Q. 1
Artificial Intelligence
Difficulty: Hard
(1 Mark)
In time-series analysis, what three conditions must be satisfied for a series to be considered 'Weakly Stationary' (Covariance Stationary)?
💡
Step-by-Step Explanation & Concept Rationale
Stationarity ensures statistical properties do not change over time, a prerequisite for linear forecasting models like ARIMA.
Q. 2
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What statistical hypothesis test is standardly used to test for the presence of a unit root (non-stationarity) in a univariate time series?
💡
Step-by-Step Explanation & Concept Rationale
The ADF test checks the null hypothesis that a unit root is present (non-stationary); p-value < 0.05 rejects the null, indicating stationarity.
Q. 3
Artificial Intelligence
Difficulty: Medium
(1 Mark)
In an ARIMA(p, d, q) model, what do the parameters p, d, and q represent?
💡
Step-by-Step Explanation & Concept Rationale
ARIMA models combine differencing 'd' to achieve stationarity with AR(p) past lags and MA(q) past forecast error terms.
Q. 4
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What visual diagnostic plots are examined to determine the appropriate AR order (p) and MA order (q) in Box-Jenkins ARIMA modeling?
💡
Step-by-Step Explanation & Concept Rationale
PACF cuts off after lag p for AR(p) processes; ACF cuts off after lag q for MA(q) processes.
Q. 5
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'SARIMA' (Seasonal ARIMA: ARIMA(p,d,q)(P,D,Q)_s) and what does parameter 's' signify?
💡
Step-by-Step Explanation & Concept Rationale
SARIMA captures both non-seasonal short-term dynamics and recurring seasonal cycle patterns (e.g. annual monsoon rainfall peaks).
Q. 6
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Prophet' (developed by Meta / Facebook) additive time-series forecasting model?
💡
Step-by-Step Explanation & Concept Rationale
Taylor & Letham designed Prophet to handle real-world business and environmental time series with missing data, trend shifts, and robust outlier handling.
Q. 7
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'DeepAR' (Salinas et al. / Amazon Research) for probabilistic time-series forecasting?
💡
Step-by-Step Explanation & Concept Rationale
DeepAR trains across large item catalogs or multi-sensor networks (e.g. 500 river gauge stations), predicting probabilistic quantile forecasts.
Q. 8
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Temporal Fusion Transformer' (TFT, Lim et al. / Google Cloud AI)?
💡
Step-by-Step Explanation & Concept Rationale
TFT handles heterogeneous inputs (static metadata, known future inputs like rain forecasts, and observed past histories) with native prediction intervals.
Q. 9
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'N-BEATS' (Neural Basis Expansion Analysis for Interpretable Time Series, Oreshkin et al.)?
💡
Step-by-Step Explanation & Concept Rationale
N-BEATS surpassed classical statistical ensembles in the M4 competition by decomposing forecasts into interpretable trend and seasonality basis functions.
Q. 10
Artificial Intelligence
Difficulty: Hard
(1 Mark)
When using LSTM networks for multi-step river flood discharge forecasting, what is 'Autoregressive Multi-Step' versus 'Direct Multi-Step' forecasting?
💡
Step-by-Step Explanation & Concept Rationale
Direct forecasting avoids cascading error accumulation inherent in autoregressive rollouts at the cost of training H independent models.
Q. 11
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'MLflow' (open-source MLOps platform) and what are its 4 core components?
💡
Step-by-Step Explanation & Concept Rationale
MLflow manages the complete machine learning lifecycle from experimentation to model registry and deployment.
Q. 12
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'DVC' (Data Version Control) in modern MLOps pipelines?
💡
Step-by-Step Explanation & Concept Rationale
DVC stores lightweight .dvc pointer files in Git while syncing large multi-gigabyte data files and model checkpoints to remote cloud buckets.
Q. 13
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Docker' containerization and why is it essential for machine learning reproducibility and deployment?
💡
Step-by-Step Explanation & Concept Rationale
Containers eliminate 'it works on my machine' dependency conflicts between local development laptops and production cloud clusters.
Q. 14
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Kubernetes' (K8s) in production machine learning infrastructure?
💡
Step-by-Step Explanation & Concept Rationale
K8s manages auto-scaling of inference pods (e.g. via HPA based on GPU utilization or request queues) and zero-downtime rolling updates.
Q. 15
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'KServe' (formerly KFServing) on Kubernetes?
💡
Step-by-Step Explanation & Concept Rationale
KServe provides production ML inference capabilities like payload logging, drift detection integration, and multi-model serving out-of-the-box.
Q. 16
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'ONNX' (Open Neural Network Exchange) runtime format?
💡
Step-by-Step Explanation & Concept Rationale
ONNX decouples training frameworks from production serving environments, enabling hardware acceleration via ONNX Runtime on Intel, AMD, ARM, and NVIDIA chips.
Q. 17
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Triton Inference Server' (developed by NVIDIA)?
💡
Step-by-Step Explanation & Concept Rationale
Triton serves PyTorch, TensorRT, ONNX, and Python models simultaneously with microsecond latency and hardware saturation.
Q. 18
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Dynamic Batching' in Triton Inference Server / vLLM?
💡
Step-by-Step Explanation & Concept Rationale
Dynamic batching dramatically increases server throughput (QPS) on concurrent traffic while maintaining low response latency for individual users.
Q. 19
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'FastAPI' and why is it preferred over Flask for building production Python ML APIs?
💡
Step-by-Step Explanation & Concept Rationale
FastAPI delivers NodeJS/Go-level performance in Python and simplifies input/output schema validation for complex ML payloads.
Q. 20
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'gRPC' compared to REST (HTTP/JSON) for high-performance ML inference microservices?
💡
Step-by-Step Explanation & Concept Rationale
In high-throughput computer vision and audio pipelines, Protobuf serialization in gRPC avoids JSON text parsing bottlenecks.
Q. 21
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Model Drift' (Model Decay) in production ML systems?
💡
Step-by-Step Explanation & Concept Rationale
Model drift requires continuous telemetry monitoring and automated retraining triggers to maintain production accuracy.
Q. 22
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Evidently AI' / 'Whylogs' library used for in MLOps?
💡
Step-by-Step Explanation & Concept Rationale
Evidently computes statistical distance tests (KS test, Wasserstein distance, PSI) to generate production drift dashboards.
Q. 23
Artificial Intelligence
Difficulty: Hard
(1 Mark)
What is 'Feature Store Online vs Offline Store' architecture?
💡
Step-by-Step Explanation & Concept Rationale
Dual-store feature architecture guarantees feature consistency while satisfying opposing throughput and latency requirements.
Q. 24
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'CI/CD for Machine Learning' (CD4ML / MLOps Pipeline)?
💡
Step-by-Step Explanation & Concept Rationale
CD4ML treats data and model pipelines with the same automated rigor as traditional software engineering.
Q. 25
Artificial Intelligence
Difficulty: Medium
(1 Mark)
What is 'Great Expectations' tool in data quality engineering?
💡
Step-by-Step Explanation & Concept Rationale
Great Expectations catches corrupt, out-of-bounds, or incomplete data at ingestion checkpoints, preventing silent model training failures.
Study Stream Progress:
Showing 25 of 82 Questions (30%)
Jump to:
Ready to Test Your Retention & Speed?
Now that you have reviewed the study questions and rationales, test yourself in our interactive 1-by-1 practice engine or take the full official timed mock exam.