch-based representation enables Sora to train on videos and images of variable resolutions, durations and aspect ratios unlike the previous works in video generation. - At inference time, the size of generated videos can be controlled by arranging randomly-initialized patches in an appropriately-sized grid. 3) Scaling transformers for video generation: - Given input noisy patches (and conditioning information like text prompts), it’s trained to predict the original “clean” patches. 4) Variable durations, resolutions, aspect ratios: - unlike past approaches to image and video generation which typically resize, crop or trim videos to a standard size—e.g., 4 second videos at 256x256 resolution. training on data at its native size provides several benefits as demonstrated by Sora. 5) Language understanding: - Training text-to-video generation systems requires a large amount of videos with corresponding text captions. - The re-captioning technique introduced in DALL·E 3 is applied to videos. A highly descriptive captioner model is first trained and then it is utilized to produce text captions for all videos in the training set for Sora. It is found that training on highly descriptive video captions improves text fidelity as well as the overall quality of video. #openai #sora #microsoft #google Google #deepmind #genai #generatieveai #artificialintelligence #machinelearning #texttovideo #videogeneration

🚀 New Project Update: Object Classification with Google Teachable Machine! 🚀 I'm excited to share my recent project where I developed an object classification model using Google Teachable Machine! 📸🧠🤖 🔧 Key Highlights: 🔹 Google Teachable Machine: Leveraged Google Teachable Machine's user-friendly platform to create a powerful object classification model. This tool made it easy to train the model with minimal coding required. 🔹 No-Code/Low-Code Approach: Embraced a no-code/low-code approach, making advanced machine learning accessible and efficient for quick prototyping and deployment. 🔹 Practical Applications: Explored various real-world applications, demonstrating the model's potential in fields like education, healthcare, retail, and more. 🛠️ Project Breakdown: 1. Data Collection & Training: - Gathered diverse image datasets for different object classes. - Trained the model using Google Teachable Machine’s intuitive interface, ensuring a robust classification capability. 2. Model Fine-Tuning: - Fine-tuned the model to enhance accuracy and performance. - Implemented additional training rounds with varied datasets to improve generalization. 3. Deployment: - Deployed the model for real-time object classification. - Created an easy-to-use interface for users to upload images and receive instant classification results. 🌟 Features: - Ease of Use:The no-code platform made it simple to develop and deploy the model, significantly reducing development time. - High Accuracy: Achieved impressive accuracy rates through iterative training and fine-tuning. - Versatile Applications:The model can be adapted for various uses, from educational tools to automated sorting systems. #AI #MachineLearning #ObjectClassification #GoogleTeachableMachine #NoCode #Innovation #Technology #DeepLearning #Automation #DataScience #AIMERS #AIMERSOCIETY #APSCHE

🚀 Exploring Human Pose Estimation with Google Teachable Machine 🚀 I recently embarked on an exciting journey into the realm of human pose estimation using Google Teachable Machine, and I'm thrilled to share my experience with you all! 🌟 Why Google Teachable Machine? Google Teachable Machine is an intuitive and accessible tool that empowers anyone to create machine learning models, regardless of their coding background. It's ideal for both quick prototyping and educational purposes. 🕺 My Project: Human Pose Estimation I decided to delve into a human pose estimation project. Here's a step-by-step overview of my process: Selecting the Project Type: I began by choosing the 'Pose Project' option in Google Teachable Machine. Labeling Poses: Using my webcam, I captured and labeled various human poses. This step was critical as the accuracy of the model heavily depends on the quality and variety of the labeled poses. Training the Model: After labeling the poses, I trained the model. The training process was smooth and efficient, showcasing the robust capabilities of Google Teachable Machine. Testing the Model: Once the training was complete, I tested the model with new poses to evaluate its performance. The results were remarkable, demonstrating the model's proficiency in accurately estimating human poses based on the training data. 💡 Key Takeaways: User-Friendly and Accessible: Google Teachable Machine makes advanced machine learning techniques like pose estimation accessible to everyone, even those without a technical background. Hands-On Experience: This project offered a practical experience that deepened my understanding of human pose estimation and machine learning. Efficient Prototyping: The ability to quickly test and refine models makes it an excellent tool for rapid prototyping. 🔗 Give It a Try! If you're interested in exploring human pose estimation or need a tool for quick prototyping, I highly recommend trying Google Teachable Machine. Start your own project here: Google Teachable Machine Feel free to reach out if you have any questions or want to discuss machine learning further! #MachineLearning #AI #GoogleTeachableMachine #PoseEstimation #TechInnovation #HandsOnLearnin #AIMERS

https://lnkd.in/eDn_Qh_B Video annotator: a framework for efficiently building video classifiers using vision-language models and active learning

Hey LinkedIn Community! 🌟 I'm Syed Ibrahim, and today I’m excited to walk you through using Google Teachable Machine for pose detection. This intuitive tool enables anyone to create machine learning models effortlessly, without any coding knowledge! 🚀 Here’s a quick breakdown of our tutorial: 1)Project Setup: Begin by creating a new project and selecting the 'Pose' project type. 2)Class Creation: We'll create two classes for this example: 'Thumbs Up' and 'Hi'. - Thumbs Up: Capture multiple images of the thumbs-up pose, varying angles and distances to enhance model accuracy. - Hi: Repeat the process with the waving pose. 3)Model Training: After collecting enough images for both classes, train the model. Teachable Machine will process the images and generate a pose detection model in just a few minutes. 4)Model Testing: Finally, test the model. When you give a thumbs up or wave 'Hi', the model will detect and classify the poses accurately. This tutorial showcases how easily you can leverage Teachable Machine to build custom pose detection models. 🌐 🎥 Watch the full video to see the entire process, and don’t hesitate to ask any questions or share your thoughts in the comments below. Your feedback is greatly appreciated! Hashtag: #AIMERS #PoseDetection #TeachableMachine #MachineLearning #AI #NoCode #DataScience #ComputerVision #DeepLearning #TechTutorial #GoogleTeachableMachine #Innovation #TechGuide #AIInAction #CustomML

"Successfully performed a similarity search using AlloyDB, pgvector, and Vertex AI! Combining the power of contextual and vector searches for efficient, meaning-driven insights. Excited to explore agent building next! #AI #MachineLearning #DataScience #TechInnovation"

🚀 Excited to Share My Latest Project! 🚀 I recently completed an image classification project using Google Teachable Machine. 🌟 This intuitive tool made it possible to create a robust machine learning model without needing extensive programming knowledge. 🔍 Project Highlights: Tool Used: Google Teachable Machine Objective: Classify images into different categories with high accuracy. Process: Collected and labeled a diverse set of images. Trained the model using Google Teachable Machine's user-friendly interface. Tested the model's performance and iterated to improve accuracy. Outcome: Successfully built a model capable of accurately classifying images into predefined categories. #MachineLearning #AI #ImageClassification #GoogleTeachableMachine #TechInnovation #Aimer society

New video: "Hands-on AI with Gemini 1.5 Pro": Images, Video, Text & Code Gemini 1.5 Pro is our AI model from Google that can handle text, images, videos, audio, and even code. With a whopping 1 million token context window, it can analyze massive amounts of data in one go. To put it to the test, I threw various challenges its way. From extracting book titles from images and videos to summarizing entire books, generating web apps based on screencasts, and processing code repositories, it impressed at every turn. I know we have lots of areas to improve, but great to see how capable it is already. If you have access to this model through AI Studio or the API, I highly recommend exploring its features and seeing how it can help your workflows. 📹 For a more in-depth look, check out this video where I walk through the examples mentioned in this post. This video uses AI Studio (https://ai.google.dev) #AI #softwareengineering #programming #lifeatgoogle