ML-Embedded Ex

In the context of embedded machine learning, why is it important to carefully monitor both accuracy and loss during model training?

• Accuracy and loss directly influence the hardware cost of the embedded device.

• High accuracy always means low loss, making monitoring both redundant.

• Embedded systems have resource constraints, so balancing accuracy and loss ensures optimal model performance and efficiency.

• Loss is irrelevant in embedded systems since only accuracy matters.

Which of the following loss functions is most commonly used for regression problems in embedded ML models?

• Binary Cross-Entropy

• Mean Squared Error

• Categorical Cross-Entropy

• Hinge Loss

Why is inference time critical in embedded machine learning model development?

• It affects the storage size of the model on the embedded device.

• Faster inference time ensures real-time performance and energy efficiency, essential for embedded systems.

• Inference time determines the accuracy of the model.

• Long inference time improves the model’s interpretability.

What is a primary cause of overfitting in machine learning models, particularly in embedded ML systems?

• Using a dataset with too few features.

• Having an excessively complex model relative to the training data.

• Using a high dropout rate during training.

• Implementing pruning techniques.

What is the CIFAR-10 dataset commonly used for in machine learning?

• Sentiment analysis of text data

• Classification of medical images

• Image classification into 10 distinct categories like airplanes, cats, and trucks

• Predicting numerical data in regression models

Which of the following stages is NOT typically part of the machine learning lifecycle?

• Data Collection and Preparation

• Model Training and Evaluation

• Hardware Manufacturing

• Model Deployment and Monitoring

In resource-constrained systems, how can you optimize model performance during training on CIFAR-10?

• Use high-resolution images directly without preprocessing.

• Train the model without a validation set to save resources.

• Use techniques like data augmentation, quantization, and early stopping.

• Use only the testing set for both training and evaluation.

How can Google Colab assist in developing ML applications for embedded systems?

• By directly deploying models to the embedded system without any preprocessing

• By providing a cloud-based environment to preprocess data, train models, and simulate edge deployments

• By limiting computation to only lightweight models for embedded use

• By replacing the need for Edge Impulse in embedded ML workflows

What is the importance of the data collection stage in the ML lifecycle?

Explain why minimizing the loss function is crucial in developing embedded ML models.