Which technique involves setting a fraction of input units to 0 at each update during training time, which helps to prevent overfitting?
- Dropout
- Batch Normalization
- Data Augmentation
- Early Stopping
Dropout involves setting a fraction of input units to 0 during training, which helps prevent overfitting by making the model more robust and reducing reliance on specific neurons.
A finance company wants to analyze sequences of stock prices to predict future market movements. Given the long sequences of data, which RNN variant would be more suited to capture potential long-term dependencies in the data?
- Simple RNN
- Bidirectional RNN
- Gated Recurrent Unit (GRU)
- Long Short-Term Memory (LSTM)
A Long Short-Term Memory (LSTM) is a suitable choice for capturing long-term dependencies in stock price sequences. LSTM's memory cell and gating mechanisms make it capable of handling long sequences and understanding potential trends in financial data.
How does ICA differ from Principal Component Analysis (PCA) in terms of data independence?
- ICA finds statistically independent components
- PCA finds orthogonal components
- ICA finds the most significant features
- PCA reduces dimensionality
Independent Component Analysis (ICA) seeks statistically independent components, meaning they are as unrelated as possible, while PCA seeks orthogonal components that explain the most variance but are not necessarily independent. ICA focuses on data independence, making it suitable for source separation tasks.
In reinforcement learning, what term describes the dilemma of choosing between trying out new actions and sticking with known actions that work?
- Exploration-Exploitation Dilemma
- Action Selection Dilemma
- Reinforcement Dilemma
- Policy Dilemma
The Exploration-Exploitation Dilemma is the challenge of balancing exploration (trying new actions) with exploitation (using known actions). It's crucial in RL for optimal decision-making.
How do the generator and discriminator components of a GAN interact during training?
- The generator produces real data.
- The discriminator generates fake data.
- The generator tries to fool the discriminator.
- The discriminator generates real data.
In a GAN (Generative Adversarial Network), the generator creates fake data to deceive the discriminator, which aims to distinguish between real and fake data. This adversarial process improves the quality of the generated data.
A company wants to develop a chatbot that learns how to respond to customer queries by interacting with them and getting feedback. The chatbot should improve its responses over time based on this feedback. This is an application of which type of learning?
- Online Learning
- Reinforcement Learning
- Supervised Learning
- Unsupervised Learning
This is an application of reinforcement learning. In reinforcement learning, an agent interacts with its environment and learns to make decisions to maximize a reward signal. The chatbot improves based on feedback (rewards) received.
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.
In which algorithm is the outcome determined based on a majority vote from its neighbors?
- K-Nearest Neighbors (K-NN)
- Linear Regression
- Logistic Regression
- Principal Component Analysis (PCA)
K-Nearest Neighbors (K-NN) is a classification algorithm where the outcome is determined by majority voting among its nearest neighbors.