Embeddings: Test Your Knowledge Return to pathway Which of the following would be good candidates for an embedding? (Choose all that apply) Choose as many answers as you see fit. Daily high temperatures for Tokyo, Japan from 1999–2024. Genomic sequences of simple viruses. A large dataset of high-definition photos of horses. Lines of code in a large software project. You encode a database of 100px by 100px black-and-white images of handwritten digits as vectors representing the pixels in the image: 0 for white and 1 for black. If you create an embedding from this encoding, roughly how many dimensions will your embedding have? 10,000 dimensions Greater than 10,000 dimensions Fewer than 10,000 dimensions Which of the following are benefits of using embedding vectors for feature data over one-hot vectors of the same data? (Choose all that apply) Choose as many answers as you see fit. A model using embedding vectors will have fewer weights to tune during training. Training the model will be faster and cheaper when using embedding vectors. A model trained on the embeddings does not need to be evaluated with a test set. Dimensionality of the data will increase when using embedding vectors, improving model performance. True or False: Weights taken from a hidden layer of a trained neural network can be used as an embedding. True False In what ways does a contextual embedding differ from a static embedding? (Choose all that apply) Choose as many answers as you see fit. Contextual embeddings encode positional information, while static embeddings do not. One token is represented by one static embedding, but can be represented by multiple contextual embeddings. Contextual embeddings have a lower computational cost when compared to static embeddings. Static embeddings allow for semantically meaningful mathematical operations between vectors in all use cases, while contextual embeddings do not. Submit answers error_outline An error occurred when grading the quiz. Please try again.
