CTO Emerging Technologies & Innovation 5 â Questions and Answers
Question 1: A CTO is evaluating graph neural networks (GNNs) for a new use case. Which problem type represents the STRONGEST fit for GNN-based approaches compared to traditional deep learning?
- Sequential time-series forecasting for stock prices
- Fraud detection in transaction networks with complex relationship patterns (Correct answer)
- Image classification with high-resolution satellite imagery
- Natural language summarization of long documents
Correct answer: Fraud detection in transaction networks with complex relationship patterns
GNNs excel at learning from relational graph structures, making them ideal for fraud detection where entity relationships and network topology carry predictive signal.
Question 2: In the context of sustainable technology strategy, what does a CTO primarily optimize for when adopting 'carbon-aware computing' scheduling?
- Minimizing total CPU cycles used by batch workloads
- Shifting flexible workloads to times and regions with higher renewable energy availability on the grid (Correct answer)
- Replacing all on-premises hardware with carbon-neutral cloud providers
- Measuring Scope 1 emissions from on-premises data center cooling systems
Correct answer: Shifting flexible workloads to times and regions with higher renewable energy availability on the grid
Carbon-aware computing schedules flexible workloads to execute when and where the electrical grid has the highest proportion of renewable energy, reducing carbon intensity.
Question 3: A CTO reviewing proposals for a new data mesh architecture encounters a team planning centralized data governance with distributed data ownership. Which tension does this reveal?
- Data mesh requires fully decentralized governance, making centralized policies incompatible
- Data mesh balances federated computational governance (global standards) with distributed domain ownershipâcentralized control of standards is expected (Correct answer)
- Data mesh eliminates the need for any data governance policies at the domain level
- Data mesh requires all domains to use the same data storage technology stack
Correct answer: Data mesh balances federated computational governance (global standards) with distributed domain ownershipâcentralized control of standards is expected
Data mesh uses federated computational governanceâglobal standards and policies enforced via platformsâwhile domains own their data products, so centralized policy governance is by design.
Question 4: When a CTO evaluates adopting a 'composable architecture' strategy, what is the primary business driver compared to traditional monolithic enterprise applications?
- Reducing software licensing costs by consolidating vendor relationships
- Enabling rapid assembly and reconfiguration of business capabilities through loosely coupled packaged business capabilities (PBCs) (Correct answer)
- Standardizing all business applications on a single cloud provider's native services
- Eliminating the need for API management infrastructure
Correct answer: Enabling rapid assembly and reconfiguration of business capabilities through loosely coupled packaged business capabilities (PBCs)
Composable architecture lets businesses rapidly reconfigure packaged business capabilities (PBCs) to respond to market changes without rebuilding monolithic systems.
Question 5: A CTO is piloting a robotic process automation (RPA) program. Which scenario represents the highest-value, lowest-risk starting point for RPA implementation?
- Automating customer-facing chatbot responses that require contextual judgment
- Automating high-volume, rule-based data entry tasks between systems that lack native APIs (Correct answer)
- Replacing the core ERP system's business logic with RPA bots
- Automating exception handling workflows that require frequent human escalation
Correct answer: Automating high-volume, rule-based data entry tasks between systems that lack native APIs
RPA delivers highest ROI on repetitive, rule-based data tasks between legacy systems where APIs don't exist, with minimal risk from process variability.
Question 6: Which approach best describes 'chaos engineering' as a discipline for improving resilience in a cloud-native system managed by a CTO?
- Randomly shutting down production systems during peak traffic without notification to test team response
- Deliberately injecting controlled failures in production or staging to discover system weaknesses before they cause unplanned outages (Correct answer)
- Stress testing by increasing load beyond capacity limits during planned maintenance windows
- Removing monitoring and alerting to assess if teams detect failures organically
Correct answer: Deliberately injecting controlled failures in production or staging to discover system weaknesses before they cause unplanned outages
Chaos engineering deliberately introduces controlled failures under scientific conditions to identify weaknesses before they manifest as unplanned customer-impacting incidents.
Question 7: A CTO is evaluating homomorphic encryption for a multi-party healthcare analytics use case. What is the PRIMARY limitation that affects its near-term production feasibility?
- Homomorphic encryption only supports symmetric key cryptography, limiting key management options
- Computational overhead is orders of magnitude higher than plaintext computation, making real-time analytics impractical today (Correct answer)
- Homomorphic encryption cannot operate on floating-point numerical data
- It requires all parties to share private keys, defeating its security purpose
Correct answer: Computational overhead is orders of magnitude higher than plaintext computation, making real-time analytics impractical today
Fully homomorphic encryption (FHE) currently carries 1,000xâ1,000,000x performance overhead compared to plaintext operations, limiting it to offline or non-time-sensitive analytics.
CTO Emerging Technologies & Innovation 5
A CTO is evaluating graph neural networks (GNNs) for a new use case.
Which problem type represents the STRONGEST fit for GNN-based approaches compared to traditional deep learning?