DeepLearning.AI founder and chief Andrew Ng recently launched a new short course “Embedding Models: From Architecture to Implementation,” covering word, sentence, and cross-encoder models, BERT training, and building dual encoder models for semantic search, taught by Vectara’s Ofer Mendelevitch.
This technical course covers the history, architecture, and capabilities of embedding models. “You may have seen embedding vectors used in many generative AI contexts. These vectors have the remarkable ability to capture the meaning of a word or phrase,” said Ng.
Further, he said that vectors with similar meanings are close to each other in vector space, allowing comparsions with low-cost algorithms like cosine similarity to find nearby vectors. “This revolutionised retrieval by enabling semantic search, which retrieves query results with similar semantic meanings rather than just matching keywords,” added Ng, saying that Vectara has built RAG systems that are deeply familiar with embedding models.
Learn more about the course here.
Andrew Ng is on a roll with his latest series of course launches. A few days back, Andrew Ng’s DeepLearning.AI introduced a two-part course on federated learning and federated fine-tuning of LLMs with private data, developed with Flower Labs, teaching distributed model training with privacy preservation, differential privacy techniques, and efficient federated learning strategies.
Last week, in collaboration with Upstage, the company launched another short course on ‘Pretraining LLMs,’ teaching data preparation, model architecture, training, evaluation, and innovative techniques like depth up-scaling to reduce pretraining compute costs by up to 70%.
Early this month, it also launched a new course on optimising RAG for cost and performance, “Prompt Compression and Query Optimization,” created with MongoDB, teaching vector search, metadata filtering, projections, boosting, and prompt compression techniques for scalable, efficient RAG applications.
Last month, DeepLearning.AI introduced “Carbon Aware Computing for GenAI Developers,” a new course taught by Google Cloud’s Nikita Namjoshi, teaching techniques to reduce AI’s carbon footprint by using real-time carbon intensity data, routing to low-carbon data centers, measuring emissions, and optimising job scheduling for clean energy availability.
In the same month, it also launched “Function Calling and Data Extraction with LLMs,” a new course created with NexusflowX and taught by Jiantao Jiang and Venkat Srinivasan, teaching how to extend LLM capabilities with external tools, using the NexusRavenV2-13B model for function calling tasks, extracting structured data, accessing web APIs, and building LLM-powered applications for customer service automation.
In June, DeepLearning.AI also Introduced “Building Your Own Database Agent,” a new course created with Microsoft Azure’s Adrian Gonzalez, teaching how to build an AI assistant that translates natural language questions into SQL queries using Azure OpenAI Service and LangChain, empowering users to access data insights directly with plain English queries and hands-on experience in working with CSV files and SQL databases.
Before that, DeepLearning.AI also launched “AI Agents in LangGraph,” a new course taught by LangChain founder Harrison Chase and TAVily founder Rotem Weiss, demonstrating how to use LangGraph to build single and multi-agent LLM applications, integrate agentic search and memory, and incorporate human-in-the-loop input, culminating in the creation of a sophisticated essay-writing agent. The list goes on and on.
