oraclebdcm

.@plburris 1 growing trend is the need to be agile. Only then can you expect enterprise adoption of big data and ML/AI in the cloud.

Agile at Oracle means iterating until you get it right. Neil Mendelson, Oracle – CUBEConversation - #theCUBE #oracle #bigdata https://video.cube36...

ML/AI will start popping up in less flashy places -- hiring, onboarding, reducing post-sale returns

The next big trend is reading about ML/AI efforts in Annual Reports.

Don't you wish you had an algorithm that pointed you to the most important sources of information and people when you were a new hire

The next big trend is for enterprises to adopt ML/AI-enabled SaaS offerings in public clouds. More enterprises of all sizes rely on their cloud ERP/CRM provider to offer these data-driven smarts are core features.

We also see the emergence of AI-based Adaptive Intelligence Apps for CX, ERP, HCM, and SCM - See @jpberkowitz on leveraging AI in Modern CX #ModernCX https://youtu.be/D8z...

Natural Language Understanding (NLU) for documents (unstructured data); and NLG (N.L. Generation = Narrative Science = automated data storytelling) from structured and un-structured data stores

Another big trend is for enterprises to use crowdsourced ML/AI capabilities--such as continuous ML training--that supplement their own more limited in-house data science resources.

AI/ML will empower machine intelligence, thus robotic process automation across many business units (out-facing and in-facing): CX, UX, and EX (Employee eXperience)

in a study by @Tractica on top AI use cases by 2025, they predict in order 1)algorithmic trading, 2)image recognition/tagging, 3)patient data processing 4)predictive maintenance and 5) content distribution on social media

The algorithmic business (#MathematicalCorporation) is here to stay => orgs must embrace a culture of experimentation around data/ML/AI => DataOps!

Another big trend is the spread of continuous delivery across enterprise apps to take advantage of continuously improving ML/AI

Checkout Oracle Day by Day in the Apps store - you can ask a question via text or voice

@PaulSonderegger That's DevOps for ML/AI in the cloud. Super-hot!

Cognitive Predictive / Prescriptive Maintenance in the IoT / IIoT era will be really big! ... https://www.slidesha... by @DataRPM

@jameskobielus @PaulSonderegger I prefer to call that DataOps ... http://www.oreilly.c...

Brilliant!

My former employer preferred to call it InsightOps. Different strokes...

.@plburris Well for one, tech shouldn't be in the driver's seat. Let orgs lead with the business questions first. Neil Mendelson, Oracle – CUBEConversation - #theCUBE #oracle #bigdata https://video.cube36...

Data governance: you need trustworthy data to select the right predictors to build into your ML/AI.

When analyzing sensitive data, like PII, masking and redaction can be effective for a data-centric approach

security should be number 1 priority. Look at Equifax. Data Capital goes to zero if you can't protect it for use.

Model governance: you need version control of ML/AI models for A/B testing, champion-challenger, et.

Algorithms can be subjected to audits. That's why some forms of AI can not be used in a regulated environment

Then there are people-centric approaches where roles and access rules limit the data science team

The intersection of the two, like multiple data zones with different access and data obfuscation rules is better

Culture eats strategy for breakfast (and lunch). Recognize and prepare for the impacts of data and ML/AI implementations: Swans and Shockwaves! https://mapr.com/blo...

For compliance, need self-explicating ML/AI models that can render plain-English explanations of why a particular classification/prediction/etc was made by a particular model from particular data under particular circumstances.

Meet regularly with your compliance officer to keep up to date with the changing regulatory rules. It can save you from wasting your effort and time

Data Governance and Compliance can be augmented and aided by ML/AI (Compliance Analytics for the CDO): http://www.kirkborne...

@neilmendelson Algorithms can be subjected to legal discovery. And that's why "explicable AI" initiatives need to render explanations that are comprehensible to lawyers, judges, and juries.

@KirkDBorne That's right. Geo-awareness will be a big part of this. Where the data originated and where it's used determine regs to comply w

@jameskobielus Agreed. Explicating AI is one of the big UX challenges of the next 10 years

A key idea that will come repeatedly here is "Use data to make data". This is the second principle of data capital. For more http://bit.ly/2xkLP0...

Algorithms create observations about their own performance that can be fed back into the system to improve their future performance. This idea is baked into ML

Apple’s chip-based ‘neural engine’ in new iPhone X creates new opportunities for AI on mobile https://goo.gl/4wtSt...

ML/AI unlocks data's potential in several ways: class (pattern) discovery, correlation (trend) discovery, novelty (interestingness) discovery, and association (link) discovery -- the business then exploits those discoveries

this is all around us. For example when Spotify recommends a playlist to you or Amazon recommends a book to u. This is ML

We need a pithy way to describe that. If robots are for boring and dangerous jobs, what are AIs for?

Using usage data (user trails, breadcrumbs) as enriched metadata to reveal, expose, and annotate how data are being used enables a feedback loop for greater uses and value creation

Numerous and numbing?

+if one takes the offer, or not, is additional data to improve the next recomendation

ML/AI is the chief statistical approach for surfacing the insight-relevant patterns in complex data (correlations, classifications, predictions, anomalies, etc.) that might otherwise go undetected.

@virtualclaude That is an example of association discovery (in both the customer graph and the product graph)

Creating the 'closed loop' allows the algorithm to learn and improve

Unlike other kinds of capital, using data does not consume it. In fact, using it creates more data capital

Several of the comments here are examples of DataOps = iterative, fail-fast, learning & improvement (which is great since ML are algorithms that learn from experience!)

.@KirkDBorne Great point. @neilmendelson shares more in data sources explored – CUBEConversation - #theCUBE #oracle #bigdata https://video.cube36...

You will need a data lab the supports the 3 main players: Business Analyst, Data Engineer, and Data Scientist. They need a platform that can enable the blending of new raw data sets with existing curated data. Highly visual tools that leverage ML

Different kinds of storage -- key/value, object stores, relational are key, but they must work in concert

First and foremost: big data/ML/AI tools that simplify the effort to get to use cases. Way too much time spent piloting infrastructure; not enough time piloting business case.

In terms of open platforms, Hadoop, the various NoSQL databases, Apache Spark, the growing range of open-source ML libraries, and, now, the growing range of open-source deep learning tools (especially TensorFlow) are key.

Let's not forget the need to secure and govern all that data

Exploratory tools to cut down on the time and effort necessary to tell spurious correlations from valuable patterns

Begin your journey with specific measurable business outcomes in mind. Hear more about it on theCUBE https://youtu.be/MGR...

Platforms are converging ... we are seeing more microservices, APIs, containers, open source implementations = no need to build your own platform, but rather deploy PaaS: Platform-as-a-Service

What you need to exploit big data effectively with ML/AI is what I call "DevOps for Data Science" platforms to help teams build, train, deploy, evaluate, and iterate models...plus a data lake with strong data/model governance.

Too much time is being wasted by technologies who think that can add value to their business by building their own infrastructure

NoSQL and Data Lakes are big help, with exploratory tools like Apache Drill, enabling "day zero" integration and exploration of new datasets

The flipside of diverse storage is unified access, like SQL across them http://bit.ly/2wpXh7...

.@neilmendelson True and data requires infrastructure. Paul Sonderegger, Oracle shares what kind - In The Studio - #Wikibon Boston #oracle #bigdata https://video.cube36...

We increasingly see object stores in the cloud as data lakes

SQL remains the lingua franca of data

"3 Database admins walked out of a NoSQL bar because they couldn't select a table"

@neilmendelson Yes. A platform for curation of the data, as well as continuous training, testing, and iteration of the statistical models (eg., artificial neural networks, support vector machines, etc) that constitute ML/AI's intelligent heart.

NoSQL = Not Only SQL

As @neilmendelson says, don't forget the Data Engineer! That role is often overlooked with all the chatter about Data Scientists

@neilmendelson For sure. Data prep is 75% of the game when it comes to developing, training, and tuning ML models. And data cleansing/augmentation is the soul of prep.

We're seeing more companies set up a durable, immutable raw data lake. Then draw from it for different kinds of prep

@PaulSonderegger Don't you mean the other way around: "data lakes incorporate object stores," but are not, strictly, the same things as object stores?

Very true. good point

@PaulSonderegger Isn't that immutable raw data the heart of serverless computing environments? Is serverless cloud the nucleus of next-gen data lakes for AI/ML?

@jameskobielus Kind of like the aquifer. It's the big resource those server less functions tap into

@jameskobielus This is a little like the iron rule of digital film editing -- never touch the original