PMI CPMAI_v7 - Cognitive Project Management in AI CPMAI v7 - Training & Certification

Under CPMAI Phase III: Data Preparation, the Data Format task includes “Data anonymization” as a core activity to remove or mask PII when it is not required for modeling, thereby protecting privacy while retaining data utility.

=========

The CPMAI™ Glossary contrasts automation (for deterministic, rule-based tasks) with AI (for probabilistic, learning-based tasks). As a rule of thumb, if a problem exhibits probabilistic patterns that can’t be captured by fixed rules, then AI is the appropriate solution; deterministic problems are better handled by simple automation.

In the Model Operationalization phase (Phase VI), when targeting edge devices, teams must capture the specific constraints around connectivity and on-device performance. The CPMAI Workbook’s Edge Model Data Needs task group instructs project teams to list all requirements and constraints for running models on edge systems—including constraints on external data access—and to determine how models will be deployed locally so they function offline when connectivity is unavailable.

Under “Edge constraints,” the Workbook requires teams to “List constraints on model usage including … requirements for model result response time, constraints on external data access,” underscoring that reliance on a remote API is infeasible without connectivity.

Consequently, the critical operationalization decision is to deploy the model directly on the edge device, ensuring it remains fully functional regardless of internet availability.

=========

In Phase VI: Model Operationalization, CPMAI specifies that batch prediction—“generating predictions offline for large datasets on a scheduled basis”—is the ideal deployment mode when analysis is needed at regular, non-continuous intervals (e.g., weekly) . Continuous methods like real-time or streaming are overkill for periodic reporting.

In CPMAI’s examination of AI patterns, Unattended bots are designed to run autonomously in back-office environments without human supervision, executing repetitive tasks like data entry at scale. This contrasts with Attended bots, which require a user to trigger or interact with them in real time.

Thought for 13 seconds

Model iteration requires regularly updating a deployed model with new data and configuration. The CPMAI Workbook’s Task: Fine-Tuning / Re-training of Pre-Trained Models prescribes defining and documenting re-training pipelines as part of the model-building lifecycle to ensure seamless iteration and ongoing performance improvements.

Training deep neural networks on CPUs is very slow. CPMAI’s Glossary highlights that tensor processing units (TPUs) and GPUs are specialized hardware accelerators explicitly recommended to “accelerate the training and inference of machine learning models” by parallelizing the heavy matrix operations in neural-network layers. Switching from desktop CPU training to GPU-based training can reduce training time by orders of magnitude.

=========

CPMAI’s Usage of AutoML task instructs teams to “Document how AutoML tools will be used for model creation” and to verify that the output can be integrated into the overall I/O flow . While AutoML excels at automating algorithm selection, model selection, hyperparameter tuning, and even preliminary performance metrics, CPMAI places iterative modeling and evaluation squarely under the manual Model Evaluation phase—where teams must interpret results against business success criteria and decide on next steps. Entrusting that high-level, iterative decision-making to an AutoML black box would undermine the human-centric evaluation that CPMAI mandates.

=========

The CPMAI methodology’s Model Iteration Approach (Phase V) explicitly calls out that “models will need continuous iteration, especially if they are only marginally providing the desired results” and requires teams to “detail approach that will be used to iterate this model to improve on any of the results in this Phase” . Failing to include a model retraining pipeline means the model cannot adapt to new data distributions, leading to performance degradation over time.

=========

A spam filter must decide between exactly two categories—spam or not spam—making it a binary (or binomial) classification task. The CPMAI Glossary defines binary classification as “a classification task where data is categorized into one of two classes (e.g., spam vs. not spam).”

=========