Select Two (Multi-Select) Questions

Apache Airflow

Why wrong: Airflow is an alternative orchestrator, not typically integrated with Kubeflow.

Weights & Biases

Why wrong: W&B is a separate experiment tracking tool; could be used but not the typical Kubeflow integration.

Amazon S3

Why wrong: S3 is object storage, not a vector database.

Snowflake

Why wrong: Snowflake is a data warehouse, not optimized for vector search.

MLflow

Why wrong: MLflow is for managing the ML lifecycle, not for containerization or orchestration of deployed models.

Kubeflow

Why wrong: Kubeflow is an MLOps platform that runs on Kubernetes, but it is not a containerization tool itself; the question specifically asks for tools to containerize and orchestrate.

Remove dropout layers to allow more gradient flow

Why wrong: Removing dropout may increase overfitting but not necessarily help convergence.

Use a smaller batch size to reduce memory

Why wrong: Smaller batch size introduces noise, which may not help convergence.

Increase the learning rate by 10x

Why wrong: A high learning rate may cause divergence, not convergence.

Replace the model with a decision tree for transparency

Why wrong: May not be feasible if deep learning is required.

Rely on the model's internal attention weights (if transformer-based)

Why wrong: Attention weights are not always reliable explanations.

Calculate global feature importance using permutation importance

Why wrong: Global importance doesn't explain individual decisions.

Remove the minority class samples to make the dataset balanced

Why wrong: Removing samples discards valuable data and reduces model's ability to learn the minority class.

Report only accuracy as the evaluation metric

Why wrong: Accuracy is misleading for imbalanced data.

It requires the base models to be of different types

Why wrong: Bagging typically uses the same type of base model (e.g., all decision trees).

It reduces the error by combining weak learners

Why wrong: That describes boosting; bagging combines strong or unstable models to reduce variance.

A relational database with BLOB storage

Why wrong: Relational databases lack vector search capabilities and are inefficient for similarity search.

A GPU cluster for model serving

Why wrong: RAG retrieval does not require GPU; the embedding model may run on CPU for retrieval, and GPU is for the generation model.

A data warehouse like Snowflake or BigQuery

Why wrong: Data warehouses are for analytical queries on structured data, not for vector search.

Decrease the learning rate

Why wrong: Learning rate affects training speed, not directly overfitting.

Increase the training batch size

Why wrong: Larger batch sizes can lead to sharper minima and may not reduce overfitting.

Regression testing only after model updates

Why wrong: Regression testing is important but typically performed after changes; the question asks for testing types to include in the plan, and the three selected (A, C, D) are core to a comprehensive test strategy.

Load testing to determine maximum concurrent users

Why wrong: Load testing is for performance and scalability, not for functional correctness or output quality.

Increase the model's context window size beyond the training limit

Why wrong: Extending context beyond training length can degrade performance and increase hallucinations.

Use a higher temperature setting during generation

Why wrong: Higher temperature increases randomness, potentially increasing hallucinations.

Fine-tune the model on safe conversational examples

Why wrong: Fine-tuning can improve general safety but is not a direct or immediate injection mitigation.

Use a larger context window

Why wrong: A larger context window does not prevent injection; it may even provide more room for injection.

Limit the maximum output token length

Why wrong: Limiting output length reduces verbosity but does not prevent injection.

Chroma

Why wrong: Chroma is open-source; managed cloud is less mature.

pgvector (PostgreSQL extension)

Why wrong: pgvector requires self-hosting and management, not a managed cloud service.

Amazon DynamoDB

Why wrong: DynamoDB is a key-value/NoSQL database, not optimized for vector similarity search.

Apply model distillation to create a smaller student model

Why wrong: Distillation can reduce model size, but it requires training a new model, which is more complex than quantization; quantization is more straightforward for this scenario.

Deploy the model on a cloud server and stream data

Why wrong: Cloud deployment introduces network latency and requires constant connectivity, unsuitable for real-time patient monitoring on IoT devices.

Use ONNX Runtime with a GPU backend

Why wrong: ONNX Runtime with GPU backend is designed for server-class hardware, not for low-power IoT devices.

Train the model on a small subset of data to reduce exposure

Why wrong: Training on a subset may reduce accuracy but does not provide formal differential privacy guarantees.

Remove all personally identifiable information (PII) from the dataset

Why wrong: Removing PII is an anonymization technique, not differential privacy.

Aggregate data into groups before training

Why wrong: Aggregation can reduce granularity but does not provide differential privacy guarantees.

Explicit consent from all data subjects

Why wrong: Consent is not always required; other legal bases may apply.

Data retention period of at least 10 years

Why wrong: Retention periods vary; no universal 10-year requirement.

Hugging Face Transformers

Why wrong: Hugging Face Transformers is primarily for NLP models, not computer vision (though it has some vision models, it's not the best fit for general computer vision transfer learning).

scikit-learn

Why wrong: scikit-learn is for classical machine learning, not deep learning or computer vision transfer learning.

Keras

Why wrong: Keras is a high-level API that runs on top of TensorFlow; it is not a standalone framework for transfer learning.

Train on the largest possible dataset

Why wrong: Larger datasets increase training time and energy use.

Use redundant backup servers

Why wrong: Redundant servers increase energy consumption without contributing to efficiency.

Increase batch size to maximum

Why wrong: Larger batch sizes may improve throughput but can increase peak power draw.

Immediately retrain the model with all available data.

Why wrong: Retraining without diagnosis may lock in errors or ignore drift root cause.

Increase the size of the training dataset by adding synthetic data.

Why wrong: Synthetic data may not reflect real-world shifts and could introduce bias.

Accuracy

Why wrong: Accuracy can be high even if the model fails on the minority class, making it inappropriate for imbalanced data.

Mean squared error (MSE)

Why wrong: MSE is a regression metric and not suitable for classification.

F1 score

Why wrong: Although F1 is useful, it is a single metric that combines precision and recall; the question explicitly asks for TWO metrics, and precision/recall are the foundational pair.