AI Foundations courses

• Supervised learning: classification, regression, decision trees, SVMs, ensemble methods
• Unsupervised learning: clustering (k-means, DBSCAN), dimensionality reduction (PCA, t-SNE, UMAP)
• Semi-supervised and self-supervised learning: leveraging unlabelled data at scale
• Reinforcement learning (RL): Q-learning, policy gradients, actor-critic, RLHF

• Convolutional Neural Networks (CNNs): image recognition, object detection, segmentation
• Recurrent Neural Networks (RNNs), LSTMs, GRUs: sequence modelling, time series
• Transformer architecture (2017): self-attention; backbone of all modern LLMs and vision models
• Graph Neural Networks (GNNs): learning on relational and network data

• Statistical NLP (pre-2012): n-gram models, HMMs, CRFs for parsing and tagging
• Word embeddings: Word2Vec (2013), GloVe - dense distributed representations
• Pre-trained language models: ELMo, BERT (2018), RoBERTa, XLNet
• Large Language Models (LLMs): GPT-3/4, PaLM, Claude, LLaMA - few-shot learners

• Image classification: AlexNet (2012), VGG, ResNet, EfficientNet milestones
• Object detection: YOLO, R-CNN family, DETR (Transformer-based detection)
• Image segmentation: semantic, instance, panoptic (Mask R-CNN, SAM)
• Generative image models: GANs (StyleGAN), Stable Diffusion, DALL-E, Midjourney

• Robot perception: sensor fusion, SLAM (simultaneous localisation and mapping)
• Motion planning and control: trajectory optimisation, model-predictive control
• Manipulation: dexterous grasping, object interaction, learning from demonstration
• Autonomous vehicles: self-driving car perception, prediction, and planning

• Knowledge graphs: Freebase, Wikidata, Google Knowledge Graph - structured world knowledge
• Ontologies and semantic web: OWL, RDF, SPARQL for machine-readable knowledge
• Automated planning and scheduling: STRIPS, PDDL, hierarchical task networks
• Constraint satisfaction and optimisation: SAT solvers, integer programming

• Classical search: A*, heuristic search, constraint propagation
• Monte Carlo Tree Search (MCTS): planning under uncertainty
• Game-playing AI milestones: Deep Blue (chess), Watson (Jeopardy!), AlphaGo/Zero (Go), AlphaStar (StarCraft)

• Explainable AI (XAI): LIME, SHAP, attention visualisation, saliency maps
• Mechanistic interpretability: understanding circuits and features inside neural networks
• Fairness and bias detection: demographic parity, equalised odds, counterfactual fairness
• Robustness and adversarial examples: attacking and defending AI models

• Value alignment: ensuring AI systems act in accordance with human values and intentions
• Reinforcement Learning from Human Feedback (RLHF): aligning LLMs to human preferences
• Constitutional AI and self-critique methods (Anthropic)
• Scalable oversight: supervising AI systems that surpass human capabilities

See also: Responsible AI courses →

• GPU computing: CUDA ecosystem; NVIDIA dominance in AI training hardware
• TPUs (Tensor Processing Units): Google's custom ASIC for ML workloads
• Specialised AI chips: Groq, Cerebras, Graphcore, Intel Gaudi
• Distributed training: data parallelism, model parallelism, pipeline parallelism

• Healthcare: diagnostic imaging, drug discovery, genomics, clinical decision support
• Finance: algorithmic trading, credit scoring, fraud detection, risk modelling
• Legal: contract analysis, case law search, regulatory compliance, AI liability
• Education: intelligent tutoring systems, adaptive learning, automated assessment
• Scientific discovery: protein folding (AlphaFold), materials science, climate modelling

• Conversational AI and chatbots: from ELIZA (1960s) to ChatGPT (2022)
• Affective computing: emotion recognition, sentiment-aware systems
• Personalisation and recommendation systems: collaborative filtering, content-based, hybrid

• Transfer learning and domain adaptation: reusing knowledge across tasks
• Meta-learning and few-shot learning: learning to learn from limited examples
• Federated learning: distributed training while preserving data privacy
• AutoML: automated model selection, hyperparameter optimisation, neural architecture search (NAS)

• Generative models: GANs (2014), VAEs, Normalising Flows, Diffusion Models (2020+)
• Mixture of Experts (MoE): sparse activation for efficient large-scale models
• Multimodal architectures: combining text, image, audio, video (CLIP, Flamingo, GPT-4V)
• Neural scaling laws: predicting model performance from compute, data, and parameters

• Machine translation: seq2seq, attention, Transformer-based (Google Translate, DeepL)
• Information retrieval and question answering: RAG, open-domain QA
• Text summarisation, sentiment analysis, named entity recognition
• Multilingual and cross-lingual NLP: low-resource languages, mBERT, XLM-R

• Video understanding: action recognition, temporal modelling, video generation (Sora)
• 3D vision: point cloud processing, NeRF (Neural Radiance Fields), Gaussian splatting
• Medical imaging: AI-assisted radiology, pathology, ophthalmology (DeepMind, Google)
• Vision-Language Models (VLMs): CLIP, Flamingo, LLaVA - joint vision-text understanding

• Drone and UAV autonomy: navigation without GPS, multi-agent coordination
• Human-robot interaction (HRI): collaborative robots (cobots), social robots
• Sim-to-real transfer: training in simulation, deploying in the physical world
• Foundation models for robotics: RT-2, PaLM-E - language-conditioned robot policies

• Commonsense reasoning: ConceptNet, reasoning benchmarks (WinoGrad, ARC)
• Neuro-symbolic AI: combining neural networks with symbolic logic
• Causal reasoning and causal inference: beyond correlation in AI systems
• Temporal and probabilistic reasoning: Bayesian networks, Markov models

• Multi-agent systems: cooperative, competitive, and mixed-motive settings
• Mechanism design and computational game theory: incentive-aligned AI systems
• Real-world planning: logistics, supply chain, scheduling applications

• Uncertainty quantification: Bayesian deep learning, conformal prediction
• Model auditing and AI governance frameworks: IEEE, NIST AI RMF, EU AI Act
• Human-in-the-loop AI: active learning, human oversight in high-stakes decisions

See also: AI Governance courses →

• Red-teaming and adversarial safety evaluation of foundation models
• Corrigibility, shutdown problem, and utility-function misspecification
• Long-term and existential risk: AI systems with misaligned goals at scale
• AI governance and policy: EU AI Act (2024), US AI Executive Orders, international regulation

See also: AI Governance courses →

• Efficient inference: quantisation, pruning, distillation, speculative decoding
• MLOps: model deployment, monitoring, versioning, drift detection in production
• Edge AI: running models on-device (phones, IoT sensors) without cloud dependency
• Energy and sustainability: carbon footprint of training; green AI research

• Manufacturing & supply chain: predictive maintenance, quality control, demand forecasting
• Agriculture: precision farming, crop disease detection, yield prediction
• Creative AI: music generation, art, writing assistance, video production
• Cybersecurity: intrusion detection, malware classification, adversarial ML
• Government & public sector: policy analysis, public service automation, defence AI

• Accessibility AI: speech-to-text, screen readers, sign language recognition
• Social media analysis: misinformation detection, community detection, opinion dynamics
• Human-AI collaboration: teaming, trust calibration, appropriate reliance on AI