Computer Β· Part D - Data, Connectivity, Intelligence and Interactive Computing Β· Major topic 21 of 39

Machine Learning, Deep Learning, Generative AI, LLMs and AI Engineering

Learning patterns from data, building foundation-model applications, measuring performance and deploying systems responsibly. This domain depends on programming, data handling, probability, linear algebra and optimization.

308 items insideAsked 1Γ— in the RPSC/RSSB papers we analysed, most recently 2024COM-21

Also look at: Science & Technology: Artificial Intelligence, Machine Learning, Robotics and Cyber-Physical SystemsChemistry: Cheminformatics, Chemical Data & AI/ML for ChemistryScience & Technology: Computer Architecture, Operating Systems, Programming and Software Systems

  1. 1

    Historical evolution β€” key dates

    • Early perceptrons and machine-learning ideas emerge.1950s
    • Rosenblatt’s perceptron becomes a landmark neural-learning model.1957
    • Backpropagation is popularized for training multilayer neural networks.1986
    • Statistical learning, support-vector machines, ensembles and large datasets grow.1990s-2000s
    • Deep-learning methods attract renewed attention; GPU training later accelerates progress.2006
    • AlexNet’s image-recognition results mark a major deep-learning inflection.2012
    • β€œAttention Is All You Need” introduces the Transformer architecture.2017
    • GPT-3 highlights few-shot language-model capabilities.2020
    • ChatGPT is released publicly as a research preview.2022
    • Multimodal foundation models, retrieval-augmented generation, tool-using agents, evaluations, safety and inference efficiency become major engineering topics.2023-2026
  2. 2

    Machine-learning problem types

    • Supervised; unsupervised; semi-supervised; self-supervised; reinforcement; online/incremental; transfer learning
      7 more points
      • Supervised
      • Unsupervised
      • Semi-supervised
      • Self-supervised
      • Reinforcement
      • Online/incremental
      • Transfer learning
    • Classification; regression; ranking; clustering; dimensionality reduction; anomaly detection; forecasting; recommendation
      8 more points
      • Classification
      • Regression
      • Ranking
      • Clustering
      • Dimensionality reduction
      • Anomaly detection
      • Forecasting
      • Recommendation
    • Inputs/features; targets/labels; training/validation/test splits; distribution shift; deployment context
      5 more points
      • Inputs/features
      • Targets/labels
      • Training/validation/test splits
      • Distribution shift
      • Deployment context
  3. 3

    Data and features

    • Dataset version; schema; collection; labeling; data quality; leakage; imbalance; sampling; deduplication
      9 more points
      • Dataset version
      • Schema
      • Collection
      • Labeling
      • Data quality
      • Leakage
      • Imbalance
      • Sampling
      • Deduplication
    • Scaling; normalization; categorical encoding; text/image/audio representation; missing-data strategies
      5 more points
      • Scaling
      • Normalization
      • Categorical encoding
      • Text/image/audio representation
      • Missing-data strategies
    • Feature engineering; feature selection; embeddings; feature store; lineage; consent and provenance
      6 more points
      • Feature engineering
      • Feature selection
      • Embeddings
      • Feature store
      • Lineage
      • Consent and provenance
  4. 4

    Classical algorithms

    Asked 1Γ— Β· last 2024

    • Linear/logistic regression; k-nearest neighbours; naive Bayes; decision trees; random forests
      5 more points
      • Linear/logistic regression
      • K-nearest neighbours
      • Naive Bayes
      • Decision trees
      • Random forests
    • Gradient boosting; support-vector machines; k-means; hierarchical clustering; DBSCAN; PCA; matrix factorization
      7 more points
      • Gradient boosting
      • Support-vector machines
      • K-means
      • Hierarchical clustering
      • DBSCAN
      • PCA
      • Matrix factorization
    • Calibration; regularization; model selection; ensemble methods; interpretable models
      5 more points
      • Calibration
      • Regularization
      • Model selection
      • Ensemble methods
      • Interpretable models
  5. 5

    Neural-network foundations

    • Neuron; layer; activation; loss; gradient; backpropagation; optimizer; batch; epoch; learning rate
      10 more points
      • Neuron
      • Layer
      • Activation
      • Loss
      • Gradient
      • Backpropagation
      • Optimizer
      • Batch
      • Epoch
      • Learning rate
    • MLP; CNN; RNN; LSTM/GRU; attention; Transformer; encoder/decoder; residual connection; normalization
      9 more points
      • MLP
      • CNN
      • RNN
      • LSTM/GRU
      • Attention
      • Transformer
      • Encoder/decoder
      • Residual connection
      • Normalization
    • Initialization; vanishing/exploding gradients; dropout; weight decay; batch normalization; early stopping
      6 more points
      • Initialization
      • Vanishing/exploding gradients
      • Dropout
      • Weight decay
      • Batch normalization
      • Early stopping
  6. 6

    Model training and evaluation

    • Baseline; split strategy; cross-validation; hyperparameter tuning; grid/random/Bayesian search
      5 more points
      • Baseline
      • Split strategy
      • Cross-validation
      • Hyperparameter tuning
      • Grid/random/Bayesian search
    • Accuracy; precision; recall; F1; ROC-AUC; PR-AUC; calibration; MAE/MSE/RMSE; ranking metrics
      9 more points
      • Accuracy
      • Precision
      • Recall
      • F1
      • ROC-AUC
      • PR-AUC
      • Calibration
      • MAE/MSE/RMSE
      • Ranking metrics
    • Error analysis; subgroup evaluation; confidence intervals; robustness; fairness; interpretability; reproducibility
      7 more points
      • Error analysis
      • Subgroup evaluation
      • Confidence intervals
      • Robustness
      • Fairness
      • Interpretability
      • Reproducibility
  7. 7

    Deep-learning specialties

    • Computer vision; object detection; segmentation; OCR; pose estimation; speech and audio; time series
      7 more points
      • Computer vision
      • Object detection
      • Segmentation
      • OCR
      • Pose estimation
      • Speech and audio
      • Time series
    • Graph neural networks; recommender systems; multimodal learning; self-supervised representation learning
      4 more points
      • Graph neural networks
      • Recommender systems
      • Multimodal learning
      • Self-supervised representation learning
  8. 8

    Generative modelling

    • Autoregressive models; variational autoencoders; GANs; diffusion models; normalizing flows; energy-based models
      6 more points
      • Autoregressive models
      • Variational autoencoders
      • GANs
      • Diffusion models
      • Normalizing flows
      • Energy-based models
    • Text/image/audio/video/code generation; conditioning; guidance; sampling; controllability; watermarking concepts
      6 more points
      • Text/image/audio/video/code generation
      • Conditioning
      • Guidance
      • Sampling
      • Controllability
      • Watermarking concepts
  9. 9

    LLM foundations and application design

    • Tokenization; vocabulary; embeddings; positional representation; attention; context window; pretraining objective
      7 more points
      • Tokenization
      • Vocabulary
      • Embeddings
      • Positional representation
      • Attention
      • Context window
      • Pretraining objective
    • Pretraining; instruction tuning; supervised fine-tuning; preference optimization; RLHF/DPO concepts; quantization
      6 more points
      • Pretraining
      • Instruction tuning
      • Supervised fine-tuning
      • Preference optimization
      • RLHF/DPO concepts
      • Quantization
    • Prompt structure; system/developer/user instructions; examples; structured output; tool definitions; context management
      6 more points
      • Prompt structure
      • System/developer/user instructions
      • Examples
      • Structured output
      • Tool definitions
      • Context management
    • Hallucination; grounding; citation checking; refusal behavior; uncertainty; prompt injection; model limitations
      7 more points
      • Hallucination
      • Grounding
      • Citation checking
      • Refusal behavior
      • Uncertainty
      • Prompt injection
      • Model limitations
  10. 10

    Retrieval-augmented generation (RAG)

    • Document ingestion; parsing; chunking; metadata; embeddings; vector index; lexical index; hybrid retrieval
      8 more points
      • Document ingestion
      • Parsing
      • Chunking
      • Metadata
      • Embeddings
      • Vector index
      • Lexical index
      • Hybrid retrieval
    • Query rewriting; filters; reranking; context assembly; citations; answer generation; abstention
      7 more points
      • Query rewriting
      • Filters
      • Reranking
      • Context assembly
      • Citations
      • Answer generation
      • Abstention
    • Evaluation set; retrieval recall; groundedness; context precision; answer relevance; latency/cost trade-off
      6 more points
      • Evaluation set
      • Retrieval recall
      • Groundedness
      • Context precision
      • Answer relevance
      • Latency/cost trade-off
  11. 11

    Agents and tool-using systems

    • Model vs agent; goal; plan; tool schema; action; observation; state; memory; orchestration
      9 more points
      • Model vs agent
      • Goal
      • Plan
      • Tool schema
      • Action
      • Observation
      • State
      • Memory
      • Orchestration
    • Single-agent workflow; multi-agent coordination; task decomposition; handoff; retry; stop conditions; human approval
      7 more points
      • Single-agent workflow
      • Multi-agent coordination
      • Task decomposition
      • Handoff
      • Retry
      • Stop conditions
      • Human approval
    • Sandboxing; least privilege; tool permissions; prompt injection; secret leakage; audit logs; failure recovery
      7 more points
      • Sandboxing
      • Least privilege
      • Tool permissions
      • Prompt injection
      • Secret leakage
      • Audit logs
      • Failure recovery
  12. 12

    AI engineering and MLOps

    • Data/model pipeline; experiment tracking; model registry; feature store; deployment; inference server; batching
      7 more points
      • Data/model pipeline
      • Experiment tracking
      • Model registry
      • Feature store
      • Deployment
      • Inference server
      • Batching
    • Offline/online inference; model serving; monitoring; drift; retraining; rollback; canary; shadow testing
      8 more points
      • Offline/online inference
      • Model serving
      • Monitoring
      • Drift
      • Retraining
      • Rollback
      • Canary
      • Shadow testing
    • GPU/accelerator basics; throughput; latency; tokens per second; caching; quantization; distillation; batching
      8 more points
      • GPU/accelerator basics
      • Throughput
      • Latency
      • Tokens per second
      • Caching
      • Quantization
      • Distillation
      • Batching
    • Versioned datasets; reproducible environments; CI/CD/CT; model card; dataset documentation; lineage
      6 more points
      • Versioned datasets
      • Reproducible environments
      • CI/CD/CT
      • Model card
      • Dataset documentation
      • Lineage
  13. 13

    AI safety, governance and frontier questions

    • Risk assessment; NIST AI RMF; generative-AI risk profile; data rights; privacy; security; provenance
      7 more points
      • Risk assessment
      • NIST AI RMF
      • Generative-AI risk profile
      • Data rights
      • Privacy
      • Security
      • Provenance
    • Evaluation contamination; benchmark overfitting; red-teaming; adversarial robustness; jailbreak resistance
      5 more points
      • Evaluation contamination
      • Benchmark overfitting
      • Red-teaming
      • Adversarial robustness
      • Jailbreak resistance
    • Synthetic data; model collapse risk; copyright questions; bias; accessibility; labour and environmental impact
      6 more points
      • Synthetic data
      • Model collapse risk
      • Copyright questions
      • Bias
      • Accessibility
      • Labour and environmental impact
    • Interpretability; mechanistic interpretability; scalable oversight; agentic-system evaluation; post-training research
      5 more points
      • Interpretability
      • Mechanistic interpretability
      • Scalable oversight
      • Agentic-system evaluation
      • Post-training research