엣지 AI 퀴즈

퀴즈를 풀면 정답과 설명이 나타납니다.

1. Which deployment is an example of edge AI?
   • Processing close to where data is collected or used, such as a phone or local gateway
   • Training a model only on a central server
   • Using a larger model than the cloud
   • Disabling all connectivity
2. An application must work offline. What should its design include?
   • An assumption that every service is local
   • A useful failure mode for functions that still need remote retrieval
   • A larger model by default
   • A permanent network connection
3. An edge feature has 2 GB available, 1.2 GB of weights, and 0.6 GB of temporary buffers. What remains before other feature needs?
   • 0.8 GB, because only weights count
   • 1.4 GB, because weights use no runtime memory
   • 0.2 GB, leaving little margin for other needs
   • 2 GB, because buffers are outside the budget
4. A short device benchmark looks good. What could it miss?
   • Only the original model download time
   • Only the peak arithmetic rating
   • Only the model weight-file size
   • Thermal throttling and competition with other applications over time
5. After quantizing a model to fit a device, what should be checked?
   • Only the new file size
   • Task performance on difficult inputs and intended environmental conditions
   • Whether the model loads once
   • A benchmark from another device
6. A phone performs local inference but syncs logs to a server. What does this imply for privacy review?
   • Privacy review is unnecessary
   • Local inference means no data leaves the device
   • Logs and synchronization still need controls
   • The app is fully offline
7. Why does the guide recommend versioning models?
   • To identify which model version ran on each device
   • To prevent all updates
   • To hide device resource use
   • To change the battery capacity
8. What should an update plan include for intermittently connected devices?
   • Deleting the previous version before validation
   • A requirement that devices never reconnect
   • An unverified moving model
   • A safe update or rollback path