Arrivaattral inspiration
Arrivaattral can be equated to a state, like a state of Arrivaattral. In a simplified view of this state, with respect to our focus area of multiple data science models, we seek features that enable evolving cognitive capabilities, primarily focusing on how knowledge is represented and organized, frameworks that allow reasoning with knowledge using predicate and first order logic, etc., and rely on nondeterministic orchestrations that adapt with experience, including refinements based on reviews from explanations, etc. In a typical data science ecosystem, models evolve, phase out, induct newer ones, and employ additional mechanisms to combine the knowledge, etc., and the Arrivaattral state requires that we consider frameworks such as symbolic processing, etc., and handle them. Aided by this wide inspiration, let's look at our IROPS example through the Arrivaattral lens.
Another Business Problem Definition
We covered an example on how airlines can recover from irregular operations (IROPS) using a consistency, collaborative and transparent framework in an earlier discussion. Please see https://www.linkedin.com/posts/prabhu-saiprabhu-sai-79b610_ai-driven-collaboration-of-multiple-data-activity-7010744021841047552-OlmM?utm_source=share&utm_medium=member_desktop for details.
In this discussion we will turn our attention to another major area - Supply Chain Management (SCM). The components that were used in the IROPS scenario will also be used for the SCM focus. However, the behavior of these components is tailored to SCM needs. We will use a restaurant business as the SCM user. SCM is a vast field in itself and one can find many good discussions elsewhere. So, we will focus on a few aspects related to SCM from the restaurant businesses perspective, to illustrate the benefits of Arrivaattral capabilities.
Consider the following business imperatives from a restaurant business irrespective of its size or scale.
Process View from our 4+1 Architecture standard
For simplicity sake we will only use some views in the 4+1 architecture framework. The following schematic depicts the process view.
Using the same platform we considered for IROPS use case, the restaurant SCM use case environment is depicted. Outside of the Arrivaattral platform, we now have a different set of applications and models. They produce similar constructs like, knowledge, prescriptions, etc. That also adhere to the same requirements, such as timeliness, responsiveness and proactiveness criteria. In this use case, we do introduce a domain knowledge mechanism to add to the knowledge that is gained through model prescriptions. This mechanism is also a route to introduce business imperatives that may either be a constraint or a conflict resolver when multiple prescriptions compete. Our knowledge representation framework is well suited for such domain knowledge inputs. Keep in mind that any knowledge should be amenable for change, even retirement or temporary pause, etc. This capability in the right environment can also be used for several what-if scenarios.
So, let’s jump right into outcomes using Neo4J graphs and describe the actions and prescriptions from the environment.
Here is a view of our menu items and ingredients. We have in all, 5 menu items that need 47 ingredients, many are common to the menu items with a few ingredients only used in some. So, in our data we have 52 things we deal with that are interconnected in 85 ways. So, the Neo4J graph view shows 52 nodes and 85 relationships as marked.
Let’s look at a menu item as shown below. The added yellow box indicates a few attributes about the menu item in consideration – Ravioli Carbonara. The customer affinity score for this item is listed as 92. There are two capacity measures tied to this item, one that is projected based on our data science model prescriptions and the other one that has been determined based on supply chain processes that this restaurant uses. In this example, we are close as we have one less procured capacity than that is predicted.
Now let’s look at another menu item – Lasagna Classico. We do have a problem. This menu item has received a rating of 99, with our data science models predicting a capacity of 24 to plan for the night, while the restaurant is unable to procure any capacity.
One of the root causes is shown in next image, where we are looking at the ingredients. Eggs seem to be not available as shown in our database; the restaurant just has 3 measures in their stock, with none available in the market.
Further digging in, we also find that we need 10 measures of this ingredient for the menu item from the RECIPE relationship and so they are unable to deliver on this menu item.
Depending on how early this process is done prior to the day in planning, the restaurant should be able to help and deal with this situation, and have better control of meeting the business objectives.
Turning to a few additional factors, let’s say that this restaurant leverages two chefs, just to show how Arrivaattral type systems can handle complex relationships. As the Neo4J graph below shows, we can see how the two chefs view the 5 menu items. We called this relationship as PREFER, where each Chef prefers or can add their special touch or in some way have a say in how the menu item is dealt with. As shown, Chef C1 has a preference to Lasagna Classico, while Chef C2 does not have a preference to that menu item; this is not necessarily a real-life or a bad scenario, but allows us to view the collaborative behavior that can be implemented. So, how strong of a preference Chef C1 has towards Lasagna Classico?
The PREFER relationship has been designed with such collaboration in mind. It does have an order and as shown below, where indicates a rating of 1, while the Five Cheese Ziti Al Forno stands at 3 on the preference order.
So, Arrivaattral state systems, by allowing consistent and collaborative capabilities can help such SCM clients be very effective.
Details that are shown using Neo4J graphs are same or a simplified view of what we have in our knowledge base. When we extend the knowledge base with additional facts, such as seasonality, region, etc., that may be fed as inputs into our data science models, we implement the transparency criteria, where the internal decision points of data science models are chained.
What next?
In the next paper we will use specific AI techniques to implement the transparency behavior. State of Arrivaattral requires capabilities to improve or select appropriate decision frameworks and so explanations are essenoviding means to accomplish this. Handling explanations to refine decision frameworks is not straightforward and is an evolving area.
There are no readily available frameworks to assist with our explanations need. We use areas of machine learning such as Explanation Based Learning and Reinforcement Learing to make a difference.
Let the fun times continue!