Mobile Devices practice questions

A team is deploying a deep learning model that uses a convolutional neural network (CNN) for image recognition. The model achieves high accuracy but is very slow to infer on edge devices. Which THREE optimization techniques should the team consider to speed up inference without significant accuracy loss? (Select three.)

Trap 1: Use larger convolutional filters (e.g., 7x7 instead of 3x3) to…

Larger filters increase computation and slow down inference.

Trap 2: Increase the number of convolutional layers to improve feature…

More layers increase computational cost and latency.

A
Use larger convolutional filters (e.g., 7x7 instead of 3x3) to capture more context.
Why wrong: Larger filters increase computation and slow down inference.
B
Use weight pruning to remove unnecessary connections in the network.
Pruning reduces computation and memory footprint.
C
Implement knowledge distillation by training a smaller model to mimic the larger one.
Knowledge distillation creates a compact model that retains much of the original accuracy.
D
Increase the number of convolutional layers to improve feature extraction.
Why wrong: More layers increase computational cost and latency.
E
Apply model quantization to reduce weight precision.
Quantization reduces model size and speeds up inference, often with minimal accuracy loss.

Question 2easymultiple choice

An organization is deploying an AI model on edge devices with limited computational resources. Which model optimization technique is most appropriate?

Trap 1: Perform additional feature engineering

Feature engineering is a training step.

Trap 2: Use an ensemble of models

Ensemble increases size and latency.

Trap 3: Increase the training dataset size

More data doesn't reduce computational requirements.

A
Perform additional feature engineering
Why wrong: Feature engineering is a training step.
B
Apply model quantization
Quantization reduces precision, making models smaller and faster.
C
Use an ensemble of models
Why wrong: Ensemble increases size and latency.
D
Increase the training dataset size
Why wrong: More data doesn't reduce computational requirements.

Question 3mediummulti select

Which TWO techniques should be considered when optimizing a deep learning model for deployment on edge devices with limited computational resources?

Trap 1: Apply adversarial training

Adversarial training improves robustness but does not reduce resource usage.

Trap 2: Use a GPU for inference

GPUs are not always available on edge devices.

Trap 3: Increase the number of layers

Adding layers increases complexity and resource usage.

A
Apply adversarial training
Why wrong: Adversarial training improves robustness but does not reduce resource usage.
B
Model quantization
Quantization reduces memory and computation requirements.
C
Use a GPU for inference
Why wrong: GPUs are not always available on edge devices.
D
Knowledge distillation
Distillation produces a smaller, faster model while retaining accuracy.
E
Increase the number of layers
Why wrong: Adding layers increases complexity and resource usage.

Question 4hardmultiple choice

A streaming data pipeline ingests sensor data from IoT devices. The data arrives at irregular intervals and contains occasional spikes. Which data transformation is most appropriate for preparing this data for a time-series model?

Trap 1: Removing all rows with values outside 3 standard deviations

Removing spikes may eliminate valid extreme events and does not address irregular intervals.

Trap 2: Using a sliding window to compute moving averages

Moving averages smooth data but do not resample to a fixed frequency; irregular intervals remain.

Trap 3: Padding missing timestamps with zeros

Zero padding introduces artificial values that can distort the time-series pattern.

A
Downsampling to a fixed frequency using mean aggregation
Mean aggregation over fixed intervals handles irregular timing and reduces noise.
B
Removing all rows with values outside 3 standard deviations
Why wrong: Removing spikes may eliminate valid extreme events and does not address irregular intervals.
C
Using a sliding window to compute moving averages
Why wrong: Moving averages smooth data but do not resample to a fixed frequency; irregular intervals remain.
D
Padding missing timestamps with zeros
Why wrong: Zero padding introduces artificial values that can distort the time-series pattern.

Question 5hardmultiple choice

An AI model is deployed to a mobile app with limited computational resources. The model is a deep neural network with high latency. Which technique is best to reduce inference time?

Trap 1: Increase batch size

Batch size affects throughput but not the latency of a single inference request.

Trap 2: Add more layers

Adding layers deepens the network and increases inference time.

Trap 3: Use a larger model

A larger model would increase computational requirements and latency.

A
Increase batch size
Why wrong: Batch size affects throughput but not the latency of a single inference request.
B
Add more layers
Why wrong: Adding layers deepens the network and increases inference time.
C
Use a larger model
Why wrong: A larger model would increase computational requirements and latency.
D
Quantization
Quantization reduces model size and speeds up inference by using lower-precision arithmetic.

Question 6hardmultiple choice

A company deploys a deep learning model for real-time object detection in autonomous vehicles. The model was trained on high-end GPUs but needs to run on edge devices with limited computational resources. Which technique is most effective for reducing model size and inference latency while maintaining acceptable accuracy?

Trap 1: Hyperparameter tuning

Hyperparameter tuning optimizes performance but does not directly reduce model size or inference latency.

Trap 2: Batch normalization

Batch normalization speeds up training and adds some regularization but does not reduce model size for inference.

Trap 3: Dropout

Dropout is a regularization technique used during training; it does not affect inference size or speed.

A
Hyperparameter tuning
Why wrong: Hyperparameter tuning optimizes performance but does not directly reduce model size or inference latency.
B
Batch normalization
Why wrong: Batch normalization speeds up training and adds some regularization but does not reduce model size for inference.
C
Dropout
Why wrong: Dropout is a regularization technique used during training; it does not affect inference size or speed.
D
Quantization
Quantization reduces numerical precision, shrinking model size and improving inference speed.

Question 7easymultiple choice

A team is deploying a deep learning model for real-time image classification on edge devices with limited computational resources. Which technique would best help reduce model size and inference time without significant accuracy loss?

Trap 1: Data augmentation

Data augmentation increases training data variety but does not reduce model size or inference time.

Trap 2: Transfer learning

Transfer learning reduces training time but does not guarantee a smaller model or faster inference.

Trap 3: Ensemble learning

Ensembles combine multiple models, increasing size and inference time.

A
Data augmentation
Why wrong: Data augmentation increases training data variety but does not reduce model size or inference time.
B
Model pruning and quantization
Pruning removes redundant weights and quantization reduces precision, decreasing model size and speeding up inference.
C
Transfer learning
Why wrong: Transfer learning reduces training time but does not guarantee a smaller model or faster inference.
D
Ensemble learning
Why wrong: Ensembles combine multiple models, increasing size and inference time.

Question 8easymultiple choice

A company wants to deploy an AI model for real-time inference on edge devices with limited computational resources. Which model architecture would be MOST suitable?

Trap 1: YOLOv4

YOLOv4 is efficient but still may be too heavy for very limited edge devices; MobileNet is more optimized.

Trap 2: ResNet-152

ResNet-152 is a deep network with many parameters, too large for edge devices.

Trap 3: BERT

BERT is a large transformer model for NLP, not suitable for resource-constrained edge devices.

A
YOLOv4
Why wrong: YOLOv4 is efficient but still may be too heavy for very limited edge devices; MobileNet is more optimized.
B
MobileNet
MobileNet uses depthwise separable convolutions to reduce computation, ideal for edge deployment.
C
ResNet-152
Why wrong: ResNet-152 is a deep network with many parameters, too large for edge devices.
D
BERT
Why wrong: BERT is a large transformer model for NLP, not suitable for resource-constrained edge devices.

Question 9easymultiple choice

An organization deploys an AI model on edge devices for real-time image classification. Which metric is most important to monitor for ensuring the device's operational health?

Trap 1: Model calibration error

Calibration is about probability estimates, not hardware health.

Trap 2: Average prediction confidence

Confidence doesn't indicate device resource status.

Trap 3: Model accuracy on local test data

Accuracy is a model metric, not device health.

A
Model calibration error
Why wrong: Calibration is about probability estimates, not hardware health.
B
Inference memory consumption
Memory is a key operational health indicator for edge devices.
C
Average prediction confidence
Why wrong: Confidence doesn't indicate device resource status.
D
Model accuracy on local test data
Why wrong: Accuracy is a model metric, not device health.

Question 10easymultiple choice

A company streams sensor data from IoT devices. The data arrives as JSON messages at high velocity. Which data pipeline architecture is BEST suited to handle this streaming data for near-real-time analytics?

Trap 1: Batch processing using Hadoop MapReduce every 24 hours.

Batch processing introduces high latency, not suitable for near-real-time.

Trap 2: Batch processing using nightly ETL jobs.

Nightly ETL adds at least 24 hours delay.

Trap 3: Single-node database with periodic inserts.

Cannot handle high velocity and lacks scalability.

A
Batch processing using Hadoop MapReduce every 24 hours.
Why wrong: Batch processing introduces high latency, not suitable for near-real-time.
B
Batch processing using nightly ETL jobs.
Why wrong: Nightly ETL adds at least 24 hours delay.
C
Single-node database with periodic inserts.
Why wrong: Cannot handle high velocity and lacks scalability.
D
Stream processing using Apache Kafka and Spark Streaming.
Kafka ingests streaming data, Spark Streaming processes it with low latency.