Audience
- Python developers interested in sequential decision-making.
- Analysts exploring simulation, optimisation and adaptive systems.
- Technical graduates preparing for deeper machine-learning study.
AI & technology
Reinforcement Learning
Explore learning through interaction, feedback and rewards.
Learning with a practical purpose.
Understand how an agent can learn to choose actions through interaction with an environment. This introductory technical module uses small simulated problems to make states, actions, rewards and policies tangible. You will investigate how a seemingly reasonable reward can lead to behaviour that does not match the original business intention.
The practical thread is a constrained resource-allocation simulation. You will compare a learned policy with a simple rule-based approach, repeat experiments and examine performance under changing conditions. The aim is to develop sound experimental judgement and a working foundation for further study, rather than to suggest that reinforcement learning is automatically the best solution to an optimisation problem.
Live onlineTeaching through Kent Business College, UK
2 hoursOne live session each teaching week
14 sessionsA semester for each module
28 live hoursPer module, plus independent study
The outline below is a 14-session online module, with one two-hour live session each teaching week. Reading weeks, breaks and additional activities follow the agreed cohort calendar. The cohort syllabus, full route, dates, fees, independent study and assessment requirements are confirmed before enrolment.
Who this module is designed for.
Starting knowledge and preparation
- Comfort with Python functions, arrays and basic plotting.
- Working knowledge of probability, averages and elementary algebra.
- An introductory understanding of machine learning is helpful.
What you will work towards.
- Express a sequential decision problem as states, actions, transitions and rewards.
- Implement a small learning environment with observable outcomes and clear constraints.
- Compare basic tabular learning approaches with a rule-based policy.
- Run repeated experiments and interpret variation in learning and evaluation results.
- Explain reward-design risks and the limits of transferring a policy beyond its simulation.
Explore the 14-session learning plan.
Explore the focus and practical output of each teaching session. Open individual sessions below, or expand the complete plan.
Session 01 Sequential decisions and the learning problem
Distinguish reinforcement learning from supervised prediction and define a manageable simulated decision task.
Practical output: A decision-problem canvas.
Session 02 States, actions and environment design
Specify what the agent observes, what it can change and how the environment responds.
Practical output: A state-and-action specification.
Session 03 Rewards and unintended behaviour
Compare reward functions and identify ways an agent could optimise a proxy rather than the desired outcome.
Practical output: A reward design with failure examples.
Session 04 Exploration through a bandit problem
Experiment with exploration and exploitation using a small repeated-choice simulation with uncertain returns.
Practical output: A comparison of exploration strategies.
Session 05 Policies, returns and value
Calculate simple returns and trace how value estimates support action selection over multiple steps.
Practical output: A worked policy-evaluation notebook.
Session 06 Dynamic programming foundations
Use a small known environment to explore policy evaluation and improvement before model-free learning.
Practical output: A tabular planning baseline.
Session 07 Learning from completed episodes
Estimate values from repeated episodes and examine the influence of sample size and variability.
Practical output: An episode-based learning experiment.
Session 08 Temporal-difference learning
Update estimates during interaction and compare learning behaviour with the previous episode-based approach.
Practical output: A documented temporal-difference implementation.
Session 09 Q-learning and action selection
Implement a tabular action-value learner and investigate the effect of learning and exploration settings.
Practical output: A trained policy and learning curves.
Session 10 Evaluating a policy fairly
Separate learning from evaluation, repeat runs and compare against simple rules under equivalent conditions.
Practical output: A repeated-run evaluation report.
Session 11 From tables to function approximation
Explore why larger observation spaces need approximation and inspect a guided neural-policy example.
Practical output: A comparison of tabular and approximated representations.
Session 12 Constraints and changing environments
Test resource limits, altered demand and unfamiliar states, documenting where the learned policy struggles.
Practical output: A stress-test and constraint register.
Session 13 Resource-allocation project workshop
Integrate the simulation, learning loop and reporting into a reproducible experiment with a clear baseline.
Practical output: A complete experimental project.
Session 14 Policy demonstration and critique
Defend the reward design, explain experimental uncertainty and assess whether further development is justified.
Practical output: A policy demonstration and next-step recommendation.
Apply the learning
A resource-allocation learning experiment
Design a small simulation in which an agent allocates limited capacity across competing demands. Train a tabular policy, compare it with a simple heuristic and repeat the evaluation. The final submission should make it possible for another learner to reproduce the experiment and understand both its benefits and its limitations.
Your project evidence can include
- An environment specification and runnable simulation.
- Training code, configuration and repeated-run results.
- A comparison report and reward-design critique.
Use anonymised, public or suitable synthetic material where workplace information is confidential. The issued assessment brief confirms the required evidence.
Assessment, feedback and certification.
- Worked exercises demonstrate understanding of rewards, returns and value updates.
- The project is assessed for reproducibility, baseline quality and fair evaluation.
- A short discussion explores constraint handling, unexpected behaviour and the limits of the simulated environment.
Teaching and the relevant course certificate are provided through Kent Business College in the UK. Completion requirements are set out in the module information; attendance alone should not be assumed to meet every assessment requirement.
Questions worth asking.
Is advanced mathematics essential?
Elementary algebra and probability are required. The module develops intuition through small examples rather than assuming research-level mathematical preparation.
Will I work with a physical robot?
No. Core exercises use simulations so that actions, rewards and repeated experiments can be inspected safely and affordably.
Does this make me an RL specialist?
It provides an applied foundation. Advanced algorithms, large-scale training and real-world control require further study and domain-specific validation.