When regular Q-learning takes too much time to converge in a high-dimensional state space (e.g., autonomous vehicle parking), what modification could help it learn faster?
- Deep Q-Networks (DQNs)
- Policy Gradient Methods
- Fitted Q-Iteration (FQI)
- Temporal Difference (TD) Learning
Using Deep Q-Networks (DQNs) is a modification of Q-learning, which employs neural networks to handle high-dimensional state spaces efficiently. DQNs can approximate the Q-values, expediting learning in complex environments.
Techniques like backward elimination, forward selection, and recursive feature elimination are used for ________ in machine learning.
- Cross-Validation
- Data Preprocessing
- Feature Selection
- Model Training
Techniques like backward elimination, forward selection, and recursive feature elimination are used for feature selection in machine learning. Feature selection helps identify the most relevant features for building accurate models and can improve model efficiency.
Which tool or technique is often used to make complex machine learning models more understandable for humans?
- Explainable AI (XAI)
- Reinforcement Learning
- Principal Component Analysis
- Gradient Boosting
Explainable AI (XAI) techniques are employed to simplify complex machine learning models, making them interpretable, and providing insights into model decisions.
One of the applications of NLP in healthcare is to assist in ________, which involves the conversion of voice-recorded notes into text format.
- Transcription
- Speech Recognition
- Note Parsing
- Audio Conversion
NLP (Natural Language Processing) can be employed for transcription tasks in healthcare, where voice-recorded notes from medical professionals are converted into text format. This conversion makes the notes more accessible and searchable for medical records.
A bank uses a machine learning model for loan approvals. However, it's observed that individuals from certain ethnic backgrounds are consistently getting rejected more than others, despite having similar financial profiles. This raises concerns related to which aspect of machine learning?
- Fairness & Bias
- Accuracy
- Data Preprocessing
- Model Training
This situation pertains to fairness and bias in machine learning, specifically, unfair outcomes or discrimination against certain ethnic groups, which highlights ethical concerns. It's important to address these biases for equitable decision-making.
How does NLP handle the nuances and variations in medical terminologies across different healthcare systems?
- Named Entity Recognition and Mapping
- Machine Translation of Medical Terms
- Simple Text Matching
- Manual Reconciliation
NLP utilizes techniques like Named Entity Recognition to identify and map medical terms to a standardized vocabulary, addressing the nuances and variations in medical terminologies across different healthcare systems. This enables consistent analysis and retrieval of information.
What is the primary objective of Generative Adversarial Networks (GANs)?
- Data Classification
- Data Generation
- Data Storage
- Data Analysis
The primary objective of GANs is data generation. GANs consist of a generator that creates data samples to closely resemble real data, aiding in tasks like image generation.
A medical diagnosis AI system provides a diagnosis but does not give any rationale or reasoning behind it. What aspect of machine learning is this system lacking?
- Interpretability
- Classification
- Model Complexity
- Feature Engineering
The system's lack of providing rationale or reasoning is a deficiency in interpretability. In medical AI, it's crucial for doctors to understand why a diagnosis was made to trust and make informed decisions based on the AI's recommendations.
Which evaluation metric would be least affected by a large number of true negatives in a dataset?
- Accuracy
- Precision
- Recall
- Specificity
Specificity is the evaluation metric least affected by a large number of true negatives in a dataset. It focuses on correctly identifying true negatives and is particularly relevant in situations where false positives should be minimized.
When using transfer learning, what part of the pre-trained model is typically fine-tuned for the new task?
- Last few layers
- First few layers
- All layers
- Random layers
In transfer learning, the last few layers are typically fine-tuned because they contain task-specific information, while the early layers retain more generic features.