Machine Learning Question Bank

Which hardware is specifically designed by Google for accelerating neural network training?

TPU (Tensor Processing Unit)

FPGAs (Field-Programmable Gate Arrays) are valuable in ML because they:

Can be reprogrammed for different ML tasks, offering flexibility

Which statement about ML hardware is MOST accurate?

Hardware selection should match the specific ML task, scale, and constraints

A privacy-sensitive healthcare application that cannot send patient data to external servers should use:

Edge devices or on-premise infrastructure

A model trained to recognize cats in images is adapted to recognize dogs with minimal additional training. This demonstrates:

Transfer learning

Transfer learning is particularly valuable because it:

Reduces training time and data requirements by leveraging pre-trained models

Which application would LEAST benefit from deep learning?

Predicting house prices from square footage (simple linear relationship)

A hospital wants to analyze thousands of patient records to discover previously unknown patterns in disease progression without any predetermined categories. Which ML approach is most appropriate?

Unsupervised learning

A medical device that performs real-time diagnostics during surgery requires:

Edge computing for low-latency local processing

Sentiment analysis of customer reviews to determine positive or negative opinions is an application of:

Supervised learning with natural language processing

Unsupervised learning is most appropriate when:

You want to discover hidden patterns or structures in unlabeled data

Which scenario demonstrates supervised learning?

Training a model to predict house prices using historical price data

A company has 100,000 labeled images of defective and non-defective products. They want to automatically identify defects on the production line. Which approach is most suitable?

Supervised learning with deep learning

A robot vacuum learning the optimal cleaning path through trial and error in different room layouts demonstrates:

Reinforcement learning

Which scenario BEST demonstrates transfer learning?

Using a model trained on ImageNet to jumpstart training for medical image classification

The primary trade-off when choosing between cloud and edge deployment is:

Scalability/flexibility vs. latency/privacy

The key advantage of using GPUs over CPUs for deep learning training is:

Massive parallelization of matrix operations, reducing training time significantly

A company wants to use a model trained on general object recognition to identify specific manufacturing defects. This is:

Transfer learning

Which statement about supervised learning is FALSE?

It discovers hidden patterns without any labels

A system that analyzes social media posts to determine public opinion on a political issue uses:

Sentiment analysis with supervised learning

ASICs (Application-Specific Integrated Circuits) for ML are advantageous because they:

Are optimized for specific ML operations, offering high efficiency

Deep learning differs from traditional machine learning primarily in:

Its use of neural networks with multiple layers to learn hierarchical features

Which statement about reinforcement learning is MOST accurate?

It learns through trial and error with reward signals

GPUs are particularly advantageous for machine learning because they:

Excel at parallel processing required for matrix operations in neural networks

Natural language processing for chatbots typically combines:

Deep learning with supervised and unsupervised techniques

Object detection and classification in autonomous vehicles primarily relies on:

Deep learning with convolutional neural networks

Which hardware configuration is LEAST suitable for training large deep learning models?

Standard laptop CPU

A fraud detection system that learns from labeled examples of fraudulent and legitimate transactions uses:

Supervised learning

TPUs (Tensor Processing Units) are specifically optimized for:

Neural network operations and tensor computations

A mobile app performing image classification on a smartphone uses:

Edge computing on the device's processor

High-performance computing (HPC) centers are most appropriate for:

Large-scale research and training of massive models requiring distributed computing

Market basket analysis, which discovers that customers who buy diapers often buy beer, primarily uses:

Unsupervised learning

Which factor is MOST important when selecting hardware for real-time autonomous vehicle navigation?

Low latency and edge processing capability

Which scenario does NOT appropriately match the ML approach?

Medical imaging diagnostics - Unsupervised learning only

A startup is prototyping a simple ML model with a small dataset. What hardware is most appropriate initially?

Standard laptop with CPU

Object detection in security camera footage to identify suspicious activities uses:

Deep learning with supervised training on labeled video data

Deep learning is particularly advantageous over traditional ML when:

You're working with unstructured data like images, audio, or text

The key characteristic that distinguishes reinforcement learning from supervised learning is:

RL learns from reward signals through interaction, not labeled examples

Cloud-based ML platforms are advantageous for:

Scalability, flexibility, and avoiding upfront hardware costs

A streaming service recommending movies based on viewing patterns of similar users primarily uses:

Unsupervised learning (collaborative filtering)

A company needs to deploy facial recognition on security cameras that must work without internet connectivity. They should use:

Edge devices for local processing

A company training multiple ML models simultaneously with varying requirements benefits most from:

Cloud platforms offering diverse instance types and scalability

Medical imaging diagnostics using deep learning to identify tumors in MRI scans is an example of:

Supervised learning with labeled medical images

A social media company analyzing billions of user posts needs hardware that prioritizes:

Scalability and distributed processing (cloud platforms, HPC centers)

A startup has only 500 labeled X-ray images but wants to build a diagnostic tool. What strategy would be most effective?

Use transfer learning from a model pre-trained on general images

Which real-world application correctly matches reinforcement learning?

Robotics navigation learning optimal paths

A self-driving car learns to navigate by receiving positive rewards for staying in lanes and negative rewards for unsafe maneuvers. This is an example of:

Reinforcement learning

A robot learning to grasp objects of various shapes through repeated attempts, improving with each try, demonstrates:

Reinforcement learning

A research institution training cutting-edge language models with billions of parameters needs:

High-performance computing centers with distributed GPU/TPU infrastructure

A recommendation system that groups users with similar preferences without predefined categories uses:

Unsupervised learning (clustering)