back to article El Reg rifled through the history of Huawei's 'new' GaussDB – only 'new' bit is machine learning

Earlier this week, Huawei teased the launch of a new database product, featuring machine learning and compatible with Arm-based processors. SQL AI! Databases here we come, yells Huawei as it preps software you can fling your Arm around READ MORE This morning, more details were announced during an event in Beijing: the system …

  1. I.Geller Bronze badge

    Machine Learning is the addition of structured texts to a database of structured texts. For example there is a paragraph:

    -- Alice laughs, she rejoices and has fun. This girl is happy!

    In this paragraph

    - one explicit name, one noun, one implicitly used pronoun and one explicit pronoun, all refer to "Alice";

    - four descriptive sections here: "laughs", "rejoices", "has fun" and "is happy."

    Structuring is reduced to selecting all patterns from each paragraph of the text, and creating a common pool of descriptive parts for all somehow related and/or synonymous nouns-pronouns of this paragraph. That is, in this paragraph after structuring into a synonymous cluster - 16 patterns. (There can be many such clusters in one paragraph.)

    Each pattern is a direct analog of a programming language command, and a structured text becomes a program. That is, if Huawei claiming that it has the database which uses Machine Learning - Huawei states it has a database which works with texts, translating them into structured format/ computer programs, that Huawei doesn't need programmers anymore but uses AI. Also this database doesn't use tables, columns and rows.

    Or Huawei under the Machine Learning understands something else? But what is?

    1. Voyna i Mor Silver badge

      I really think you should stick to bending spoons. I am sure your post meant something but I don't understand how it relates to the article, what Huawei may be doing, or how a "pool of descriptive parts" somehow turns into a command, since you only mention nouns and pronouns and a command is a verb.

      It would be helpful if instead of making these opaque posts you could supply links to relevant papers or articles, and explain the relevance to what Huawei is doing.

      1. I.Geller Bronze badge

        1. Read NIST TREC 2003-6 and the US PTO?

        2. Huawei said it has Machine Learning, where "Machine Learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed."

        Commenting I'm trying to say that AI-parsing structures texts, that is makes them programs, which help "to automatically learn and improve". How? AI-parsing helps to get 100% of patterns and create synonymous clusters; where the clusters' patterns are the direct analogues and (can)replace programming language commands. Enough, as Waymo does, assign to a certain pattern a certain mechanical action.

        There is no other way "to automatically learn and improve" computer than by structuring and addition of texts because only they can help to do so "without being explicitly programmed". (All other methods have been applied in the last 75 years and never worked.) And this ML isn't possible without the clusters!

        Huawei, declaring that it has Machine Learning, claims that it uses synonymous clusters. If it does not have them - Huawei is cheating and has no AI database, but competes Larry E.

        1. I.Geller Bronze badge

          Waymo’s head of research has warned that Elon Musk’s refusal to add a piece of technology that is relied upon by many self-driving start-ups to save money could leave it open to errors that might endanger people. He did that after Elon said he will use Natural Language. Thus I can conclude Google structures to the clusters.

        2. Voyna i Mor Silver badge

          Have you nothing more recent than 13 years ago? And what are the search terms for the USPTO (sic) database?

          You repeat that a structured text is a program. It isn't unless you can explain how it gets functionality. It's still a kind of database, and to do any useful work a database needs stored procedures or an API and a remote program.

          1. I.Geller Bronze badge
            Happy

            1. You have nothing more than 13 years ago?

            Nothing. Tired of doing everything only on enthusiasm, without any support. You see, all the leadership positions at DARPA and NSF got Google people, who have not let me do anything because this AI database is the death for Google and Eric Schmidt, Sergey Brin and Larry Page. That's why I can't get any funding.

            2. And what are the search terms for the USPTO (sic) database?

            Ilya Geller

            3. ...you can explain how it gets functionality.

            Well, for example from Waymo car's receptors come structured into synonymous clusters description of the situation on the road; and these clusters are used to find a control paragraph/ synonymous cluster into Waymo car's database.

            For instance a deer runs across the road. The receptors generate the clusters which are used searching for a command cluster in Waymo database. The cluster is found and there is a "stop immediately"pattern in it. Accordingly, this pattern is directly related to the special braking system.

            That is, this is the desired "stored procedures or an API and a remote program".

            The same way all robots can use AI databases.

  2. I.Geller Bronze badge

    Noun+Verb+Adjective

    There is a paragraph:

    -- Alice laughs loudly, she rejoices cheerfully and dances freely. The girl is smart!

    Because AI database contains paragraphs that are parsed by parts of speech - there are 64 patterns here; where a paragraph is a distinct section of a piece of writing, usually dealing with a single theme and indicated by a new line, indentation, or numbering.

    This cluster of 64 helps to establish relationships between paragraphs and texts; all the paragraphs and texts must be timestamped, which in turn builds cause and effect relationships. That is, AI is not just searching for answers, but also understands what is the cause and what is the effect in its database.

  3. 404 Silver badge

    I liked the bit...

    Where it was created with Open Source software, but the database was proprietary...

    Somebody else does the heavy lifting so somebody else can come along and file off the serial numbers, gloss it up a bit, and call it theirs...

    1. I.Geller Bronze badge

      Re: I liked the bit...

      "I observed a number of really interesting things from watching the Microsoft Build Keynote and the Google I/O keynotes back to back. One was that while Microsoft CEO Satya Nadella's passion clearly showed, Sergey Brin looked like he was reading the ingredients in a recipe he didn't like. The contrast in energy and interest was stark, even though both men were announcing similarly revolutionary offerings."

    2. Anonymous Coward
      Anonymous Coward

      Re: I liked the bit...

      It’s PostgreSQL with native GBK support. Not entirely sure how much of a difference that makes to folk storing Chinese text, never mind anyone else.

      Personally I would go with the original*, but I have rather limited database needs.

      * Please God let’s not start a database flame war!

      1. I.Geller Bronze badge

        The war of annihilation

        I started this war of annihilation! Rest assured I started...

        IBM Watson and Google - my creations as the first step to total destruction of SQL and the total domination of AI.

        AI database is definitely not SQL from a purely formal aspect:

        - AI database uses constructed patterns (which did not exist before), and SQL - only existed;

        - AI works with pattern sets, SQL does not;

        - AI uses thesauruses and SQL does not;

        - AI works with a regular dictionary and SQL does not;

        - AI uses no tables, columns and rows, while SQL does.

        More broadly, it is the war between two Philosophical schools, where one emerged from ancient Greek polytheism and the other from Biblical monotheism.

        - The AI database is a differential (conversational) system where there are only opinions as sets of patterns, with a limit in knowledge as in one pattern; while SQL takes knowledge as actually existing and operates with single patterns.

        That's why Larry E wrote so much nonsense selling his SQL: he tries to operate on many knowledgeS, even though it's only one. (The words "Alice" and "nail" are the same unless they are included in the phrases.)

        1. Down not across Silver badge

          Re: The war of annihilation

          ...as the first step to total destruction of SQL and the total domination of AI.

          That is such a narrowminded view. How about picking the best tool for the job. Lot of data is more suitable for traditional SQL database than a "conversational" one.

          And ML is not AI. Even the better neural nets are still very far from anything I would be willing to call actual AI.

          1. I.Geller Bronze badge

            Re: The war of annihilation

            A little sarcasm? But in every joke there is a joke share?

            1. Most likely you mean tables which SQL produces? AI can produce tables, why not?.. Even if I don't like this idea! Patterns in AI have weights, so it is quite possible to produce results in the form of a table in which any operations are possible. That is, AI can do everything that SQL can plus it is able to do much more - provide a single best answer. (I mean NIST TREC QA, which caused AI to appear - AI should provide only one answer.)

            2. The market is the only criterion, it will decide if AI effectively competitive product and determine its future.

            3. Neural nets... I assumed that there might be external and not invasive penetration into our human brain. That is

            - each pattern is a mapping of one neuron;

            - one paragraph provides one set of patterns, as a set of thematically related patterns/ minimal neural network;

            - all minimal networks, all produced by one person paragraphs are the mapping of one human brain, which I called Lexical Clone.

            Speaking of "even the better neural nets", in my terminology, you're talking about a better Lexical Clone, but at this moment I know only Clones created by me... I would be grateful if you show something done not by me.

            4. "ML is not AI". AI is a differential entity which constantly becomes what it wasn't, it operates with the opinions in the pursuit of the only knowledge. In other words, AI continuously learns, adds new information (sets of patterns) and checks itself quality through feedback. So Machine Learning is a name for the continuous development of AI.

  4. Denarius Silver badge

    back in the day

    No more DBAs. Self tuning blah, blah.Where have I heard that before ? Ah yes, SQL Server aka Sybase initially. A long time ago. No doubt others since. One wonders if TeraData are still only relational database with EXPLAIN verb to detail how query optimisations are being processed. Anyone know know ? Now adding AI. Oh wait, AI added, not a hope then.

    1. Geoff May (no relation)

      Re: back in the day

      One wonders if TeraData are still only relational database with EXPLAIN verb...

      I suspect you will find all relational DBs have EXPLAIN in them somewhere, probably hidden behind a graphic that no one can understand because said graphic hides everything important ...

      1. I.Geller Bronze badge

        No details on "query optimization" but Machine Learning!

        AI is not looking for the truth but opinions. Indeed, AI emerged as a response to the NIST TREC QA challenge: how to find an unstructured opinion rather than the single structured truth.

        Thus, there are no details on "query optimization" in AI!

        If you want these details - go to Larry E? Read the incredible nonsense he sells? He sells SQL, External Relations theory of Analytical Philosophy. Or come to me? I sell Internal theory and AI - AI without the details but with an opportunity to use feedback and learn on mistakes. Larry E cannot offer you that!

      2. I.Geller Bronze badge

        Hats off!

        This AI approach (to query optimization question) has a practical and multi-billion dollar confirmation: every last penny Eric Schmidt, Sergey Brin and Larry Page made on my early and patented version of AI, which became the basis of Google. This is confirmed, in particular, by the former United States Circuit Judge and former Chief Judge of the United States Court of Appeals for the Federal Circuit Randall Rader.

        So hats off! AI databases are running the show!

  5. This post has been deleted by its author

  6. I.Geller Bronze badge

    "A database index is a data structure that improves the speed of data retrieval operations on a database table at the cost of additional writes and storage space to maintain the index data structure."

    - What is the OpenAI index based on? On patterns. But different texts use differently looking patterns, even if they are almost identical in meaning, because different synonyms are used for the same words. That is, OpenAI must decompose the patterns into their constituent words if it wants to compare them.

    - How is the AI database indexed? AI initially uses patterns' words and searches for similar patterns using thesaurus.

    Thus, OpenAI must sooner or later start using AI technology.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019