Interview questions in AI & ML, page 12
Describe the difference between online and batch inference.
Online serves single requests in ms on live endpoints; batch processes data offline with elastic compute.
How do you detect and handle outliers in multivariate datasets?
This tests statistical judgment beyond univariate thresholds. A strong answer names multivariate methods like Mahalanobis distance or isolation forests, PCA or pair plots for visualization, and checks for measurement error before removal.
Classic image captioning architecture
A CNN encoder extracts image features, a recurrent or transformer decoder generates the caption word by word, and attention lets the decoder focus on image regions per word.
Brightness constancy and small-motion assumptions
Brightness constancy says a point's intensity is invariant under motion; small motion lets you linearize via Taylor expansion.
Where to place feature transformations: client, serving API, or upstream service?
Tests separation of concerns in ML systems. Client causes duplication and skew; serving API couples compute to requests; dedicated service adds a network hop but centralizes logic. Red flag: ignoring training-serving skew.
How do you visualize clusters in 100-dimensional numerical data?
Tests PCA versus t-SNE tradeoffs for high-dimensional visualization. Strong answers use PCA first for global variance and outliers, then t-SNE with perplexities 5-50 run to stability. Red flag: interpreting t-SNE cluster sizes or distances as literal metrics.
Early versus late modality fusion
Early fusion merges raw or low-level features so the model learns cross-modal interactions, while late fusion processes each modality separately and combines outputs.
Kalman filter for bounding-box tracking
State, transition, measurement models, and process plus measurement noise; predict then correct each frame. State holds box position and velocity; measurement is the detected box.
Design a system for batch scoring millions of customer records daily
Shard jobs via scheduler; partition storage; right-size CPU/GPU on spot; retry.
How does a U-shaped EDA pattern influence feature engineering?
Add age squared or splines; trees handle splits but explicit terms aid linear models.
Designing a Visual Question Answering system
Encode the image with a vision backbone, encode the question with a text encoder, fuse them via cross-attention into a joint representation, then decode or classify the answer.
3D CNNs vs two-stream action recognition
3D CNNs learn spatiotemporal filters end to end but are heavy; two-stream splits RGB appearance and precomputed optical flow, strong but costly to compute flow.
How would you systematically debug an inference API latency breach?
This tests structured debugging across the full inference stack. A strong answer traces the request path from ingress to GPU, splits TTFT from token-generation latency, inspects queuing and batching, then applies targeted fixes.
How would you visualize high-cardinality categorical relationships?
This tests dimensionality reduction when categories exceed roughly fifty levels. A strong answer proposes top-N aggregation with an other bucket and density or hierarchy plots like heatmaps or treemaps. A red flag is scrolling a bar chart or sampling rows.
How Stable Diffusion generates images
The text encoder turns the prompt into embeddings, the U-Net predicts noise to remove conditioned on those embeddings, and the scheduler controls how noise is stepped down over iterations in…
Re-identification in multi-object tracking
Re-ID matches an object to its prior id using appearance embeddings, not just position; store track features and match re-entering detections by embedding similarity.
Architectural challenges for deploying ML models on resource-constrained edge devices
Tests Edge MLOps architecture under severe constraints. Strong answers hit quantization and delta OTA updates for flaky networks, power-aware scheduling, and closed-loop drift detection.
Describe your systematic approach to interpreting an 8-feature pair plot
Check 8 diagonals for skew; scan 28 off-diagonals for nonlinear trends, variance patterns, and hue clusters; flag redundancy; spot outliers.
Aligning text and image representations
Contrastive learning like CLIP pulls matched image-text pairs together and pushes mismatches apart; alternatively projection layers map one modality into a frozen model's space.
Self-supervised pretraining for video understanding
Define a label-free task like temporal order prediction or contrastive clip matching that forces temporal reasoning, then fine-tune on labeled action data.
We are hiring for this. Every open role lists the topics its interview covers, so you can prepare for the real thing rather than guessing.
See open roles