Ateneo - UPO

The course examines the use of Artificial Intelligence and quantitative methods for the formulation, analysis, and support of strategic business decisions in dynamic and uncertain environments.
The course introduces the main elements of sequential decision problems, with particular attention to the representation of states, decisions, information, transitions, and objectives. It then covers Dynamic Programming, resource allocation problems, Markov Decision Processes, and value and policy iteration algorithms.
A second part focuses on the limitations of exact methods and on approximate decision policies. The course also addresses the main problems of sequential learning, with particular reference to the exploration-exploitation trade-off and to Temporal-Difference Learning, SARSA, and Q-Learning.
The course concludes with the design, comparison, and evaluation of simple decision policies applied to business problems, jointly considering performance, uncertainty, constraints, and the dynamic consequences of decisions.
Quantitative applications and computational exercises are developed in Python.

The compulsory material for exam preparation is provided by the lecturer on the course DIR page. Any books, articles, or other sources suggested during the course are optional and intended only for further study.

The course aims to provide students with the conceptual and quantitative tools required to formulate, analyze, and support strategic business decisions in dynamic and uncertain environments.
In particular, the course aims to develop the ability to represent sequential decision problems by identifying states, decisions, information, transitions, and objectives, and to use Dynamic Programming and Markov Decision Processes to analyze simple resource allocation and control problems.
A further objective is to develop the ability to understand when exact methods become difficult to apply and to compare alternative approaches based on approximate decision policies and sequential learning methods, including Temporal-Difference Learning, SARSA, and Q-Learning.
The applied activities in Python are designed to strengthen the ability to translate theoretical models into simple computational implementations, compare alternative policies, and critically interpret their results.

Basic knowledge of mathematics, probability, and statistics is required, sufficient to understand functions, derivatives, expected value, random variables, and probability distributions.
Introductory knowledge of Python programming is also required, particularly the use of variables, data structures, functions, and basic libraries for numerical computing and data visualization.

The course combines lectures, quantitative examples, exercises, and computational activities in Python.

The information provided in the Syllabus concerning course content, teaching materials, teaching methods, and assessment applies both to attending and non-attending students.

DISABILITY AND SPECIFIC LEARNING DISORDERS
Students with disabilities, Specific Learning Disorders (SLD), or Special Educational Needs (SEN) may request dedicated services and specific support by contacting the University Staff for Student Careers and Services and consulting the dedicated University webpage.
After contacting the relevant University Staff, students with disabilities, SLD, or SEN may contact the course lecturer regarding the adaptation of examination arrangements and teaching-related aspects.

The exam consists of an oral examination aimed at assessing knowledge and understanding of the course topics, as well as the ability to apply the concepts and models studied to simple business decision problems.
The examination may include both theoretical questions and applications, numerical examples, interpretation of results, or discussion of computational procedures covered during the course.
To pass the exam, students are required to demonstrate adequate knowledge of the fundamental concepts and the ability to apply them correctly in simple contexts. Higher grades will be awarded for deeper understanding, ability to connect different topics, autonomy in analysis, and clarity of exposition.

The course is organized into ten blocks.

1. Artificial Intelligence, strategy and decision framing
Role of AI in business decision-making. Identification of decisions, objectives, metrics, and sources of uncertainty. Distinction between static and sequential decisions. Introduction to the concept of policy.

2. Modeling sequential decision problems
Representation of state, decisions, information, transitions, and objective function. State sufficiency, decision timing, and simulation of simple decision processes.

3. Dynamic Programming and the principle of optimality
Multistage problems, open-loop and closed-loop decisions. Value function, Bellman equation, principle of optimality, and backward recursion. Applications to simple planning problems.

4. Strategic resource allocation
Allocation of scarce resources, knapsack problems, and budget decisions. Dynamic formulation, sensitivity analysis, and comparison of alternative allocation choices.

5. Stochastic business decisions and Markov Decision Processes
Introduction to decision-making under uncertainty. Markov property, states, actions, transitions, rewards, and evaluation criteria. Applications to inventory and resource-management problems.

6. Value iteration and policy iteration
Bellman operators, policy evaluation, and optimal policy search. Value iteration and policy iteration algorithms. Comparison between exact methods and simple rollout strategies.

7. From exact optimization to approximate policies
Computational limits of exact methods. Curse of dimensionality and main classes of approximate policies: Policy Function Approximations, Cost Function Approximations, Value Function Approximations, and Direct Lookahead Approximations.

8. Learning while deciding: exploration and exploitation
Sequential learning problems. Multi-armed bandits, action values, greedy and epsilon-greedy strategies, optimistic initialization, non-stationarity, and Upper Confidence Bound.

9. Temporal-Difference Learning, SARSA and Q-Learning
Model-free learning, Temporal-Difference Learning, SARSA, and Q-Learning. Distinction between on-policy and off-policy approaches and comparison with model-based methods.

10. Designing AI-driven business policies
Design, comparison, and evaluation of decision policies in business problems. Applications to supply chain and inventory management. Analysis of cost, service, stability, bullwhip effect, and response to shocks.

Quantitative applications and computational exercises are developed in Python.

Knowledge and understanding: by the end of the course, students will be able to understand the main concepts and methods for modeling and analyzing sequential business decision problems, including Dynamic Programming, resource allocation, Markov Decision Processes, approximate policies, and Reinforcement Learning methods.

Applying knowledge and understanding: students will be able to formulate simple dynamic decision problems by identifying states, decisions, information, transitions, and objectives, apply the main methods covered in the course, and understand and discuss simple computational implementations in Python.

Making judgements: students will be able to critically assess the assumptions, advantages, and limitations of alternative decision approaches, compare different policies, and identify the most appropriate method for the specific business problem considered.

Communication skills: students will be able to clearly and correctly describe and discuss the main models, algorithms, and quantitative results covered in the course, using appropriate technical language.

Learning skills: students will be able to independently explore topics related to decision analytics, sequential decision problems, and Artificial Intelligence applications to business decision-making using course materials, specialist texts, and other relevant sources.