Just like Santa needs his naughty or nice list on Christmas, organizations have recently noticed a need for a similar naughty or nice registry for AI models and use cases.   In today's rapidly evolving technological landscape, the reliance on models, particularly those powered by artificial intelligence (AI), has grown exponentially. These models drive decision-making processes across industries. However, as their influence expands, so too does the risk they pose. Misaligned assumptions, data quality issues, or unintended biases in models can lead to catastrophic outcomes—financial losses, reputational damage, and legal consequences. Model Risk Management (MRM) has thus emerged as an essential discipline to ensure that models operate as intended, align with regulatory requirements, and serve their intended purposes without causing harm.   The necessity for robust MRM practices stems from the inherent uncertainties and complexities in modeling. Models are approximations of reality, built on assumptions and trained using datasets that might not fully represent the environment in which they are applied. Therefore, it helps with efficiency, transparency and the speed of constant development that all vital information is captured in a centralized location so that the assumptions and risks of AI models and use cases are understood by all who want to utilize them. What has made MRM even more relevant today is the European Union's Artificial Intelligence Act (EU AI Act) that represents one of the most comprehensive regulatory frameworks aimed at governing AI systems. Introduced to establish trust in AI while ensuring compliance with ethical and legal standards, the Act emphasizes transparency, accountability, and risk management.   The EU AI Act classifies AI systems into categories of risk: unacceptable, high, limited, and minimal. High-risk AI systems, such as those used in critical sectors like critical infrastructure, healthcare, law enforcement, and finance, are subject to new requirements. Robust MRM practices can help with the compliance with the Act. Organizations must not only validate models during development but also monitor their performance and impact post-deployment. This ensures that risks are mitigated proactively and that systems remain compliant as real-world conditions evolve.   MRM is available as a solution on the SAS Viya platform. To learn more visit www.sas.com/fi_fi/software/model-risk-management.html or contact me at antti.heino@sas.com   Merry Christmas, Antti Heino Principal AI Advisor, SAS Institute ... View more

This Juletip explores different options for named entity recognition (NER). NER is a natural language processing (NLP) task that involves the identification and classification of entities, such as names of people, organizations, locations and other specific categories, within a body of text. The primary goal of NER is to extract useful structured information from unstructured text data or it can be used to remove that data to anonymize documents. If you want to dive deeper into the details and try these yourself, I recommend to check out this video I also made recently: https://youtu.be/HC6UJDeoddw   NER systems typically use a combination of machine learning algorithms and rule based approaches, to analyze and understand the context of words in a given text and assign them to predefined categories.   There are 4 typical options to combine to get best possible results. Each option has their own benefits and drawbacks so I recommend to use both machine learning and rule based approaches. From simple to more complex:   Rule based methods that utilize regex: SAS PRXMATCH / PRXCHANGE functions Rule based methods that utilize NLP & REGEX: SAS Visual Text Analytics concept rules Large language models (LLMs) such as BERT (millions of parameters) Recent generative LLM such as Llama 2 or other generative AI models (billions of parameters)   PRXMATCH / PRXCHANGE Benefits: Uses regex rules to find entities. It is especially useful for entity types that follow a particular pattern, such as social security numbers. It is quick and easy to implement. PRXCHANGE gives the options to anonymize the entity simultaneously so required code is minimized. Drawbacks: Does not understand context or language. Concept rules Benefits: Concept rules bring together language understanding and versatile rule types, including regex. These rules can take into account the context in a sentence or a paragraph. They can be used to pick up patterns, but also work well for different types of names. Drawbacks: It takes time to develop rules manually. It is easy to detect entities in their base form, but in some languages inflected word forms might take additional effort. Large Language Models, such as BERT Benefits: BERT models tuned for named entity recognition can be highly efficient at spotting entities and the model gives out also a probability score and entity type. Drawbacks: It is much harder to modify results of a LLM compared to rule based methods. It might not be feasible to do additional training to spot entities that are missed by the model so rule based methods should be still used. Generative Large Language Models, such as Llama 2 Benefits: LLMs can generate answers to wide variety of queries. The new generative models have been trained on massive amounts of text and can detect entities well while understanding context and nuances of language. Drawbacks: Generative models can produce false results known as hallucinations. It is also possible that the positions of the recognized entities are incorrect. They can also refuse to detect some entity types such as social security numbers because of their safeguards. It might require prompt engineering to get the desired result. The required compute is also on a different scale compared to the other options so I would recommend to first try the other options as pure named entity recognition is not necessarily generative.   If the topic interests you further, check out the linked video for details. It provides practical examples on how to get started with combining these options on SAS Viya.   Links to SAS documentation: VTA Concept rule documentation: https://go.documentation.sas.com/doc/en/capcdc/v_022/ctxtcdc/ctxtug/p1kf71w7npr9ecn1gysvovfs42x2.htm PRX documentation: https://go.documentation.sas.com/doc/en/pgmsascdc/v_045/lefunctionsref/n0r8h2fa8djqf1n1cnenrvm573br.htm   Happy holidays! ... View more

Seamless digital customer experiences are essential for banks that want to remain competitive. Customers increasingly expect real-time interactions and decisions, but making it happen might be a challenge without the right technology. In this video I show how an "AI Bank" can use hyperautomation to create seamless digital customer experiences for the bank's customers and how many individual components can be integrated with decisioning technology into various customer journeys. The showcased integrations are examples and can be replaced depending on the needs of regulations and the preferences of banks.   The video shows the power of SAS analytics combined with our Intelligent Decisioning technology from the end customer point-of-view when they are using the bank's website. Loan origination and onboarding journeys are completed in real-time with the help of predictive models, fintech integrations and rule-based checks.   ... View more

Santa needs a lot of helpers to bring presents to everyone around the world. The same way users of SAS Viya analytics platform can benefit from helper bots so that they can get data and analytical results faster to everyone in their organization. SAS Viya always includes the Conversational Designer functionality which allows you to build chatbots to interact with the platform and provide capabilities of Viya through a chat interface.   You can build many types of bots, but in this Juletip I want to focus on a simple helper bot that can perform data preparation tasks for anyone using Viya.   The first step is to create a new bot in Conversational Designer. In order to understand what a Viya user wants to do, we have to define an intent for the bot. Then we can give the bot few examples of different ways users can express their intent (uttrances) and finally we have to map that intent to a dialogue flow.   1. Create a new intent of "Join tables". Add some example utterances and save the most important words as entities. Defining intents   2. Start building your dialogue flow. The first node should map to the user intent that triggers the dialogue flow (=the "Join tables" -intent). Creating dialogues   3. After the dialogue has been triggered by the intent, the bot should answer something back to the user. In this case, we ask the user to provide the name of the first data table they would like to include in the join. Defining responses   4. We record the user input to parameter variables (in this case named "firsttable" that we later use to execute the join code. In similar way, we can ask the user for the name of the second table they want to include to the join and also the key variable that should be used. Depending on the data preparation step you want the bot to execute, the variables you need to ask from a user can change. Setting input parameters   5. When all the required parameters have been given by the user, the join can be executed using the SAS Code Data Provider step. I have included the join table example code below that uses 3 parameters that are required from the user. The code assumes that the source tables are found in CASUSER library and the bot puts the resulting joined table into the CASUSER library as well. Of course if you want to make your bot more versatile, these could be filled by parameters as well so that the user has more control over the join. SAS Code Provider     cas CASAUTO; caslib _all_ assign; %global tableone; %let tableone=&firsttable; %global tabletwo; %let tabletwo=&secondtable; %global jkey; %let jkey=&joinkey; proc sql; create table final_table as select * from casuser.&tableone as x left join casuser.&tabletwo as y on x.&jkey = y.&jkey; quit; data casuser.joinedtable (promote=yes); set work.final_table; run;     6. After everything has been configured, it is possible to test the bot using the "Try It Now" button. Test your bot   7. The resulting table can now be found in your CASUSER library. Congrats, you have now developed your first dialogue for your helper bot! 7 Bot has created a new table as a result of the join 8. Typically after the bot has been tested, it will be deployed on a web server and integrated on a website, a Visual Analytics report or MS Teams.   Thanks for checking out this SAS Viya Conversational Designer Juletip! Merry Christmas and happy New Year! ... View more

Santa needs to decide what presents everyone gets. That is a momentous task which might require an efficient machine learning (ML) model. Factorization machines fit this description as they can handle large data with high volumes of missing values (sparse data).   This ML model can learn preferences of people to recommend any products such as movies or books. The purpose is to predict how a person would react to a set of products, and then recommend products that the person is likely to prefer the most.   Let's look at an example about books as they are a common xmas present. Typically, people are more likely to buy a book if their predicted rating for the book is higher than 4.5. Therefore, we predict ratings for multiple books and recommend the ones that the person is likely to rate 4.5 or above. If rating data would not be available to Santa, other indicators of popularity like pageviews in online stores could be used instead.   Santa's database could look like the table below. Each person has read and rated some books, but most observations are empty. This type of data can be challenging if there are a lot of products to choose from as that causes also a lot of empty observations. However, factorization machine is built for this type of sparse data and can calculate the predictions for unrated books based on the available data.     Book A Book B Book C Book D Person 1 3 5 - - Person 2 - 3 1 - Person 3 4 - 3 5 Person 4 5 3 - -   The Factorization Machine procedure estimates factors for each of the predictors (Book and Person), in addition to estimating a global and a level bias. After specifying the target variable (Book Rating), the procedure computes the biases and factors by using the stochastic gradient descent algorithm, which minimizes the prediction error during the learning process.   As a result of training the model, Santa gets statistics how well the predicted ratings match the actual values overall. The trained model seems to be quite good at predicting the ratings and it can now be used to predict ratings that a person would give to the selection of books Santa has available this year. Santa can then easily select the books that have the highest predicted rating and are not yet owned by the person.   Predicted vs Actual Ratings ... View more