“Bring AI to Data” is a term I recently heard during the Oracle AI World in Singapore last week and it really caught my attention. It sounded simple but the idea behind it is quite powerful. So I thought it’s worth exploring a bit more!
Normally, working with data and
AI meant one thing, which is, moving data around. We would extract data from
databases, send it to external tools or platforms, build machine learning
models and then push the results back into the database. But this approach adds
complexity, increases costs and introduces security risks.
But now, Oracle is changing that
model by bringing AI to where the data already is!
The concept of “Bring AI to Data”
is straightforward but powerful. Instead of moving large volumes of data across
systems, Oracle allows you to run AI and machine learning directly inside the
database. This means that data do not have to leave its secure environment. This
results faster processing, reduced data duplication, improved security and a
much simpler architecture. For organizations dealing with sensitive data, such
as banks, healthcare organizations and government institutions, this approach is especially
valuable.
Lets consider a simple banking
scenario where we want to predict customer churn. Traditionally, we would
extract customer data, move it to a machine learning platform, train a model
and then integrate the predictions back into our system. With Oracle we can
create and run the model directly using SQL.
For example, we can build a
churn prediction model using a built in package as below.
BEGIN
DBMS_DATA_MINING.CREATE_MODEL(
model_name => 'CHURN_MODEL',
mining_function => DBMS_DATA_MINING.CLASSIFICATION,
data_table_name => 'CUSTOMER’,
case_id_column_name => 'CUSTOMER_ID',
target_column_name => 'CHURN_FLAG'
);
END;
/
Once the model is created,
generating predictions is just another SQL query.
SELECT
CUSTOMER_ID,
PREDICTION(CHURN_MODEL USING *) AS
CHURN_PREDICTION,
PREDICTION_PROBABILITY(CHURN_MODEL USING
*) AS PROBABILITY
FROM CUSTOMER;
As per above example, in just a
few lines, we can implement AI without exporting a single row of data. This
simplicity is what makes Oracle’s approach much practical for real world
systems.
Another useful example is
building recommendation systems. Whether it’s suggesting products in an ecommerce
platform or recommending financial services to customers, Oracle allows us to
generate recommendations directly within the database. Using built in models
we can produce recommendation sets with simple queries, without the need
for external processing engines.
Oracle also supports integrating
Python with the database, thus giving us the flexibility to use popular machine
learning libraries while still keeping data nearer. For eg, we can load data
into a Python environment directly from Oracle, train a model using libraries
and store the model back in the database. This approach allows us to combine
the familiarity of Python with the performance and security of Oracle.
One of the biggest advantages of
this approach is security. When data stays within the database, there are fewer
risks associated with data movement. There are no intermediate files, no
unnecessary copies and fewer integration points that could be exposed. This
makes compliance with data protection regulations much easier as well.
Performance is another key
benefit. Since AI models run where the data is, there is no network latency or
overhead from data transfer. Oracle databases are already optimized for high performance
processing and by executing AI workloads within the databases, organizations
can take full advantage of parallel processing and optimized query execution.
In real world, this approach can
be applied to fraud detection, customer segmentation, predictive maintenance etc.
Financial institutions can run fraud detection models in real time on live
transactions. Retail companies can segment customers and personalize marketing
campaigns. IoT systems can analyze sensor data directly within the database to
predict equipment failures before they happen.
What makes Oracle’s “Bring AI to
Data” approach stand out? It is its simplicity!
Instead of building complex data
pipelines and managing multiple tools, everything happens in one place. This
reduces operational overhead and allows us to focus more on insights rather
than infra.
So, “Bring AI to Data” is
a smarter way of thinking about data and analytics. By keeping data in the same place
and bringing AI to it, organizations can build faster, safer and more efficient
solutions.
Comments
Post a Comment