Machine Learning

CPMAI · 26 questions

• Meridian Regional Airlines' data science team is building a model that reads sensor readings from an aircraft component and outputs one of two labels: "likely to fail within 30 days" or "not likely to fail." The training data includes past component records already tagged with the outcome that actually occurred. Which type of algorithm is Meridian's PM overseeing here?
• Meridian's loyalty-fraud team combines the votes of several different models — a decision tree, a logistic model, and a gradient-boosted model — into a single fraud/not-fraud verdict for each mileage-redemption transaction, rather than relying on any one model alone. What technique is the team using, and why would a PM want it here?
• Meridian's network-planning team wants to group its ~120 routes into segments with similar demand-seasonality patterns, but they have no pre-existing labels defining which routes "should" belong together. Which type of algorithm fits this need?
• Meridian's crew-and-gate scheduling team is prototyping a system that repeatedly tries different gate assignments in a simulated version of a busy hub, receiving a numeric reward each time based on minimized taxi time, fewer connection misses, and no FAA duty-time violations, and gradually improving its assignment strategy over many simulated days. What kind of system is this, and what are its two core components a PM should be able to name?
• During a review of the gate-scheduling reinforcement learning project, a Meridian ops director asks why the data science team keeps letting the system try unusual, seemingly worse gate assignments instead of always picking the best-known assignment so far. What concept should the PM use to explain this design choice?
• Meridian's computer-vision team proposes a baggage-mishandling detector that achieves 97% overall accuracy in testing. The PM reviewing the results notices the model still misses a meaningful share of true mishandling events (false negatives), even though its accuracy score looks strong. What should the PM do?
• Meridian's virtual assistant needs to route each incoming customer message into one of four known intents — rebooking, baggage status, loyalty questions, or check-in — using thousands of past messages that support agents have already tagged with the correct intent. Which approach best fits, and why is it not clustering?
• At a Go/No-Go review for the ramp-safety camera system, a Meridian union representative asks the PM to explain, in plain terms, how the neural network decides a ground crew member has entered an unsafe zone. Which explanation correctly describes the basic structure the PM should reference?
• At a Go/No-Go review for the ramp-safety camera system, a Meridian union representative asks the PM to explain, in plain terms, how the neural network decides a ground crew member has entered an unsafe zone. Which explanation correctly describes the basic structure the PM should reference?
• Meridian's data science team recommends a neural network over a hand-written rules engine (e.g., "flag if pixel-brightness delta exceeds X near a marked zone") for the ramp-safety camera system. What is the core justification a PM should expect to hear?
• Meridian's PM notices the same general neural-network approach is being proposed both for cargo-demand forecasting (a numeric prediction) and for ramp-safety violation detection (a category decision). Is this a red flag, and why or why not?
• Meridian's dispatch team is asking the data science team to move beyond a shallow one-hidden-layer network to a deep architecture with many hidden layers for interpreting complex, multi-sensor engine-vibration patterns tied to developing faults. What is the PM-relevant justification for adding depth?
• Meridian's engineering team is comparing two deep learning approaches for two different projects: recognizing foreign-object debris in ramp camera footage, and modeling the trend in a time-series of turbine vibration readings over the past 90 days. Which pairing correctly matches architecture family to problem for a PM sizing the two efforts?
• Meridian has a small predictive-maintenance dataset — only 400 labeled failure events collected over several years for one aircraft type. The data science team proposes a very deep, high-capacity neural network. What should the PM raise as a concern?
• Meridian wants a system that maintenance and dispatch staff can ask natural-language questions of, drawing on the airline's maintenance manuals and dispatch procedures, and receive a synthesized, readable answer rather than a list of document links. Which technology is an appropriate fit, and why?
• During testing, Meridian's generative-AI dispatch assistant confidently states a specific regulatory requirement that turns out not to exist in any actual FAA or Meridian document. What should the PM take away from this, and what should the project do about it?
• A union representative at Meridian, unfamiliar with AI, asks the PM to explain "in plain terms" how the maintenance-manual chatbot actually produces its answers. Which explanation is both accurate and appropriate for the PM to give?
• Meridian's maintenance-manual assistant gives inconsistent answers to nearly identical questions from different technicians, and the data science team has not yet retrained or fine-tuned anything. What is the appropriate first step the PM should direct the team to try?
• After exhausting prompt-engineering improvements, Meridian's data science team still finds the general-purpose base LLM struggles with the airline's specific maintenance terminology and internal procedure codes. What is the appropriate next step, and what should the PM understand about it?
• For the dispatch-procedure assistant, the PM directs the team to design it so it always drafts a suggested answer with source citations for a human dispatcher to review and approve before any procedure change is acted on, rather than letting the model act autonomously. What principle does this design reflect?
• Meridian's customer-service chatbot, built on a general-purpose LLM, is asked by a passenger whether their specific flight departing in two hours is currently on time. The chatbot has no live connection to Meridian's operational systems. What limitation does this expose, and what does it imply for the project?
• During the training phase of the cargo-demand forecasting model, the PM's data science lead asks whether the PM needs to personally review each hyperparameter setting before training runs. What is the PM's appropriate role during this phase?
• To speed up development of the predictive-maintenance neural network, Meridian's data science team proposes starting from a pre-trained model already trained on general sensor/vibration data, then adapting it to Meridian's specific aircraft components, rather than training an entirely new network from scratch. What is this technique, and why does a PM care?
• Meridian's predictive-maintenance team uses one cloud vendor's ML platform, while the cargo-demand-forecasting team independently adopted a different open-source ML toolkit, and the two teams' outputs and data formats are incompatible with each other. What does this situation illustrate, and what is the PM's responsibility?
• After the platform-fragmentation issue surfaces, Meridian's PM proposes a shared feature store and a common data/model versioning environment so the predictive-maintenance and demand-forecasting teams stop working from disconnected spreadsheets and ad hoc scripts. What is this PM action an example of?
• Meridian is evaluating two cloud ML platform vendors for the ramp-safety computer-vision system, which needs GPU-accelerated real-time inference and the option to run inference at the edge (on-site, near the cameras) due to connectivity constraints on the ramp. One vendor is cheaper but offers only cloud-hosted batch inference with no edge deployment option; the other costs more but supports real-time GPU inference and edge deployment. What should guide the PM's platform decision?