Winter_school

The gyroscope in an IMU is mainly used to measure:

• Linear acceleration

• Angular velocity

• Magnetic Field strength

• Orientation with respect to gravity

Which library is commonly used to deploy TensorFlow Lite models on ESP32?

• tflite-micro

• PyTorch Mobile

• scikit-learn embedded

• ONNX Runtime

A vanishing gradient problem is most common in:

• Shallow neural networks

• Deep networks with sigmoid/tanh activations

• Decision trees

• K-Nearest Neighbors (KNN)

To improve underwater data acquisition reliability, hydrophones are often arranged in:

• Star topology

• Sensor fusion arrays

• Antenna towers

• Random placement

Transfer learning is particularly effective when:

• The new dataset is very small but similar to the pretrained dataset

• The new dataset is completely unrelated to the source dataset

• Training must be done from scratch

• The model must be used without any fine-tuning

Which type of error commonly accumulates in IMU-based dead reckoning navigation?

• Quantization noise

• Bias drift and integration error

• Signal reflection

• Packet loss

Which of the following best explains the difference between ML and DL?

• ML requires GPUs, DL does not

• DL can automatically extract features, ML often needs feature engineering

• ML is faster but less accurate, DL is slower but always more accurate

• ML models always outperform DL

In TinyML workflows, the ESP32 is usually responsible for:

• Model training

• Cloud-based hyperparameter tuning

• Model inference after deployment

• Labelling dataset

Multipath propagation in underwater acoustics primarily occurs due to:

• Interference from WiFi signals

• Sound reflections from the water surface and seabed

• Poor sensor calibration

• Magnetic field distortion

Sensor fusion algorithms like Kalman filter in IMUs are used to:

• Reduce energy consumption

• Combine sensor outputs for better accuracy

• Increase sampling frequency

• Automatically calibrate WiFi signals