Informix

 View Only

  • 1.  Vote now for : Multi Vector database functionality as Database Extend to Informix ( INFX-I-612 )

    IBM Champion
    Posted 25 days ago
    Edited by Henri Cujass 25 days ago
    We discuss during the leolo Mastermind Conference that will be an exceptional feature if Informix 15 has a Multi-Vector Database functionality (like Milvus).
    May an additional functionality like time-series or as datablade like geospatial and text-search or as part of the core machine.
    IBM Informix recommended I create a request because it's a long-term idea of the Product team since Version 12.10 . If we get enough votes for this feature from the community and customers, there's a large chance that R&D will put the Vector Database functionality into Informix Products.
    Small background: It's a mandatory feature for AI to use RAG use cases in the future !
    Please vote for Vector Database functionality inside Informix - and bring up Informix to the next AI level!
    Best Regards
    Henri



    ------------------------------
    Henri Cujass
    leolo IT, CTO
    Germany
    IBM Champion
    henri.cujass@leolo.com
    ------------------------------



  • 2.  RE: Vote now for : Multi Vector database functionality as Database Extend to Informix ( INFX-I-612 )

    IBM Champion
    Posted 25 days ago

    Mind explaining "Multi-Vector Database functionality (like Milvus)", to the dummies among us?



    ------------------------------
    Andreas Legner
    ------------------------------

    Original Message


  • 3.  RE: Vote now for : Multi Vector database functionality as Database Extend to Informix ( INFX-I-612 )

    IBM Champion
    Posted 25 days ago

    Hi Andreas,

    thank you for the great question.

    I mean a new datatype to store multi-dimensional vectors - maybe in the SLOB DBSpace? 
    Special Index types on this column type "vector" with different approximation methods and a set of functions to use this vector datatype for AI use cases (RAG, LLM).

    Great to start the discussion on this subject. I pleased Andreas Weininger to share some knowledge about Vector Database Functionality with us, too. 
    Maybe other specialists can complete the needed functionality also.

    Best Regards

    Henri



    ------------------------------
    Henri Cujass
    leolo IT, CTO
    Germany
    IBM Champion
    henri.cujass@leolo.com
    ------------------------------

    Original Message


  • 4.  RE: Vote now for : Multi Vector database functionality as Database Extend to Informix ( INFX-I-612 )

    IBM Champion
    Posted 25 days ago

    Hi Andreas,

    three slides I got from Andreas Weininger about this topic. The first Image to illustrate Vector Data inside a Database. The second and third to explain the search functionality which can be supported by special indexes. 

    Best Regards

    Henri



    ------------------------------
    Henri Cujass
    leolo IT, CTO
    Germany
    IBM Champion
    henri.cujass@leolo.com
    ------------------------------

    Original Message


  • 5.  RE: Vote now for : Multi Vector database functionality as Database Extend to Informix ( INFX-I-612 )

    IBM Champion
    Posted 25 days ago

    Hi Andreas,

    one more explanation Andreas Weininger pleased me to post:

    Vector databases store (as the name implies) vectors of numerical components. The number of components of a vector is called the dimensionality of the vector. Vector databases can retrieve the most similar vectors for a given search vector. Several similarity metrics can be used e.g., Euclidian distance. For performance reasons vector databases will give an approximate result instead of the exact answer. Vector databases use indexes/methods like HNSW (https://en.wikipedia.org/wiki/Hierarchical_navigable_small_world) or inverted files for efficient retrieval of the results.

    A typical use case for vector databases is Retrieval Augmented Generation where information encoded via so-called embeddings in vectors with hundreds or thousands of dimensions is stored in a vector database. A query to the vector store is used to get additional information which can be fed into the prompt of a large language model. The large language model can now answer a query and include information which wasn't part of its original training.

    Best Regards, Henri



    ------------------------------
    Henri Cujass
    leolo IT, CTO
    Germany
    IBM Champion
    henri.cujass@leolo.com
    ------------------------------

    Original Message


  • 6.  RE: Vote now for : Multi Vector database functionality as Database Extend to Informix ( INFX-I-612 )

    Posted 25 days ago

    A vector database stores (as the name implies) vectors of numeric components. The number of components are the dimensionality of the vector. The vector database allows to find for a given search vector the most similar vectors in the database. Similarity may be defined according to different metrics e.g., Euclidian distance. For performance reasons vector bases often provide an approximate answer instead of the exact answer. To answer queries fast there are several methods/indexes in vector databases e.g., HNSW (https://en.wikipedia.org/wiki/Hierarchical_navigable_small_world) or inverted files. A typical use case for vector databases is Retrieval Augmented Generation where information is encoded in vectors via so called embeddings, and a vector search provides additional information which is fed via the prompt to a large language model to get an answer from the large language model which contains information that was not initially part of the trainings data of the large language model.



    ------------------------------
    Andreas Weininger
    ------------------------------

    Original Message