This year at Ignite was a little different in that it was virtual, but it was the same in the increase in capabilities brought from a widely diverse set of areas. Not surprisingly, huge amounts of effort has been placed in advancing Artificial Intelligence capabilities both in commodity AI (Cognitive Services) and custom AL/ML (Azure ML, Databricks, etc.). I appreciated how during the keynote Satya was showed the areas that AI has advanced over the last several years.
Capability 1: Azure Cognitive Services Metrics Advisor
A very interesting cognitive service that was released this year was metrics advisor, a service that helps with prescriptively finding outliers and issues in a set of data coming from a monitored source. You can check more about it here. This brings a number of great capabilities to our organization:
This was typically a more complicated process that required custom ML work to find the anomalies. The goal with Metrics Advisor is to simplify the output against the incoming data and accelerate time to value. You can see an example below of data auto-generated to help find opportunities for improvement.
Does it eliminate the need for custom ML? No, but it does give options to bring value quickly, which custom ML can then be built on top of. You can see a detailed breakdown of one such anomaly from the platform which we can dig into.
Capability 2: Azure Cognitive Services Spacial Analysis
The next capability is Spacial Analysis, which is oriented around understanding the locations of individuals in a facility for purposes of safety and revenue. In the cases of safety it’s around current problems like social distancing. In the cases of revenue it might be around dwell time, wait time, areas of interest, etc. You can see an example of this analysis below in the visualization of the spacial distance of the individuals:
You can see here the description of wait time with an intention of both spacing and how well our backlog is working. For instance, do we have enough checkout people, are purchases being made during the checkout process, what is the dwell time?
Here is another look at this implemented for the purpose of Social Distancing and the visualization of such:
An obvious concern with Spacial Analysis is privacy protection for individuals in public spaces. I was impressed with the clear retention rules and usage requirements that Microsoft placed around this service. They don’t retain the images, only the output that is non-personally identifiable.
Capability 3: Custom Vision
Even more work has been done in custom vision this year, including the efforts surrounding remaining capabilities. For instance, another capability of integrating custom vision with low/code no/code technologies in Power Apps. In an organization intending to do quality management based on completed work, cognitive services can review that completed product against what it *should* look like, then provide feedback on whether the job was completed successfully.
Here is an example of training that model with correct or non-correct images, such as the flat tire below:
Here we have an example of feeding that into a Low/Code No/Code interface, allowing the quick creation of this interface (in this case at Toyota), talked about in the Ignite keynote.
Capability 5: Smart Narratives in Power BI
Microsoft has started to build in automated AI into core interfaces used to visualize and review information. At Ignite an announcement was made surrounding “Smart Narratives”, which includes the AI platform analyzing data surfaced through Power BI and then providing insights based on quick analysis of that data. This might not be perfect, but it represents how over time we’ll see AI continue to be a partner for a human worker and inform it with ever-improving insights about the data being viewed.
You can see above on the left the Power BI visualization and on the right the AI feedback based on reviewing the data.
Capability 6: Responsible AI
I find something Microsoft has done well is to drive responsible AI in everything they do. Satya talked about how technology needs to be driven with purpose. This includes core principals that we are building something purposeful and meaningful.
- Fundamental rights
The core ideas surrounding Ethical AI include:
The core AI ethical principals include:
Capability 7: Updated Azure AI Designer
Easing the capabilities tied to building AI models and understanding the data feeding it, Microsoft has updated it’s Azure ML AI designer. This provides a visual framework to build ML models, clean them, and understand them visually.
Cleaning data and visualization:
Creating models visually:
For more on all of these check out Eric Boyd’s post here. What a great time to be in AI and Machine Learning!