CENG440 Introduction to TinyML

Why is a non-linear activation function important in a neural network model?

• It helps in performing linear transformations.

• It allows the model to learn complex patterns and representations.

• It reduces the size of the neural network.

• It simplifies the training process by making it faster.

What role does the non-linear activation function (represented by the symbol g) play in the neural network structure depicted in the image?

• It combines the weighted sum of inputs.

• It adds bias to the input data.

• It transforms the linear combination of inputs into a non-linear output.

• inputs into a non-linear output.
  D)

How can you evaluate the performance of a linear regression model in a TinyML application?

• By comparing the predicted values to random values.

• By calculating the difference between predicted and actual values using Mean Squared Error (MSE).

• By counting the number of correct classifications.

• By adjusting the weights randomly during training.

Why is the loss function of a linear regression model considered convex in TinyML applications?

• Because it has multiple local minima.

• Because it results in non-linear relationships between inputs and outputs.

• Because it ensures faster convergence due to the absence of local minima.

• Because it only works with small datasets.

What role does the learning rate play when minimizing a convex loss function in a linear regression model in TinyML?

• It determines the final accuracy of the model.

• It increases the complexity of the model.

• It ensures the model avoids overfitting.

• It controls how much to adjust the model parameters at each step during training.

What can happen if the learning rate is set too high or too low when minimizing a convex loss function in a linear regression model in TinyML?

• A low learning rate can cause the model to converge instantly, while a high learning rate leads to underfitting.

• A high learning rate can cause the model to oscillate around the minimum, while a low learning rate can result in slow convergence.

• A high learning rate can lead to faster convergence, while a low learning rate always ensures accurate results.

• A high learning rate prevents overfitting, while a low learning rate ensures maximum generalization.

Which of the following statements about TensorFlow Lite for Microcontrollers (TFLite Micro) is NOT true?

• TFLite Micro allows running machine learning models on resource-constrained devices such as microcontrollers.

• TFLite Micro is specifically designed for low-power and low-memory environments.

• TFLite Micro supports all TensorFlow operations without any restrictions.

• TFLite Micro enables the deployment of models trained using TensorFlow on TinyML devices.

Which of the following statements accurately describes a simple feedforward neural network architecture?

• It contains feedback loops that allow information to flow backward through the network.

• It consists of an input layer, one or more hidden layers, and an output layer, with each layer fully connected to the next.

• It is primarily used for recurrent tasks like time series prediction.

• It utilizes convolutional layers to process spatial data effectively.

Which of the following characteristics of TensorFlow in TinyML is MOST significant for deploying machine learning models on edge devices?

• High computational power requirements for model training.

• Extensive support for multi-threading and parallel processing.

• Optimized model size and memory usage to fit on low-power devices.

• Reliance on cloud computing for all inference tasks.