Hands-on Machine Learning with AWS and NVIDIA Quiz Answers

All Weeks Hands-on Machine Learning with AWS and NVIDIA Quiz Answers

Graded Quiz 01

Q1. True or false? Machine learning is the process of getting computers to perform certain tasks by being explicitly programmed.

• True
• False

Q2. Which of the following is NOT a machine learning build stage that has been brought together in Amazon SageMaker Studio?

• Prepare, such as data preparation
• Design, such as built-in and bring your-own algorithms
• Train and tune, such as Amazon SageMaker Experiments
• Deploy and manage, such as Amazon SageMaker Model Monitor

Q3. Which of the following are reasons why developers struggle to go beyond experimentation to operationalize machine learning applications? (Please select all that apply.)

• Debuggers, project management tools, and collaboration tools, which are available when building traditional software, are not available for machine learning development.
• Collecting and preparing data can be a time-consuming process.
• Choosing the right machine learning algorithm is often done by trial and error.
• The short cycle time for model tuning is unable to be monitored.
• All of the above.

Q4. True or False? In Amazon SageMaker, if you have sensitive data, you can encrypt your volume as well as control root access.

• True
• False

Q5. Which of the following are things you can do with Amazon SageMaker Studio’s feature called Data Wrangler? (Please select all that apply.)

• Write your own transformation code.
• Connect to and query various data sources.
• Pull data to a centralized location for you to explore with built-in data transformations.
• Maintain your current level of data preparation time.
• See the model’s estimated accuracy in the data preparation phase itself.

Q6. Which of the following is NOT an Amazon SageMaker Studio Feature?

• Amazon SageMaker Feature Store
• Amazon SageMaker Pipelines
• Amazon SageMaker Clarify
• Amazon SageMaker Intelligence

Q7. The Amazon SageMaker Experiments feature lets you:

• Launch, track, compare and evaluate thousands of model training experiments and model versions to quickly identify best performing models.
• Automatically manage all experiments you perform after you manually set it up.
• Track up to 100 experiments.
• Track and record datasets, algorithms, parameters, and model metrics only.

Q8. True or false? NVIDIA optimizes and makes available for free, a plethora of popular machine learning and deep learning software libraries and frameworks to ensure that whenever you’re building or deploying a model on an NVIDIA GPU, you’re tapping into the maximum potential of that GPU.

• True
• False

Q9. Which of the following processes is NOT one that the NVIDIA NGC Catalog helps simplify and accelerate:

• Building containers
• Creating your own AI models
• Developing AI SDKs
• Designing Helm Charts

Q10. Data parallelism and model parallelism help you to deal with what challenge?

• Dealing with hundreds of thousands of jobs, keeping track of metrics, grouping jobs by experiment, comparing jobs, and querying the past jobs.
• Ensuring that large models do not exceed the memory of a single GPU and create an out-of-memory error.
• Quickly turning your model to be as accurate as possible.
• Training large datasets can take a long time and create a bottleneck for data science teams.

Graded Quiz 02

Q1. True or False? RAPIDS is a suite of open-source software libraries that can accelerate your machine learning code with GPUs.

• True
• False

Q2. The most popular frameworks in machine learning, such as xgboost, pandas, scikit-learn, and numpy, are not:

• Used by data scientists on a daily basis.
• Able to help you to complete the end-to-end flow of a data scientist.
• GPU accelerated.
• Available as classes on Coursera enabling you to go in more depth with these tools.

Q3. Which of the following would best describe a typical machine learning dataset?

• Single-source
• Complete data
• Contains highly dimensional data points
• Single data type

Q4. The core concepts of ETL are:

• Extract, transform, and load
• Exhume, transform, and load
• Extract, transform and launch
• Exhume, transform, and launc

Q5. ____________________ is a great tool for telling us how independent different features are from each other within a dataset.

• Model design
• Cross-correlation
• ETL
• Optimized Row Columnar file

Q6. True or false? Machine learning models tend to be more complex than neural networks where neural networks excel at structured data.

• True
• False

Q7. What is the last step of the machine learning workflow?

• Release
• Insinuation
• Inference
• Model Optimizatio

Q8. Which of the following is NOT one of Triton Inference Server’s benefits or capabilities?

• It is an open-source tool that can serve both deep learning and machine Learning models.
• It can simultaneously manage multiple models while executing multiple workloads simultaneously on the same GPU resources.
• It has backends for FIL, ONNX, and most deep learning libraries.
• It allows model optimization of inference requests and model pipelines.

Q10. Which of the following is true about the AutoGluon library? (Select all that apply.)

• It can run on a single machine with as little as one GPU and takes only about 3 lines of code.
• It will automatically try to clean your data.
• It will automatically try to determine the best model architecture.
• Launch virtual machine

Q10. Which parameter is used to train an XGBoost model using RAPIDS and an NVIDIA GPU?

• max_depth
• n_estimators
• max_features
• tree_method

Graded Quiz 03

Q1. Computer vision:

• is an interdisciplinary scientific field.
• deals with how to allow machines to see images and video.
• deals with how machines process images and video.
• All of the above.
• None of the above.

Q2. Who created the first convolution neural network?

• Dr. Kunihiko Fukushima
• Isaac Privitera
• Yann LeCun
• Alex Krizhevsky

Q3. Which of the following is NOT one of the top 3 computer vision tasks?

• Image identification
• Image classification
• Object detection
• Semantic segmentation

Q4. How many unique data points are there in a color 10×10 pixel image?

• 255
• 300
• 765
• 100

Q5. Which of the following are benefits of using a DataLoader? (Select all that apply.)

• Works on large image datasets usually too difficult to fit in memory.
• Can specify multiple workers to load objects in parallel.

Q6. What is the correct sequence for conducting convolution using the torch.nn.functional module?

• Run a 2-dimensional convolution with a stride that you determine, create an image tensor, then define the kernel.
• Define the kernel, run a 2-dimensional convolution with a stride that you determine, then create an image tensor.
• Run a 2-dimensional convolution with a stride that you determine, define the kernel, then create an image tensor.
• Define the kernel, create an image tensor, then run a 2-dimensional convolution with a stride that you determine.

Q7. True or false? Intersection over Union, or IoU, is an object detection measure of the level of overlap between a predicted label and a ground truth label. Given two bounding boxes, we compute the area of the union and divide by the area of the intersection.

• True
• False

Q8. Instead of having two pieces of the model that needed to train separately, _____________ introduced a region of interest pooling and a single stage training algorithm that jointly learned to classify object proposals and refine their locations.

• OverFeat
• RCNN
• Fast-RCNN
• Faster-RCNN

Q9. ______________________ is the process of using the learned weights from a model and applying them to a different but related task.

• Transfer learning
• Model training
• Model Transformation
• Transfer scaling

Q10. Choose the answer below that correctly identifies and sequences the main steps needed to deploy a running Amazon SageMaker endpoint.

1. Create a model in Amazon SageMaker.
2. Find the model components.
3. Create a HTTPS endpoint.
4. Provide the endpoint configuration to Amazon SageMaker.

1. Create an endpoint configuration for an HTTPS endpoint.
2. Create a HTTPS endpoint.
3. Provide the endpoint configuration to Amazon SageMaker.

1. Create a model in Amazon SageMaker.
2. Create a HTTPS endpoint.
3. Provide the endpoint configuration to Amazon SageMaker.
4. Create an endpoint configuration for an HTTPS endpoint.

1. Create a model in Amazon SageMaker.
2. Create an endpoint configuration for an HTTPS endpoint.
3. Provide the endpoint configuration to Amazon Sagemaker and create a HTTPS endpoint.

Graded Quiz 04

Q1. What is Natural Language Processing (NLP)? (Please select all that apply.)

• Computational processing of natural or human languages.
• It is useful in tasks like search, spam detection, language translation, and grammar correction.
• The probability of a given sequence of words occurring in a sentence.
• Represents language units, such as words or characters, in a continuous vector space.

Q2. Which common task used for evaluating NLP models is summarized by the goal of identifying persons, locations, or products in a given text?

• Sentiment classification
• Question answering
• Named entity recognition
• Machine translation

Q3. True or false? BERT, or Bidirectional Encoder Representations from Transformers, is a method of pretraining language representations which obtains state-of-the-art results on a wide array of NLP tasks.

• True
• False

Q4. _________________ are a neural network architecture that learns the human language using self-attention, where a segment of words is compared against itself.

• Bidirectionals
• Encoders
• Representations
• Transformers

Q5. Which of the following is the correct BERT workflow for a question answering task?

• Raw data, preparation, refinement, and inference
• Raw data, pre-training, fine-tuning, and inference
• Raw data, pre-training, fine-tuning, and implication
• Raw data, preparation, refinement, and implication

Q6. True or false? Next Sentence Prediction takes an input segment and withholds a random 15% of the words. The model then tries to predict the missing word and compares it with the word that was initially removed.

• True
• False

Q7. With Mixed Precision, you can train with __________________ while maintaining the network accuracy achieved with __________________.

• full Precision, single Precision
• half Precision, full Precision
• half Precision, single Precision
• full Precision, zero Precision

Q8. The parameter _________________ ensures we are training using Mixed Precision. Setting this parameter will save time without compromising the accuracy of the training.

• doc_stride
• do_train
• train_file
• fp16

Q9. True or False? Triton is an open-source, scalable inference server that addresses deployment challenges and makes it easier for organizations to deploy their models in production at scale.

• True
• False

Q10. Which of the following is NOT accurate about Triton in Amazon SageMaker?

• You can only run one model at a time on different types of inference queries and use-cases.
• It allows you to run multiple copies of a model on your compute instance.
• Supports many popular frameworks like TensorFlow and PyTorch as backends.
• You can run Inference on CPUs or GPUs and it ensures that you are maximizing the utilization of your hardware.

Final Graded Quiz

Q1. True or false? A common way to develop and run ML models is on a set of powerful servers which may seem less expensive upfront, but either they are under-utilized or need time and money to scale, maintain, and make them available. Amazon SageMaker removes a lot of this effort for you.

• True
• False

Q2. In any model’s lifecycle, there are stages that machine learning builders typically deal with. All these tools for the stages are brought together in one place through an integrated workbench called _____________________.

• Jupyter Lab
• Amazon Web Services
• NVIDIA
• Amazon SageMaker Studio

Q3. ________________ is an open-source framework that is an API-compatible framework that allows existing machine learning code to be ported and benefit from GPU acceleration.

• scikit-learn
• RAPIDSAI
• MapReduce
• Apache Hadoop

Q4. The process of downloading and transforming data, often referred to as _________________, is one of the most common and often most time-intensive aspects of data science.

• Extract Transform Load
• Exhume Transform Launch
• Extract Transfer Load
• Extract Transfer Launch

Q5. True or false? XGBoost is an optimized inference server to parallelize predictions on CPU and GPU.

• True
• False

Q6. ____________________ was the first convolutional neural network. Instead of breaking down the image into expert generated features, it used convolution to gain a greater understanding of the spatial features within images.

• WieselWeb
• Neocognitron
• Fast-RCNN
• LeNet

Q7. ____________________ is assigning a label to an entire image.

• Image identification
• Image classification
• Object detection
• Semantic segmentation

Q8. _______________ used region proposals, a piece of the network that used selective search to generate around 2,000 category independent region proposals for the input image.

• OverFeat
• RCNN
• AlexNet
• ResNet-50

Q9. The input to the BERT Model is an encoded representation of the input sentence which does NOT include which of the following?

• Token embeddings
• Classification embeddings
• Segment embeddings
• Positional embeddings

Q10. Which of the following is NOT one of the 3 key challenges related to deploying an inference solution?

• Ease of use
• Cost of deployment
• Automation of inference
• Lack of integration with ecosystem tools

Find more related Quiz Answers >>

Introduction to Cybersecurity Tools & Cyber Attacks

Cybersecurity Roles, Processes & Operating System Security

Cybersecurity Compliance Framework & System Administration

Network Security & Database Vulnerabilities

Penetration Testing, Incident Response, and Forensics