A code-first course on building temporal intelligence systems for forecasting, reinforcement learning, and world models. Part of the Hands-On AI Science series, designed around Innovation-First Learning principles.
Real systems unfold over time: markets move, sensors stream, agents act. Forecasting, decision-making, and control under uncertainty are central to robotics, trading, operations, and adaptive products. This course equips students to reason about and build systems that learn from sequential experience.
Core concepts, models, and ideas behind temporal and sequential learning: autoregressive and state-space models, Markov decision processes, Bellman equations, temporal difference learning, policy gradients and actor-critic methods, model-based reinforcement learning, and world models that simulate future dynamics.
PyTorch, statsmodels, Prophet, Darts, NeuralForecast, Gymnasium, Stable Baselines3, Ray RLlib, TensorBoard, and Weights & Biases.
A specific course syllabus is built for each audience: graduate or undergraduate, across engineering, digital health, or computer science.
As AI and mature libraries handle standard tasks, professional developers must focus on innovation. Student projects tackle new use cases by generating unique data and fine-tuning task-specific temporal and sequential models.
Students begin their projects while learning the material and enrich them as new concepts arrive. Each team gives several in-class presentations for discussion and feedback.
Time Series Fundamentals
statsmodels, pandas, autocorrelation
Classical Forecasting Models
ARIMA, Prophet, seasonal decomposition
Deep Forecasting
Darts, NeuralForecast, temporal CNNs
Anomaly Detection & State-Space Models
Kalman filters, autoencoders
Project Proposal Presentations
Student proposals, peer feedback
Markov Decision Processes
Gymnasium, value/policy iteration
Deep Reinforcement Learning
Stable Baselines3, DQN, PPO
Interim Project Presentations
Progress demos, instructor feedback
Policy Gradients & Actor-Critic
A2C, SAC, Ray RLlib
World Models & Model-Based RL
Dreamer, learned dynamics
Temporal Transformers
Temporal Fusion Transformer, PatchTST
Multi-Step Planning & Exploration
MCTS, curiosity-driven exploration
Final Project Presentations
Live demos, peer evaluation
